Applied Mathematical Modelling 40 (2016) 9625–9655

Contents lists available at ScienceDirect

Applied Mathematical Modelling

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

Review Article

An overview of smoothed particle hydrodynamics for

simulating multiphase flow
Zhi-Bin Wang a,b, Rong Chen a,b,∗, Hong Wang a,b, Qiang Liao a,b, Xun Zhu a,b,
Shu-Zhe Li a,b
a
Key Laboratory of Low-grade Energy Utilization Technologies and Systems (Chongqing University), Ministry of Education,
Chongqing 400030, China
b
Institute of Engineering Thermophysics, Chongqing University, Chongqing 400030, China

a r t i c l e i n f o a b s t r a c t

Article history: Smoothed particle hydrodynamics (SPH), is a meshfree, Lagrangian, particle method, which
Received 9 March 2015 is advantageous over conventional grid-based numerical methods in the aspect of inter-
Revised 25 May 2016
face treatment. Therefore, it has shown promising potential for simulating the multiphase
Accepted 16 June 2016
flow problems. Over the past decades, numerous efforts have been devoted to the simula-
Available online 8 July 2016
tion of the multiphase flow by using the SPH method. In this review, recent advances in
Keywords: SPH for simulating the multiphase flow are reviewed. Firstly, the basic concept of SPH is
Smoothed particle hydrodynamics briefly introduced. Attention is then paid to how to treat the pressure of incompressible
Multiphase flow flow, boundary condition and the surface tension in SPH. In addition, modeling free sur-
Surface tension face flow is briefly introduced. The modified interfacial models and their applications in
High density ratio handling high density ratio are addressed. Finally, a summary of SPH in the application of
the multiphase flow is given.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction

Multiphase flow is a typical phenomenon which widely exists in nature and engineering applications [1], such as mud-
slides [2,3], gas–liquid flow in flow channels [4,5], fluid-particle flow in fluidized beds [6,7] and so on. Therefore, the inves-
tigation of the multiphase flow is extremely important in not only theory but also engineering [8,9]. In general, there are
three approaches to study the multiphase flow, including the experiments, theoretical analysis and numerical methods. With
the rapid development of computers, numerical methods have been widely adopted because they can offer the advantages
of safety, high efficiency and low cost. Although numerical methods have been successfully applied, there still exists one of
the most critical obstacles in the simulation of the multiphase flow, i.e., the interface tracking, because the interfacial behav-
iors are rather complex in the numerical process [10], which involves complex solid wall and large surface deformation and
moving interfaces. Right now, typical methods for tracking the interface mainly include Particle in Cell (PIC) [11,12], Marker
and Cell (MAC) [13], Volume of Fluid (VOF) [14] and Level Set Method (LSM) [15]. However, these methods are basically grid
or semi-grid method, making them difficult to accurately track and describe the interface. Moreover, meshing and remesh-
ing may be required in some cases. In addition, the problems of mesh distortion, deformity, overlay and twisting may occur
when tracking the interface [16,17]. At present, these drawbacks are hardly avoided because they are inherent with grid-
based methods [18]. Nevertheless, the above-mentioned problems encountering in the grid based methods can be resolved

∗
Corresponding author at: Institute of Engineering Thermophysics, Chongqing University, Chongqing 40 0 030, China. Fax: +86 23 65102474.
E-mail addresses: [email protected], [email protected] (R. Chen).

http://dx.doi.org/10.1016/j.apm.2016.06.030
0307-904X/© 2016 Elsevier Inc. All rights reserved.
9626 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

by the meshfree methods [19]. Over the past decades, the meshfree methods have experienced an exponential development
[20–23]. Smoothed particle hydrodynamics (SPH), one of the meshfree methods, has also received much attention. SPH was
firstly developed by Lucy [24] and Gingold & Monaghan [25] and initially used for astrophysical problems. Afterwards, the
SPH method has also been used to solve applied mechanics problems. Therefore, the term of hydrodynamics should be in-
terpreted as mechanics in general. If this method is applied to the branch of mechanics rather than classical hydrodynamics,
some researchers called it smoothed particle applied mechanics (SPAM) to make a distinction with the classical hydrody-
namics [26–28]. As compared to other meshfree methods [29–34], the SPH method is more mature and thus widely used
[20,22,35].
Basically, SPH is a pure Lagrangian, meshfree, particle method [36–39]. The basic idea is that the whole continuous sys-
tem is represented by a series of particles. Each of them has independent information and follows the governing equations.
The simulation can be carried out after determining the locations and the physical information of the particles. Owing to its
nature of smoothed particle and pure Lagrangian, SPH offers many merits of free distribution of particles, self-adaptability,
ease to capture the interface, natural concurrent computation. All these features make it particularly suitable to simulate
the multiphase flow. Extensive works have been made to develop and apply the SPH method to study the multiphase flow
behaviors [23,40–48] and a significant progress has been attained.
The purpose of this review is to summarize recent advances in SPH for simulating the multiphase flow. The review is
organized as follows. Section 2 briefly introduces the basic principle of SPH and various discretized equations are discussed
in Section 3. Section 4 gives an overview about pressure treatment for incompressible flow. In Section 5, the boundary con-
dition treatments are discussed. The overview of kernel and improvement about interpolation approximation are introduced
in Section 6. In Section 7, modeling free surface flow is briefly introduced. Section 8 summarizes various SPH models to
treat the surface tension. In Section 9, the corrected models for high density ratio between phases are discussed. Section
10 provides some applications of SPH on the multiphase flow. Finally, a summary of the SPH method for the multiphase
flow is provided.

2. The basic principle of SPH

SPH is a method using a series of particles which carry physical properties to represent the fluid. For all particles, the
changes of motion and physical properties follow the governing equation. Based on these particles, approximate numerical
solutions are obtained using kernel and particle approximations, which are the most important steps in SPH. Generally,
the kernel approximation employs a weight integral representation for the field function approximation, where the weight
function used in the integral representation is called smoothing kernel function. The particle approximation is done by
replacing the integral representation of the field function with the sum of all the corresponding values at the neighboring
particles. The detailed procedure about the kernel and particle approximations is presented as follows.
In the kernel approximation, Dirac delta function δ needs to be used, by which the function ƒ(r) can be converted to an
integral form,
    
f (r ) = f rδ r − r dr , (1)


where dr denotes a volume element, ƒ(r) is a function at the position r and r represents the integral domain. To simply
calculate the function of ƒ(r), a smoothing function is introduced to replace Dirac delta function, by which Eq. (1) can be
transferred to the following equation,
    
 f (r ) = f r  W r − r  , h dr  , (2)


where the symbol  denotes the approximation and W (r − r  , h) is the weight function with h standing for the smoothing
length that is the influencing area of the weight function, also called smoothing kernel function in SPH. The weight function
is a smoothing function of |r − r  | and becomes a Dirac delta function when h→0. Basically, the kernel function must satisfy
that the integration of W over the entire domain is unity. In practice the kernel functions are similar to a Gaussian function
[36], although they are usually vanished when |r − r  | is sufficiently large (2 h or 3 h). More details will be discussed in the
Section 6.
After completing the kernel approximation for the field function, the kernel approximation about the function derivation
should be performed. According to the Gauss integral formula, Eq. (2) can be further changed to,
  
∇ · f (r ) = [∇ · f (r )]W r − r  , h dr 
         
= ∇ · f (r )W r − r , h dr − f (r )∇W r − r , h dr = − f (r )∇W r − r , h dr . (3)
  
Because fluid is represented by a set of particles in SPH, particle j has mass mj , density ρ j and position rj , respectively.
Eqs. (2) and (3) can be further approximated using the interpolant formula, which is called particle approximation in SPH.
The interpolant formula for ƒ(r) and ∇ · f(r) can then be respectively transferred to the integral interpolant forms,
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9627


N     mj
f ( ri ) = f r j W ri − r j , h , (4)
j=1
ρj

N
mj    
∇ · f ( ri ) = f rj ∇ W ri − r j , h , (5)
j=1
ρj
where N is the amount of particles in the calculation region. Eqs. (4) and (5) constitute the basic formula of the SPH method.
With these two equations, the function value and the gradient value about the particle i can be calculated using other
particles j in the support region of the kernel function of particle i.

3. Governing equations and discretization

It is well known that for the fluid dynamics, the governing equations include the continuity equation, namely mass
conservation equation, and momentum equation, so-called Navier–Stokes equation. These two equations with respect to the
Lagrangian forms are as follows.

Continuity equation:
dρ
= −ρ∇ · v. (6)
dt
Momentum equation:
dv 1
= −ρ∇ p + ∇ (μ∇ · v ) + f , (7)
dt ρ
where ρ is density (kg/m3 ), p is pressure (Pa), μ is dynamic viscosity (Pa s), f is the external force (m/s2 ). By using the
approximations of the SPH method, these two partial differential equations can be converted to ordinary differential equa-
tions in the discretized form with respect to time, namely the SPH discretized equations. Actually, prior to the discretization
of these two equations, the approximation of the SPH method needs to be firstly discretized. According to the differential
quotient or product expansion form, Eq. (5) can be rewritten by introducing the density to the differential operator with the
following two formulas [36],

1 
N    
∇ · f ( ri ) = m j f r j − f (ri ) · ∇iWi j , (8)
ρi j=1

  

N
f rj f ( ri )
∇ · f ( r i ) = ρi mj + · ∇iWi j . (9)
j=1
ρ 2j ρi2
Compared with Eq. (5), Eqs. (8) and (9) bring about a good feature, that is, the gradient term is expressed in the form of a
pair of particles. They are fully conservative due to the pairwise anti-symmetric formulation. Therefore, these two formulas
are more accurate than Eq. (5) [49]. Because of this, the direct use of Eq. (5) is rarely found in the discrete equations
currently.
In addition, because the second-order derivative is included in the momentum equation, how to discretize the second-
order derivative is of importance for using the SPH method. One way is to directly use the formula (8) or (9) for two
times [50–52], which involves the nested sum over the particles. Another way is to employ second-order derivative of
smoothing kernel function [53], but such way may result in larger error at low resolution, especially for low-order spline
kernels [54,55]. In order to improve the accuracy of the second-order derivative, Monaghan [56] developed a new approach
to discretize the second-order derivative in the simulation of heat conduction. The approximate formula is as follows,

   

N
mj f ( ri ) − f r j ri − r j
 f ( ri ) =   · ∇iWi j . (10)
j=1
ρj ri − r j 2
Morris et al. [55] also used this method to calculate the second-order derivative items for simulating incompressible flow
with low Reynolds number. In this method, a standard SPH first-order derivative and a finite difference approximation of
first-order derivative are combined together. Therefore, they also called it as hybrid expression. Following this idea, many
researchers used it to discretize the second-order derivative [57–59].
Based on the above discretization methods, the continuity and momentum equations can then be discretized. In the SPH
method, there are two typical forms of the continuity equation. One is the sum of particles using Eq. (4) when the field
function is the density, the other is to use Eq. (8) to derive the continuity equation. These two forms are as follows.


N
ρi = m jWi j . (11)
j=1
9628 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

d ρi  N
= m jvi j ∇iWi j . (12)
dt
j=1

For Eq. (11), it strictly follows the mass conservation, but may produce the error at the interface because of the density
drop near the interface of the fluid [36]. To handle this error, two ways have been proposed. One is the normalized form
by summing the smoothing function itself [60,61] and the other is to use Moving Least Squares (MLS) developed by Dilts
[62].
For the momentum equation, the situation becomes more complex and different terms have different forms. Regarding
the term of the pressure gradient, three forms are usually used [22,63].
⎧ N  
⎪  pj
⎪
⎪− mj
pi
+ ∇iWi j
⎪
⎪ ρi2 ρ 2j
⎪
⎪  
⎪
⎨ 
j=1

1
N
m j pi + p j
− ∇p = − ∇iWi j . (13)
ρ ⎪
⎪ ρi ρ j
⎪N
⎪
j=1
 
⎪
⎪ m j pi − p j
⎪
⎪ ∇iWi j
⎩ ρi ρ j
j=1

For the viscous term, the above-mentioned hybrid method is usually employed, which can be presented as follows
[37,55],
⎧N

⎪   1 1 ∂ W
⎪ m j μi + μ j
⎪ v  
⎨ ρi ρ j i j ri j  + η ∂ ri j
∇ (υ∇ · v ) = Nj=1
    , (14)
⎪
⎪ 2 μi μ j 1 1 ri j · ∇iWi j
⎪
⎩ m j μ + μ ρ 2 + ρ 2 vi j r2 + η2
i j i j ij
j=1

where ƞ is a small parameter used to smooth the singularity when rij is close to zero [64,65]. In general, ƞ = 0.01 h and ƞ2
= 0.001h2 are typically employed. The selection of them can be determined by the demand of the simulation.
It should be noted that only the continuity and momentum equations are not closed because the pressure term in the
momentum equation is unknown. For compressible flow, the pressure of fluids can be got from the state equation of fluids.
For incompressible flow, two ways can be used to solve this problem. One is to introduce the state equation to describe the
relationship between the pressure and density, which is called weakly compressible SPH (WCSPH). The other is to solve the
partial differential equation for pressure (Poisson equation) through the projection method, which is named as incompress-
ible SPH (ISPH). This two pressure treatments for incompressible flow will be discussed in the next section.

4. Pressure treatment

As mentioned above, there are two ways used to determine the pressure, i.e., WCSPH and ISPH. In the following, the
detailed discussions about these two methods will be given.

4.1. WCSPH

WCSPH is to introduce the state equation to describe the relationship between the pressure and density. Presently, the
most commonly-used state equation is the Tait equation [66],
 ρ γ 
p = p0 −1 , (15)
ρ0
where γ is a constant ranging from 1 to 7 and p0 = c02 ρ0 /γ with c0 denoting the speed of sound at the reference density
ρ 0 . Here, the realistic speed of sound cannot be used because the associated time step imposed by the Courant–Friedrich–
Levy (CFL) condition is rather small, making the computational cost high. In the work by Monaghan and Kos [67], they
artificially slowed down the sound speed, by which the satisfied results on the fluid propagation were obtained. The sound
speed should be at least 10 times higher than the maximum fluid velocity [67]. Hence, from the viewpoint of the numerical
stability, the sound speed has a direct effect on the permissible time-step, which affects the computational cost [68].

4.2. ISPH

The principle of ISPH is to solve the partial differential equation for the pressure through the projection method. The
projection method is originally proposed by Chorin [69,70] and firstly implemented to the SPH method by Cummins and
Rudman [71], termed as the standard projection method. Some researchers [72–75] have also improved and modified the
projection method to make it more accurate and efficient. Compare to WCSPH, ISPH is a typically implicit by dealing with
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9629

Table 1
Three algorithms of ISPH.

Methods Algorithms

ISPH_DF Step 1:ri = rin + vi t;

1 n
Step 2: vi = vni + ∇ (μ∇ · v) + f t;
ρ
ρ i 
Step 3: pni +1 ← ∇ · (∇ pni +1 ) = ∇ · vi ;
 t
Step 4: vni +1 = vi − (∇ p/ρ )ni +1 t;
Step 5: rin+1 = rin + (vni + vni +1 )t /2;
Step 6: n = n + 1 t = t + t .
1 n
ISPH_DI Step 1: vi = vni + ∇ (μ∇ · v) + f t;
ρ i
Step 2: ri = rin + vi t;
ρ0 − ρ  i
Step 3: pi ← ∇ · (∇ pi ) =
n+1 n+1
;
ρ0 t 2
Step 4: vni +1 = vi − (∇ p/ρ )ni +1 t;
Step 5: rin+1 = rin + (vni + vni +1 )t /2;
Step 6: n = n + 1 t = t + t .
1 n
ISPH_DFDI Step 1: vi = vni + ∇ (μ∇ · v) + f t;
ρ i
Step 2: ri = rin + vi t;
Step 
3: ρi ← ri ; 
ρ0 − ρ  i
Step 4: pni +1/2 ← ∇ · (∇ pni +1/2 ) = ;
ρ0 t 2
Step 5: ri = ri − (∇ p/ρ )ni +1 t 2 /2;
Step 6: ρi ← ri ;
Step 7: Repeating step 4 to 6, until that the relative density difference is less a
certain criterion (typically 1%);
1 
Step 8: vi = vi + ∇ (μ∇ · v) + f t;
ρ
ρ i 
Step 9: pni +1 ← ∇ · (∇ pni +1 ) = ∇ · vi ;
 t
Step 10: vni +1 = vi − (∇ p/ρ )ni +1 t;
Step 11: rin+1 = rin + (vni + vni +1 )t /2;
Step 12: n = n + 1 t = t + t .

the pressure and velocity as primitive variables. In ISPH, the density is constant for incompressible flow. The continuity
equation is firstly reduced to the divergence-free velocity field as follows,

∇ · v = 0. (16)

With the divergence-free velocity field and the momentum equation, the velocity-pressure coupling problem can then
be solved by the projection method. Solving the momentum equation consists of two steps. The first step is the prediction
step based on the viscous and body forces to achieve the auxiliary velocity. The expression of the auxiliary velocity vn+1 is
as follows,
1 
vn+1 = vn + ∇ (μ∇ · vn ) + fn t. (17)
ρ
The second step is the correction step based on the pressure gradient,
1
vn+1 = vn+1 + ∇ pn+1 · t. (18)
ρ
Substituting Eq. (18) into Eq. (16) leads to Poisson equation for pressure. The expression is shown as follows,
ρ
∇ · (∇ pn+1 ) = ∇ · vn+1 . (19)
t
By doing this, the pressure can then be determined by Eq. (19) and the velocity field is then updated by Eq. (18).
It is clear that the main idea of the above algorithm is based on the velocity divergence-free such that it is called
as incompressible divergence-free SPH method and referred to as ISPH_DF. In addition, Eq. (16) implies that the velocity-
divergence-free condition comes from zero density variation. Following this idea, Shao and Lo [76] proposed a projection-
based incompressible method to impose density invariance. This method is so-called density-invariant SPH method and re-
ferred to as SPH_DI. Besides, Hu and Adams [77] suggested that a divergence-free and density-invariance algorithm should
be applied. This requires two pressure Poisson equations to be solved in this algorithm. As a result, this combined in-
compressible SPH method is referred to as ISPH_DFDI. The ISPH algorithms about these three methods are summarized in
Table 1.
These three methods about ISPH have their own advantages and disadvantages. Comparing with ISPH_DI, ISPH_DF pro-
vides more accurate prediction, but its stability is lower because the particle spacing easily becomes distorted and errors
9630 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Fig. 1. Truncated support domain for boundary particle.

caused by the anisotropic particle spacing may be accumulated [78]. Although ISPH_DFID combines the advantage of the
accuracy from ISPH_DF and the stability from ISPH_DI, the computational cost becomes rather high because of usually-
required multi-internal iterations and solving two pressure Poisson equations. In some cases, the computational cost may
be five times higher [78].
In general, WCSPH can show the advantages over ISPH in some aspects, including the ease of programming and better
ordered particle distributions [68]. That is because solving the Poisson equation in WCSPH is not required. Moreover, it is not
straightforward in most cases of the multiphase flow especially involving wave breaking and loss of continuity in the fluid
because solving the pressure Poisson equation is rather difficult [79]. For this reason, WCSPH is more widely used at present.
However, some researchers [76,80–82] suggested that ISPH was more accurate especially in the pressure representation. The
reason is that when handling fluid flow with larger Reynolds number (typically>100), the standard WCSPH method has
been found to suffer from large density variations. As such, it is required to use a much smaller Mach number than 0.1 to
avoid the formation of unphysical void regions in the computational domain [55,68,83]. Hu et al. [77] also pointed out that
WCSPH was computationally less efficient than ISPH when the flow is not dominated by the viscous force or the surface
tension in the case of the fluids with moderate density ratios.

5. Boundary condition treatment

In order to solve the governing equations, appropriate implementations of boundary conditions are required. Although
SPH has been developed for many years, the treatment of solid wall boundary is still one of the main difficulties in the
SPH applications [84,85]. This difficulty is caused by two reasons. Firstly, the boundary is complex in engineering problems.
Secondly, for those particles close to the boundary, the kernel will be truncated by the boundary, as illustrated in Fig. 1.
Therefore the accuracy of the SPH approximation is reduced. In this case, the proper and correct boundary treatments have
been an ongoing concern for the accurate and successful implementation of the SPH method [85–87].
To well implement the solid boundary conditions and improve the numerical accuracy, many researchers have proposed
different boundary treatment algorithms for the SPH method. At present, there are four main methods to treat the boundary
conditions, including boundary particle force model, image particle model, dynamic particle model (also termed as coupling
boundary model), and semi-analytical model (also termed as normalizing condition model), which will be discussed in the
following.

5.1. Boundary particle force model

The basic idea of the boundary particle force model is that the boundary particles are assigned at the boundary zone
and these particles exert the forces on the inside fluid particles. Monaghan [66] firstly proposed a set of particles fixed at
the boundary to prevent the fluid particles from non-physical penetration by the repulsive force, which is similar to the
Lennard–Jones (L–J) equation,
  n1  n 2   
K r0
− r0 ri j ri j  < r0
f B = | |
ri j ri j 2
ij
ri j   , (20)
0 ri j  ≥ r0
where the values of n1 and n2 are often 12 and 4, respectively, and rij is the distance of particle i and j, r0 is the initial
space between particles, and K is commonly taken as the square of the maximum velocity [88]. If K is too large, strong
force exerted by boundary particles will be generated, leading to the calculation failure. It should be pointed out that it
is difficult to prevent fluid particles from penetrating through the boundary if K is not large enough. Moreover, this force
model is sensitive and strong so that it is easy to produce disturbance during the calculation. In order to overcome the
disadvantage of the L–J force model, Monaghan et al. [89] presented another repulsive force model,

f B = K ri j 2m j
β r 2 i j mi + m j
ij W , (21)
ij

where the constant β ensures the invariance of the forces acting on the fluid particles in the case that the spacing of the
boundary particles is changed. For example, if the spacing is halved, the number of boundary particles will be doubled and
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9631

Fig. 2. Diagrams of three image particle models.

simultaneously β is also doubled. K is another constant parameter depending on the simulating problem. Therefore, this
force model is soft unlike L–J equation.
In the above two force models, the boundary particle spacing is usually required to be smaller than the fluid particle
spacing, otherwise the fluid particle may cross over the boundary through the space between the boundary particles. Only
when these forces are chosen correctly and the boundary particle spacing is small than the fluid particle spacing with a
factor of 2 or more, the total boundary forces on a fluid particle can then be perpendicular to boundaries with negligible
error.
Qiang et al. [90] proposed a new repulsive force model based on the Galerkin function. And then Han et al. [88] presented
a corrected model as follows,
⎧
    D 
⎪
⎨ min vi − v j · n
 j , −1 hi + h j  
− v2max   Wi j n
j vi − v j · n
j < 0
f B = r · n j  2 , (22)
ij
⎪
⎩  
0 vi − v j · n
j < 0
where n j is unit normal vector of boundary particle j, vi and v j are the velocity of fluid particle i and boundary particle j, h
and D are the smoothing length and the number of dimension of simulating problem. Comparing with the above two force
models, this repulsive force is inversely proportional to |ri j · n
 j | instead of |rij |, which can prevent unphysical penetration in
the case of less boundary particles. Another new approach [70] is to set an interpolation procedure for the force normal to
the boundary exerting on the fluidic particle. This model is improved by Rogers and Dalrmple [91],

ij
 j R(y )P (x )ε (z, u⊥ ),
f B = n (23)
where n  j is the unit normal vector of the boundary particle, R is the repulsion function, P is a corrected function, ε is a
term to adjust the magnitude of the force according to the local fluid depth z and velocity of the fluid particle normal to
the boundary u⊥ , y is the perpendicular distance from the fluid particle to the wall, x is the distance between the location
of fluid particle projected at the boundary and the boundary particle. R(y) is determined in terms of the normal distance
to the wall. The function P(x) is designed to ensure that the forces acting on the fluid particles are constant when the
fluid particles move in parallel with the boundary. It is clear that not only the distance but also the normal and tangential
direction are necessary to be determined before the repulsive force calculation.

5.2. Image particle model

The image particle model is also called as ghost particle [92] or mirror particle model [93]. In this model, the image
particles are outside of the fluid but alongside of the boundary. These particles have the same density, pressure and temper-
ature as the fluid particles. If there is an external force acting on the fluid, the pressure for the image particles may need to
be adjusted [93]. For the velocity, these image particles have the perpendicular component of the velocity of fluid particle
with the opposite sign, and the tangential component with the same sign (slip) or opposite sign (non-slip), as illustrated in
Fig. 2a. The image particles are generated by mirroring or reflecting fluid particles along the solid boundary in each step of
the calculation. The disadvantage of using these image particles is that when the boundary changes sharply, how the image
particle should be placed becomes an issue.
In the method by Morris et al. [55], the locations of the image particles are unchanged with time. The velocities of the
image particles are obtained by linear extrapolation from the velocity of fluid particles close to the wall, depending on the
normal distances from the fluid particle and image particle to the wall, respectively, as shown in Fig. 2b. The velocity of the
image particle can be expressed as vi = v f + (1 + di /d f )(vw − v f ), the subscript i, f and w represent the image particle, fluid
particle and wall boundary, respectively. In this way, the normal projection of the image particle onto the boundary is not
coincident with the corresponding normal projection of the fluid particle onto the boundary. This feature may lead to error
9632 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

of the extrapolation. In order to reduce the extrapolation error, Fang et al. [94] proposed another method by adding a series
of fixed virtual particles inside the domain so that the normal projection is coincident (see Fig. 2c). The velocities of the
virtual particles are also obtained by the interpolation, such as the MLS method, from the surrounding fluid particles.

5.3. Dynamic particle model

Dynamic particle model is also called as coupling boundary model by some researchers [95]. The idea of this model has
been mentioned by Liu and Liu [21], which has been successfully employed in modeling the free surface flow by Gomez
et al. [96]. In this model, the virtual particles are also introduced and fixed at the outside of the solid boundary. The thick-
ness of the virtual boundary particle located region is determined by the smoothing kernel function and initial particle
spacing. The virtual particles are employed to approximate the variables of the fluid particle. Thus the accuracy is improved
by compensating the boundary deficiency. At the same time, the field variables of the virtual particle are obtained from the
governing equation, such as the continuity equation and the equation of state. It should be noted that although the accuracy
is improved, the problem of the boundary deficiency for the virtual particles still exists, which can lead to inaccurate results
associated with the pressure oscillations. In order to address this issue, Gong et al. [95] proposed a method to smooth the
pressure field, which has shown better results by removing the pressure oscillations.

5.4. Semi-analytical model

The semi-analytical model is based on the introduction of a renormalizing factor to reconstruct the governing equations
for the fluid particle close to the boundary. This factor depends on the boundary shape and the position of a fluid particle
relative to the wall. Kulasegaram et al. [86] developed a formulation to determine the renormalizing factor accounting for
the missing area of the kernel support. This model was then improved by Ferrand and his co-work [97] by solving a dynamic
equation for the renormalizing factor. Mayrhofer et al. [98] then extended it to 3D model. The discretization of the governing
equations can be expressed as follows.
The operator of the gradient for an arbitrary field function f can be presented as,
   
ρi  fi fj ρi  f i fs
∇ fi = m + ∇ Wi j − + ρ ∇γ . (24)
γi j∈F j ρi2 ρ 2j γi s∈S ρi2 ρs2 s as
The operator of the divergence for an arbitrary field vector can be presented as,
1  1 
∇vi = − m jvi j  ∇ Wi j + ρsvis ∇ γis . (25)
γi ρi j∈F
γi ρi s∈S

The Laplacian operator can be expressed as,

1 1  μi + μ j vi j 1 
∇ · (μ · ∇vi ) = mj ri j · ∇ Wi j − |∇ γis |(μi ∇vi + μs ∇vs ) · n s . (26)
ρi γi j∈F
ρi ρ j r i j
2 γi ρi s∈S
In the above equations, the correction parameters are defined,

γi = W (r − ra )dr, (27)

 re2

∇ γis = W (r − ra )dl n
s. (28)
re1

Different types of the elements in this model are shown in Fig. 3. ∇γ is is defined as the contribution of the segment
s with the value of the gradient of γ i at the position ra , determined by a polynomial approximation [86,99], an analytical
solution [87] and a discrete summation over the boundary [100]. In order to more easily account for complex shape of the
boundary, in the work by Ferrand and his co-work [97], the computation of the renormalization term of the kernel support
near a solid wall is obtained with a time integration scheme.
In summary, a good boundary treatment algorithm should consider the accuracy, efficiency and adaptivity to complex
boundaries. It is clear that the repulsive force solid boundary treatment is conceptually simple, easy for enforcement and
preventing unphysical penetration, and adaptive to complex solid boundaries. But the accuracy is still poor because of ex-
isting boundary deficiency problem. Meanwhile, the coefficient of the specific repulsive force has an important effect on
the accuracy. The image particle algorithm improves the numerical accuracy by supplementing the support domain of fluid
particles with image particles. However, the image particle algorithm is usually applicable to simple boundaries to generate
the image particles or place the fixed image and virtual particles. For the dynamic particle algorithm, the computational
accuracy is improved by removing the deficiency problem for fluid particles, which is suitable for complex solid bound-
aries. In many cases, however, unphysical penetration of fluid particles through the solid wall may occur. Besides, existing
algorithms for approximating field variables of the virtual particles usually employ the conventional SPH method, leading
to poor accuracy. With respect to the semi-analytical boundary algorithm based on computing the surface integral, both
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9633

Fig. 3. Illustration of different types of elements.

linear and angular momentum of the governing equations are preserved. In this case, it is difficult to allow this model to be
incorporated with other SPH formulations. As a result, these four algorithms have their own advantages and disadvantages.
A universal model that gives accurate results for all free-slip, non-slip, partial slip boundary conditions is still unavailable
[101]. The development of new boundary treatment algorithms is one of the important research directions for the SPH
method.

6. Kernels and interpolation methods

6.1. Kernel functions

As mentioned above, the smoothing kernel function is closely related to the computational accuracy and stability of the
SPH method. For this reason, the study on the smoothing kernel function is reviewed in this section. At present, there are
many kinds of smoothing kernel functions used in the SPH method. Various requirements and properties for the smoothing
kernel functions have been discussed. Major properties or requirements are now summarized in the following.
(i) The smoothing kernel function must be normalized over its support domain.
(ii) The smoothing kernel function should be compactly support.
(iii) The value of the smoothing kernel function within the support domain is positive. This property is not mathematically
necessary under the convergent condition but important to ensure a meaningful representation of some physical
phenomena.
(iv) The smoothing kernel function value for a particle should be monotonically decreasing with increasing the distance
away from this particle.
(v) The smoothing kernel function should be an even function.
(vi) The smoothing kernel function should be satisfied with the Dirac delta function condition as the smoothing length
approaches zero.
(vii) The smoothing kernel function should be sufficiently smooth.
Theoretically, any function having the above properties can be used as smoothing kernel function in SPH. Liu et al.
[102] presented the method to construct the smoothing kernel function. Many open literatures have reported many kinds
of smoothing kernel functions. The main commonly-used smoothing kernel functions and their characters are summarized
in Table 2. If the Nth order derivative of smoothing kernel function is continuous and non-zero, this function is termed
as Nth order smoothing kernel function. For example, the new quartic function is 4th order smoothing kernel function.
In this aspect, high order is more beneficial for improving the stability and accuracy. The Gaussion kernel is used in the
original paper of Gingold and Monaghan [25], and this kernel is sufficiently smooth even for critically high-order derivatives.
However, it requires large support domain to ensure the value of smoothing kernel function to approach zero in bound.
In order to overcome this problem, the super-Gaussian kernel is presented. However, this smoothing kernel function is
negative in some region of its support domain. It is noted that unlike other smoothing kernel functions, the derivatives of
the smoothing kernel functions for No. 6, 10 and 12 monotonically decreases as the particles move closer, and they can
relieve the problem of compressive instability [39]. However, their derivative values at the point q = 0 is not equal to zero,
which reduces the suitability of the smoothing kernel functions. In addition, the derivative of the smoothing kernel functions
of Quadriatic 2 monotonically increases as the particles move closer, and it can relieve the problem of tension instability
[39]. Some researchers [50] have given some criterions to evaluate the smoothing kernel function. However, the criterion for
determining suitable smoothing kernel function is still uncertain.
9634 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Table 2
Commonly-used kernel functions.

No Name Expression (q = |r|/h) Characteristics

Smooth Non- negativity ∂ W /∂ q

1
e−q ∂ W /∂ q|q=0 = 0
2
1 Gaussian √ Sufficient Yes
πh
1
√ (1.5 − q2 )e−q ∂ W /∂ q|q=0 = 0
2
2 Super Gaussian Sufficient No
π
⎧
⎨2/3 − q2 + 0.5q3 0≤q<1
1
3 B-spline ( 2 − q )3 /6 1≤q<2 2nd order Yes ∂ W /∂ q|q=0 = 0
h⎩
0 q≥2
⎧
⎪ (3 − q )5 − 6(2 − q )5 + 15(1 − q )5 0 ≤ q < 1
⎨
1 ( 3 − q )5 − 6 ( 2 − q )5 1≤q<2
4 Quintic- Spline 4th Order Yes ∂ W /∂ q|q=0 = 0
120h ⎩⎪ ( 3 − q )5 2≤q<3
0 q≥3
5 Quartic spline ⎧ 3rd order Yes ∂ W /∂ q|q=0 = 0
⎪ (2.5 − q)4 − 5(1.5 − q)4 + 10(0.5 − q)4 0 ≤ q < 0.5
⎨
1 ( 2.5 − q )4 − 5 ( 1.5 − q )4 0.5 ≤ q < 1.5
⎩ ( 2.5 − q )
24h ⎪ 1.5 ≤ q < 2.5
4

0 q ≥ 2.5
3 ( 2 − q )2 0≤q<2
6 Quadriatic 1 2nd order Yes ∂ W /∂ q|q=0 = −0.75
16h 0 q≥2
 Monotonic
3 4 − q2 0≤q<2
7 Quadriatic 2 2nd order Yes ∂ W /∂ q|q=0 = 0
32h q≥2
 Monotonic
1 ( 2 − q )2 ( q + 1 ) 0≤q<2
8 Thrice 3rd order Yes ∂ W /∂ q|q=0 = 0
8h 0 q≥2

3π 2 (1 − q2 /4 )2 (1 + cos(π q/2 )) 0≤q<2
9 Cosin Sufficient Yes ∂ W /∂ q|q=0 = 0
8h ( π 2 + 3 ) 0 q≥2
 1 −6
1 + q16 0≤q<2
10 1/X, 2 2+q Sufficient Yes ∂ W /∂ q|q=0 = −1.38
h (ln 4 − 1.25) 0 q≥2
2 Monotonic
9 19 3 5 4
1 3 − + q − q 0≤q<2
11 New quartic 8 24 32 4th order Yes ∂ W /∂ q|q=0 = 0
h 0 q≥2
[102] ⎧
⎨q3 − 6q + 6 0≤q<1
1
12 Cubic spline ( 2 − q )3 /6 1≤q<2 2nd order Yes ∂ W /∂ q|q=0 = −6/7
7h ⎩
[103] 0 2≤q Monotonic

1 4 cos(π q/k ) + cos(2π q/k ) 0≤q<k
13 Double cosine Sufficient Yes ∂ W /∂ q|q=0 = 0
6kh 0 q≥k
[104]

6.2. Improvement of interpolation approximation

The SPH approximation can cause the inaccuracy in the zone which is near the boundary or free surface or when the
distribution of particles is irregular. Therefore, extensive efforts have been devoted to the improvement of the SPH approxi-
mation, which will be discussed as follows.

6.2.1. Kernel and kernel gradient correction

In order to improve the accuracy of the SPH approximation, Randles and Libersky [38] derived a normalization formula-
tion for the function approximation.
N mj
j=1 ρ j f j Wi j
fi = N m j . (29)
j=1 ρ j Wi j

Correspondingly, the corrected kernel gradient should be used to calculate the force in the governing equation instead of
the normal kernel gradient. The corrected kernel gradient is as follows,

∇ W̄i j = L−1
i
∇ Wi j , (30)


N
mj
Li = ∇iWi j  r ji . (31)
j=1
ρj
Since the 2×2 corrected and inverse matrix for two-dimensional problems and 3×3 matrix for three-dimensional prob-
lems need to be calculated in this corrected method, the computational cost is usually high. Another corrected kernel and
kernel derivative correction was developed by Chen et al. [60,61], based on Taylor series expansion for the SPH approx-
imation of a function. For this method, the kernel approximation of a function is the same as Eq. (29). And the kernel
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9635

Fig. 4. Particle approximations for a function with a discontinuity [39].

approximation of a function derivative is,



N 
  mj
 N   mj
fix = fi − f j Wi jx xi − x j Wi jx . (32)
j=N
ρj j=1
ρj

Compared with the corrected method by Randles and Libersky [38], this corrected method is simpler. Generally, both
methods are referred to as CSPH, which have better accuracy than the conventional SPH method. It is also reported that
both methods can reduce the tensile instability.

6.2.2. An SPH formulation for discontinuity

In the method developed by Chen et al. [60,61], Taylor series analysis on the kernel approximation of a field function
and its derivatives are required. To this end, the function should be sufficiently smooth. As a result, CSPH is not applicable
to problems with discontinuities. Liu and his co-workers [105] proposed a method for SPH to resolve discontinuity prob-
lems, and called it as a discontinuous SPH, or DSPH. The detail for one-dimensional case can be referred to the references
[21,39] and multi-dimensional DSPH method can be found in the reference [106]. Taking one dimensional problem as an ex-
ample, the approximation of this method is presented as follows. First of all, the particle approximation of the discontinuous
function is performed by,
N mj N mj
j=1 ρ j f j Wi j ( fk − fi ) j=k ρ j f jWi j
f i = N mj − N m j . (33)
j=1 ρ j Wi j j=1 ρ j Wi j

Similarly, the particle approximation of the derivative of the discontinuous function is done by,
N   m
 N mj N      m

j=N fi − f j Wi jx ρ jj ( fk − fi ) j=k Wi jx ρ j j=k x j − xk fkx − x j − xi fix Wi jx ρ jj
fix = N   m − N   m + N   mj . (34)
j=1 xi − x j Wi jx ρ jj j=1 xi − x j Wi jx ρ jj j=1 xi − x j Wi jx ρ j

Clearly, both the particle approximations are composed of two parts. One is the first primary part similar to the CSPH
approximation and the other one is the second additional part for the discontinuity treatment. The summations of the
numerators in the additional part are only carried out for particles in the right portion of the support domain shaded in
Fig. 4. In this method, when the kernel approximation shifts to particle approximation, an arbitrary point xk is in association
with a particle k, which is the nearest particle on the right-hand side of the discontinuity, as shown in Fig. 4. Only the
particles, xj (j = k, k+1,..., N) that satisfy (xk − xj )(xk − xi ) ≤ 0, need to be considered. As a result, this method is beneficial not
only for simulating the discontinuity problem but also for remedying the boundary deficiency problem.

6.2.3. Reproducing kernel particle method (RKPM)

As one of the earliest meshless methods, the accuracy and instability are low especially near the region of boundary. To
overcome this problem, Liu et al. [31] proposed the reproducing kernel particle method (RKPM) by introducing a corrected
function for the integral transformation in the SPH method to satisfy the boundary compatibility, and the instability of the
SPH method can be eliminated. The idea of RKPM is to introduce a corrected function for the integral transformation in the
SPH method, which can be expressed as follows,
    
f¯(r ) = f r  W r ; r − r  dr  , (35)


where f¯ is the reproduced function of f. W (r ; r − r  ) is the modified smoothing kernel function and expressed by,
     
W r ; r − r  = C r ; r − r  Wa r − r  , (36)
 
  1 r−r 
Wa r − r  = W , (37)
a a
9636 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

where a is the dilation parameter of the kernel function Wa (r − r  ), C (r ; r − r  ) is the correction function and obtained by
imposing the reproducing conditions such that the reproduced Eq. (36) can be exactly polynomial. The correction function
is proposed by Liu et al. [31] and expressed by a linear combination of polynomial basis function in the following form,

  
N    
C r; r − r = bi (r ) r − r  ≡ H T r − r  b(r ), (38)
i=1

where
      2  N 
H T r − r  = 1, r − r  , r − r  , . . . , r − r , (39)

bT (r ) = [b0 (r ), b1 (r ), · · · , bN (r )], (40)

H is a vector of basic function and bi (r) is a function of r to be determined. Considering Taylor series expansion of f(r),
the transposed matrix, bT (r), can be achieved by,
M (r )b(r ) = H (0 ), (41)
where M(r) is the moment matrix defined by
      
M (r ) = H r − r  H T r − r  Wa r − r  dr  . (42)

Solving b(r) from Eq. (41), the correction function can then be obtained,
   
C r ; r − r  = H T (0 )M−1 (r )H r − r  . (43)
More detail can be found in the references [31,107]. In general, RKPM can provide a multi-scale decomposition to enhance
the physical interpretation.

6.2.4. Finite particle method

Liu et al. [108,109] proposed another particle consistency restoring approach to improve the precision of traditional SPH,
which retained the conventional smoothing function. This approach is finite particle method, termed as FPM, in which a set
of basic functions is used in the numerical approximation. Similarly, FPM is also based on the Taylor series expansion. The
Taylor series expansion for a function f(r) in a multi-dimensional space is as follows,
  β

  r α − riα r β − ri
f (r ) = fi (r ) + r α − riα fi,α + fi,αβ + · · · . (44)
2!
where α and β are the dimension indexes ranged from 1 to D (D is the number of dimensions). Multiplying both sides
of Eq. (44) with a smoothing function Wi (x) and its first order derivatives Wi,β , and neglecting the second and high or-
der derivatives, and then using the particle approximation to present the expression, the above equation can become the
following equations, respectively.


N
mj   
N
mj
N 
  mj
f r j Wi j = fi Wi j + fi,α r α − riα W , (45)
j=1
ρj j=1
ρj j=1
ρj ij

N
mj   
N
mj
N 
  mj
f r j Wi j,β = fi Wi j,β + fi,α r α − riα W . (46)
j=1
ρj j=1
ρj j=1
ρ j i j,β
It can be seen that there are D+1 equations for D+1 unknowns (fi and fi,α ). Therefore, Eqs. (45) and (46) are closed for
solving fi and fi,α . The approximations of fi and fi,α in the continuous form (kernel approximation) are
   
 N N   mj    N N   mj 
 f jWi j mρ j α α
r j − ri Wi j ρ j    mj
r j − ri Wi j ρ j 
α α
 j=1   j=1 Wi j ρ j 
   ,
j
j=1 j=1
fi =  N N     N   (47)
    mj 
N
 f jWi j,β mρ j α α
r j − ri Wi j,β ρ j
mj
  mj
r j − ri Wi j,β ρ j 
α α
 j=1 j
j=1
  j=1 Wi j,β ρ j j=1

   
 N N   N   m N 
 f jWi j mρ j Wi j ρ jj 
m
r αj − riα Wi j ρ jj Wi j ρ jj 
m
 j=1   
fi,α =  N   j=1 .
j
j=1 j=1
     (48)
  f jWi j,β mρ j  Wi j,β mρ j    rαj − riα Wi j,β mρ j  Wi j,β mρ j 
N N N

 j=1 j
j=1
j   j=1
j
j=1
j 

Finally, the basic expressions of FPM are obtained, in which no modification for conventional smoothing function is
made. FPM shares some common features with CSPH since they both use Taylor series expansion on the original SPH kernel
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9637

approximations. However, FPM has the first order kernel consistency and first order particle consistency for both the interior
and boundary regions. Hence, its consistency is higher than CSPH that has the first order kernel consistency for the interior
regions and zero order kernel consistency for the boundary region.

6.2.5. Moving least square particle hydrodynamics

One simple way to improve the accuracy of the SPH kernel function is to adopt a moving least square (MLS) interpolant
as the SPH kernel function. By using a MLS interpolant, Dilts [62] used a collocation/kernel Galerkin approximation to derive
a so-called moving least square particle hydrodynamics (MLSPH). The basic idea of MLSPH is introduced as follows. Firstly,
f(r) is approximated using fh (r),

N
f h (r ) = φ j (r ) f j , (49)
j=1

where φ j (r) is the shape function and can be calculated by,

 
φ j (r ) = PT (r )A−1 (r )P r j W j (r ), (50)


N    
A (r ) = P T r j P r j W j (r ). (51)
j=1

In the above expressions, P(r) is the basic function vector, which is typically the quadratic polynomial basic function
vector in the two dimensional problem. Wj (r) is the smoothing function at the position of rj . After that, some definitions in
association with φ j (r) are introduced.

The volume of the MLS particle, Vi =  φi (r )dr.

The area matrix of MLS, Ai j =  φi φ j dr.
 0 / ∂  or
i∈ /∂ 
j∈
The boundary matrix of MLS, Bi j = ∂  φi φ j n
 ds = { , where n
 is the exterior normal unit vector of
Ai j +A ji otherwise
surface ∂ . Finally, the approximation of the governing equation in MLSPH is obtained.
⎧  N 
⎪ dρ ρi    
⎪
⎪ = vi − v j · Ai j − A ji + Bi j
⎨ dt 2mi
i j=1
1  . (52)
⎪ 1    
N
⎪
⎪ ∇ = + · − +
⎩ ρ p
2m
p i p j A ij A ji B ij
i i
j=1

The determination of Aij can refer to the references [65,110]. This method promises to greatly increase the dependability
of SPH. However, it should be noted that this original MLSPH is not conservative. Some modifications on the original MLSPH
are developed to make it locally conservative.

7. Modeling free surface flow

Free surface flow is a typical simulation case with the interface. Modeling the free surface flow is also briefly reviewed
in this paper. SPH was extended to simulate the free surface flow in 1992 [66]. In the context of the free surface flow, there
are two critical issues resulting from a free surface: (i) physical boundary conditions have to be satisfied at this surface,
and (ii) the interpolation accuracy has to be preserved in the region close to the boundary. Basically, the physical boundary
condition is naturally satisfied with the SPH method [111,112]. For the cases with considering the surface tension force, the
calculation of the surface tension force is the same as the multiphase flow, which will be discussed in next section.
When modeling free surface flow, there are some terms related to the free surface in both the pressure gradient and
velocity divergence, which need to be accounted. Therefore, some treatments have been developed to address the consis-
tency of kernel interpolation near the free surface. The evaluation on the treatments has been given in the literature [112].
In general, these treatments can be divided into two major groups [113].
The first one is to retain the kernel interpolation but attempt to address the inconsistency by reformulating the dis-
cretized forms of the governing equations, such as CSPH, FPM and DSPH. Since these methods have been introduced in
Section 6, the detailed descriptions are not repeated here. Basically, such formulations consider the mass conservation.
Moreover, the change of the particle density in kernel interpolation is caused due to the particle density normalization,
which decreases the consistency of kernel interpolation. Because existing formulations on the momentum equation focus
on the conservation of the global momentum and do not account for the incompleteness of kernel interpolation near the
free surface, the core issue of using kernel interpolation in the free surface problems is not addressed. It is noted that the
semi-analytical model introduced in Section 5.4 can be used to treat the free surface. However, the normal vector of the
free surface needs be calculated.
The second one is to modify or eliminate the use of kernel interpolation. For example, the kernel bandwidth in time is
evolved [114–116], which is typically used for shocks and has little apparent effect near the free surface. In addition, other
9638 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

unstructured spatial discretization schemes are used to replace kernel interpolation, e.g., MLSPH and RKPM, both of which
are to remain complete even in the presence of a free surface. However, these schemes may become locally singular in the
absence of sufficient neighboring particles so that further treatment is required [117], such as a local increase in the kernel
bandwidth and a local decrease of the order of the approximation [62].

8. Surface tension

In the simulation of the multiphase flow, including the gas–liquid and liquid–liquid two-phases, the surface tension is
one of the most important physical properties. Particularly at the micro scale, it becomes a dominant factor. Therefore, how
to accurately describe the surface tension is a critical issue in numerically studying the two-phase flow by SPH. Physically,
the surface tension is due to the unbalanced force of the molecules in the neighbor of the interface. To describe such a
physical phenomenon, there are three typical models of the surface tension in the SPH method, including Cohesive Pressure
Model (CPM), Inter-particle Force Model (IPF) and Continuum Surface Force Model (CSF). The first two models are also
regarded as inter-particle interaction force (IIF) model by some researchers. The detailed descriptions and discussions of
these three models are given in the following subsections. It is noted that the momentum conserving form of the pressure
gradient term actually has an inherent or inbuilt surface tension when it is used at the interface where the two phases
have different densities. This inbuilt surface tension is usually called artificial surface tension. However, this artificial surface
tension is different from the physical surface tension and difficult to physically describe the surface tension.

8.1. Cohesive pressure model

Cohesive pressure model is based on the van der Waals equation of state, by which the cohesive pressure term is pro-
posed on the consideration of the interactions between molecules. Because the surface tension is caused by the interactions
between molecules, the cohesive pressure term can be used to characterize the surface tension. For van der Waals liquid,
the pressure and internal energy can be characterized as follows.

ρ k̄B T
p= − āρ 2 , (53)
1 − ρ b̄
ξ
U= k̄B T − āρ 2 . (54)
2
In these two equations, ξ is the number of degrees of freedom for the molecules, k̄B = kB /m, b̄ = b/m and a = a/m,
where kB is the Boltzmann’s constant, b is a constant due to the finite volume of the molecules, a is a measure for the
forces of cohesion between neighboring molecules and m is the mass of the molecules. In Eq. (53), āρ 2 represents the force
of cohesion between neighboring molecules, which is unbalanced for the molecules at the interface region and thus results
in the surface tension. Nugent et al. [118] in 20 0 0 proposed to use the cohesive pressure term to represent the surface
tension. In this method, for internal particles, the cohesive pressure term is balanced so that the force is zero, but for the
particles within the interface region, a net acceleration rate and calorific will be produced, whose formulas are given by,

dvi  N
= 2ā m j ∇iWiHj , , (55)
dt
j=1

dUi   N 
= 2ā m j v j − vi ∇iWiHj , (56)
dt
j=1

where WiHj = W (ri − r j , H ) is a kernel function that is the same as the kernel function of other terms except for H≥2 h [118–
120]. Similarly, the method can also be used to characterize the behavior of the liquid-solid interface [59], in which the
cohesive pressure term, kaρ 2 , is introduced between the liquid and solid particles.

8.2. Inter-particle force model

As mentioned above, the surface tension is caused by molecular forces. As a particle based method, the inter-particle
force fi j representing the interaction between the particles i and j can be introduced to describe the surface tension in SPH,
where the sum of the inter-particle forces is zero when the particles are located in the region far away from the interface
and but is unbalanced at the interface region. The net force at the interface region turns out to be the surface tension,
which takes action toward the inner side. As a result, the inter-particle force should be attractive. The addition of these
inter-particle forces into the SPH momentum equation and the sum of the forces acting on these inter-particles can then
describe the surface tension force. The formula of the net acceleration produced by the surface tension force in this method
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9639

Table 3
Summary of inter-particle force model.

Inter-particle force model Expression of inter-particle force References

 1.5π  r −r
Cosine force f i j = si j cos |ri − r j | j i |r j − ri | < kh [58]
kh
 1.5π  |rr j −
− ri |
r
Modified cosine force f i j = si j cos |ri − r j | j i
W |r j − ri | < kh [121]
 kh |r j − ri| i j
57 r 2 45 r 3 | r ij|
Cubic polynomial force f i j = si j 1 − ( ij ) + ( ij ) ≤2 [122]
48 h 96 h h
r
f i j = −smi m j C (ri j ) |ri j |
⎧ ij

Cohesion force
⎨(kh − r )3 r3 kh > r > 0.5k
[123]
C (r ) = 32
2(kh − r )3 r 3 − (kh )6
0.5kh > r > 0
π (kh )9 ⎩ 64
0 otherwise
Cohesive force f i j = s(2.0W (ri j , 0.8h) − 1.0W (ri j , h )) ri j |ri j | ≤ kh [124]
|ri j |
Attractive force f i j = −s m j Wi j [125]
mi
(The interpolation kernel W is B-cubic spline function with a support of length 4 h.)
 12  6
Lennard–Jones force fi j = − m j ϕi j ∇ Wi j , ϕi j = 4ε
r0
−
r0
. [126]
ρj ri j ri j

Fig. 5. Comparison of the inter-particle force.

can be given by,

 
1 
N
dvi
= fi j . (57)
dt mi
s j=1

In order to reduce the computational cost, the inter-particle force fi j is usually set to be zero when the distance between
particles i and j is greater than the radius of support domain. Presently, there are several forms describing this pair-wise
force. The typically-employed expressions of these inter-particle force models are summarized in Table 3 and the variations
of normalized interaction force with the distance between the pair-wise particles for different inter-particle force models
are given Fig. 5. As seen, Tartakovsky et al. [58] in 2005 constructed a cosine force which is repulsive at the short distance
and attractive at the long distance to characterize the surface tension force for the first time. Then, this interaction particle
force were modified to make the force function become smoother by Aly et al. [121]. In 2007, Zhang et al. [126] used the
Lennard–Jones force to simulate the surface tension force, since the Lennard–Jones force is repulsive at the short distance
and attractive at the long distance. Su et al. [122] proposed a cubic polynomial force between particles to simulate the
surface tension force. Kordilla et al. [124] constructed the inter-particle force function consisting of two superposed kernel
function. Akinci et al. [123] proposed a spline function to describe the inter-particles. The above six forms of the inter-
particle force function are repulsive at the short distance and attractive at the long distance. Although such description
similar to the interaction between molecules has been well declared, Becher and Teschner [125] proposed a form with
the only consideration of the attractive force using the smoothing kernel W as a weighting function, which has also been
9640 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Fig. 6. Schematic of the transition region in the inter-particle force model [128].

successfully used to simulate the surface tension force. Hence, some researchers [58,124] pointed out that the description
and accuracy of this force was not so important in the simulation of the multiphase flow.
In summary, except the attractive force model, the forces determined by these inter-particle force models are similar to
the force between the molecules, which is repulsive at the short distance and attractive at the long distance. The repulsive
force at the short distance can effectively prevent the particles from the cluster formation, especially for the Lennard–Jones
force model. Besides, the first five forms of the inter-particle force model listed in Table 3 are anti-symmetric, which ensures
the momentum conservation of the inter-particle interaction force. Moreover, the first six forms of inter-particle interaction
are soft to allow reasonably large time step to be used. These force models not only are suited for the multiphase flow but
also can be easily generalized to the interaction between the solid wall and the fluid. Although the attractive force model is
unable to prevent the particles from cluster formation, it is better suited for the problems with high curvatures.
In the above inter-particle force models, the force acts on each particle and the computational cost is usually high.
In order to reduce the computational expense, Zhou et al. [127] proposed the interface particle force, in which only the
particles near the interface are affected by the force from the surrounding particles to describe the surface tension and
remaining particles have no such kind of force. As schematically displayed in Fig. 6, the transition region of the interface is
about two times of the support radius of weighting function (3 h for the quintic spline function). For liquid–liquid surface
tension, the inter-particle force is repulsive force, which decreases with increasing the distance and only exists between
different kinds of fluid particles near the interface. With respect to the wetting behavior of the fluid on a solid wall, a
Lennard–Jones like force is introduced between the fluid particles and wall particles if the wall is wetted by the fluid, while
a repulsive force is introduced when the wall is not wetted by the fluid. Such manner can reduce the computational expense
for simulating the interactions of fluid–fluid or fluid-wall. However, it becomes unavailable for the case of involving the free
surface.

8.3. Continuum surface force model

Continuum surface force (CSF) model was proposed by Brackbill et al. [129] in 1992. In this model, it is assumed that
the interface is represented by a transition region with a finite thickness. The surface force is converted into a body force,
i.e. f = fs δs , where δ s is the normalized function and a peak value at the interface, fs is the force per unit surface area. The
surface tension force per unit surface area can be calculated according to the following formula [130,131],
fs = σ κ n
 + ∇s σ , (58)

where σ is the coefficient of the surface tension (N/m), к is the curvature of the interface (m), n  is the vector of interface,
∇ s σ is the surface tension gradient. The first term at the right of Eq. (58) is a net force caused by the local curvature of
the interface. The greater curvature, the greater force normal to the interface. It tends to reduce the interface area so as to
minimize the surface energy. The second term is the interface tension gradient, which is parallel to the interface. This part
makes the micelle at the interface to flow from the zone with low surface tension to the zone with high surface tension. Of
course, if a surface tension is given and fixed, this term becomes zero. Currently, there are two main ways for calculating the
surface tension using CFS model. One is the use of the color function to calculate the interface curvature and the interface
normal vector, the other is based on the interface reconstruction.
Morris [130] used CFS model by the color function to calculate the surface tension for the first time. In this method, a
color function was used to describe different phases, and the interface is defined as a finite transitional band, where the
color gradient does not vanish. Within this band, the surface tension is approximated as a continuous force. The curvature
and normal vector of the interface is calculated using the color function, the basic idea of this method is similar to the VOF
method. A color function, which is also called color index, is introduced as follows.

s particle i belongs to phase s
Ci = s = 1, . . . , n. (59)
0 else
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9641

Table 4
Comparison of interface tension models.

Model CSF IPF CPM

Accuracy High Low Low

Complexity Complex Simpler Simplest
Calibration No Yes Yes

The normal vector of the interface can then be obtained, particle i belongs to phase s
 = ∇ C/|∇ C |.
n (60)
And the curvature of the interface is,
κ = −∇ · n . (61)
Finally, the surface tension force is converted into a body force,
f = −σ (∇ · n
 )∇ Ci . (62)
Although using the color function to calculate the normal vector and curvature of interface is relatively simple, large
error may come out when directly using the SPH method to discretize Eqs. (60) and (61) [130]. To solve this problem,
some researchers [130,132,133] have carried out some works. For example, Hu and Adams [132] directly used a sharp color
function with a discontinuity at the interface. The surface curvature was achieved by introducing a surface stress tensor that
only depends on the color gradient. They also proposed a new reproducing divergence approximation without the demand
of a matrix inversion to obtain a stable and accurate scheme for the interface curvature [133]. A density-weighted color-
gradient formulation is used to reflect the reality of an asymmetrically distributed surface tension force.
The other method for reconstructing the interface curvature is to identify the interface particles by the capturing technol-
ogy [110]. The positions of interface particles are fitted to obtain an interface curve. This curve is then used to calculate the
normal vector and curvature of the interface. The detailed process of this method can be found in the references [134,135].
In summary, the above two methods have their virtues and faults, respectively. The interface curvature reconstruction
method can avoid the calculation of the first and second order derivatives using the SPH approximation. For the color
function method, the accuracy is reduced due to the lack of particles of support area especially for calculating large cur-
vature region [136]. Therefore, the interface curvature reconstruction method is more accurate in general. On the other
hand, the transitional band is narrow in the interface reconstruction based approach, which makes the results more close
to the Laplace formula. However, this approach is complex because the technology of the boundary particle detection is
complicated and influences the precision. In addition, the interface reconstruction based approach may become invalid if
the particles within the boundary are dispersed and the cost of calculation is high. In order to better simulate the surface
tension force, some methods were proposed to calculate the surface tension force rather than the above-mentioned two
main methods. For example, Hieber et al. [137] presented a Lagrangian particle level set method for capturing interfaces.
Zhang et al. [138] then used this method to calculate the surface tension force in SPH. Duan et al. [139] also developed a
contour continuum surface force model. Although these methods improve the accuracy of the curvature, the complexity is
highly increased and the curvature is difficult to directly calculate using the algorithm of SPH.

8.4. Summary of the surface tension models

The comparison of three surface tension models is summarized in Table. 4. Basically, the CPM and IPF are based on
the microscopic inter-phase attractive potential, while the CSF model is based on a macroscopic force. As compared to the
CSF model, the CPM and IPF models show some advantages. First, the algorithm of them is simple, particularly for CPM.
However, the calculation of the cohesive pressure term requires the smooth length not to be less than twice of other terms.
The discretization of the cohesive pressure term may require the correction for modified SPH [140,141]. In addition, the
normal vector and curvature of the interface need not be calculated. They trivially conserve the momentum force when the
applied force are pair-wise symmetric. Besides, the CPM and IPF can be easily generalized to the interaction between the
structure and the fluid. Although the implementation of CPM and IPF is straightforward, the particle in SPH represents the
mass of fluid rather than a molecule. Therefore, there are some difficulties for the use of the CPM and IPF models. First, the
resulting surface tension needs to be calibrated through the numerical experiment. Second, the surface tension is dependent
of the resolution and does not converge to a fixed value with increasing the resolution. On the contrary, the CSF model can
recover the prescribed surface tension and converge to an exact value with increasing the resolution [133]. Furthermore, the
forces between fluid particles in the models of CPM and IPF cannot guarantee the minimization of the surface area because
the forces can be trivially balanced with each other in a form without the necessary correspondence to the smallest surface
area [142]. Fig. 7 shows the oscillation of a 2D droplet with the initial square shape under surface tension force using CSF
and IPF method, respectively [143]. It can be found that the results of CSF are better than IPF for particles distribution at the
interface. In summary, although there have existed some models to describe the surface tension, more accurate description
of the surface tension is still essential for the application of SPH to the multiphase flow.
9642 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Fig. 7. Comparison of oscillation for 2D drop with initial square shape: (a) the results using CSF and (b) the results using IPF [143].

9. High density ratio

Typically, at normal pressure and temperature, the density ratio (the dense fluidic density to the light fluidic density)
of most two liquids is mainly in the range of 1–1.3 and but much larger for gases [144]. For example, the density ratio of
neon and hydrogen is about 10, and the density ratio was as high as 20.6 in the experiments of Grobelbaure et al. [145].
However, for most cases of the gas–liquid multiphase flow, the density ratio can be as high as hundreds, even thousands.
One typical case is the multiphase flow with air and water, whose density ratio is about 800. Air cannot be neglected be-
cause its thermodynamic pressure has a significant effect on the dynamics [144]. As mentioned above, the particles in the
SPH have a spatial distance (smoothing length), over which their properties are distributed by a smoothing kernel function
[10,146]. When densities and masses of neighboring particles vary within the smoothing length, the distributed quantities
of a particle show false values. Such problem can be encountered near the interface with high density difference. The false
value is acceptable only when the density gradient is much smaller than that of the smoothing kernel [40,66,67,146]. Val-
izadeh et al. [146] pointed out that the standard SPH method could well adapt to the multiphase flow with the density
ratio smaller than 10. The numerical experiments by Colagrossi et al. [147] suggested that the standard SPH method never
succeed for the multiphase flow when the density ratio is as high as 10. The reason is that the erroneous quantities result
in undesirable effects. Unphysical density and pressure variations may generate spurious and unnatural surface tension, and
even severe numerical instabilities [10]. However, the density ratio is usually larger than 10 (so-called high density ratio)
in most cases of the multiphase flow. Hence, how to apply the SPH method to simulate the cases with high density ratio
becomes challenging. Right now, if the interface is treated to be a sharp interface, the main way to handle this problem is
to use the particle number density to replace the particle density. In addition to the sharp interface, the diffuse interface is
also incorporated into the SPH to handle the interface with high density ratio.

9.1. Particles number density

The main idea of using particles number density is that the contribution of other particles within the particle support
region is the volume of the particles instead of density or mass. Based on this idea, Colagrossi and Landrini [145] pro-
posed a modified approximation form of spatial derivative to handle the high-density-ratio multiphase flow. In their work,
Eqs. (8) and (9) are rewritten as the following Eqs. (64) and (65). The density evolution equation, i.e., Eq. (66), is used to
solve the density. The pressure is obtained by using weakly compressible equation of state. The equation of state for dense
fluid is Tait equation. The equation of state for light fluid is the modified Tate equation, namely, the Tait equation coupled
cohesive pressure term to characterize the surface tension. In the meantime, XSPH [66,148] is used to prevent non-physical
penetration of different phases.
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9643

Fig. 8. Behaviors of two fluids with different densities modeled with standard and modified SPH equations: (a) initially square drop, (b) standard SPH
equations and (c) modified SPH equations [152].

 mj    
N
∇ · f ( ri ) = f r j − f (ri ) · ∇iWi j , (64)
j=1
ρj

  

N
mj f rj f ( ri )
∇ · f ( ri ) = + · ∇iWi j , (65)
j=1
ρj ρ 2j ρi2

d ρi  N
m
= ρi vi j · ∇iWi j j . (66)
dt
j=1
ρj

It should be noted that in this method, the mass conservation is not satisfied because of using the density evolution
equation instead of the density summation. Although some authors used the reconstructed density to reduce the error of
mass conservation [62,149], error will be accumulated and finally affect the flow behavior after a long calculation. Moreover,
the reconstruction of density is quite complex and requires huge computational time because three-dimensional matrix
needs to be solved. In addition, this method requires that the artificial sound of gas is larger than that of liquid and the ratio
of artificial sound is as high as about (ρ d γ l /ρ l γ d )0.5 , by which the physical meaning cannot then be realized. A relatively
small time step is also required according to the CFL conditions, increasing the computational cost.
Chen et al. [150] present an improved SPH model to simulate the multiphase flow with complex interface and high
density ratio. This model is based on the assumption of the pressure continuity across the interface. Near the interface, all
the neighboring particles of other phase in its support domain are regarded as the interpolation points with the same phase.
Only position, velocity, volume and pressure of these particles are taken into account when solving the acceleration and
density of the central particle. In addition, the inter-particle pressure was not applied in this model to reduce the influence
of the fluid with large density on the fluid with small density when the density ratio is high. The density re-initialization is
also applied. The density re-initialization algorithm is corrected based on the pressure continuity. To avoid negative particle
pressure that may cause numerical instabilities in computation, a cut-off value of the particle density is employed.
Unlike the modified spatial derivative approximation, Ott et al. [151] introduced a modified SPH approach based on
discretizing the particle number density other than the mass density. Similarly, Tartakovsky et al. [152] and Solenthaler
et al. [10] proposed the particle number density σ i to replace mj /ρ j , and proved that this approach could remove the artifi-
cial surface tension caused by high density ratio. As shown in Fig. 8, an initial square shape droplet composed of fluid 1 is
simulated, in which the droplet is surrounded by fluid 2 and the physical surface tension is not accounted. The density of
fluid 1 is 5 times smaller than fluid 2, and the system pressure is uniform. At t = 30, the square droplet became a circular
droplet (Fig. 8b) using standard SPH equation with the artificial surface tension, and the square droplet remained unchanged
because the artificial surface tension has been removed (Fig. 8c). Also Hu and Adams [132] performed the approximations
for particles averaged by particle smoothing functions, in which neighboring particles only contribute to the specific volume
but not to the density. This method defines the inverse of the particle volume σ to be the same as the particle number den-
sity, which is calculated using Eq. (67). Here, the density ρ is replaced by the inverse of the particle volume σ to discretize
the governing equations.


N
σi = Wi j , (67)
j=1


N
ρi = m i σ i = m i Wi j , (68)
j=1

  

N
f rj f ( ri )
∇ · f ( ri ) = σi + · ∇iWi j . (69)
j=1
σ j2 σi2
9644 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

This method not only adapts to density discontinuities when there are large particle-mass differences between adjacent
particles, but also satisfies the conservation of mass, because the method employs the density summation formula. However,
the method developed by Hu and Adams [132] may cause the problem of the particle agglomeration and aggregation, and
give rise to a false density value in the neighbor of the free surface due to using the summation of density. In 2006, an
incompressible SPH model was proposed [77], which used the projection formulation to solve the Poisson equation to get
the pressure. Inspired by this work, some researchers [153] used exact projection method to solve the Poisson equation,
which ensures a constant density in the process of simulation. However, although subsequent works can solve the problem
to some extent and do not require the introduction of artificial speed of sound, the computational cost is greatly increased
because of solving the Poisson equation.
Besides, Grenier et al. [154] proposed a new multiphase flow SPH model, which is actually an extension of the work by
Colagrossi and Landrini [147]. The formulation is based on the Shepard kernel [149] for the density evaluation by Eq. (70),
and a new discretized formula, Eq. (71), used for the smoothed physical quality gradient. The authors defined their method
as a Hamiltonian interface SPH formulation.
⎧ 
⎨ ρ= 1
iχ
m jWi j
j∈χ
χ  . (70)
i ∈ fluidχ
mj
⎩ i = ρ j Wi j
j∈χ
⎧
N  
⎪
⎪ 
⎨∇ f = ∇iWi j
fi fj mj
i + j ρj
j=1
. (71)
⎪ 
N
⎪
⎩i =
mj
ρ Wi j
j=1

This SPH scheme allows an accurate treatment of the discontinuity of quantities at the interface (such as the density) so
as to model the multiphase flow. For the interface flow where the surface tension effect is negligible, a small repulsive force
is introduced in the pressure gradient to prevent a spurious fragmentation of the interface, which is similar to the method
by Monaghan [155]. The fluids are considered as weakly compressible fluids, and Tait equation is used to describe the state
of fluid. Similarly, the artificial sound speed of light fluid is larger that of dense fluid for simulating the multiphase flow, as
discussed above.
In addition, Monaghan et al. [144] proposed an alternative SPH algorithm based on the standard SPH algorithm
with the inclusion of a repulsion term in the force between different fluids, which is similar to the work by Grenier
et al. [154]. As suggested by Monaghan [155], for the tension instability, they used Eq. (65) to calculate the density and
the acceleration equation for particle i was presented as follows.
   
dv 
N
Pi + Pi
=− mj + Ri j + i j · ∇iWi j + f , (72)
dt
i j=1
ρi ρ j
  
 ρ0i − ρ0i  Pi − Pi 
Ri j = 0.08  , (73)
ρi ρ j  ρi ρ j 
16υi υ j v · r
i j = −   i j i j . (74)
υi ρi + υ j ρ j ri j  0.5hi + 0.5h j
Note that this method does not use the particle number density, or velocity smoothing, or an artificial surface tension
and but uses Wendland kernel [156] as smoothing kernel function. It is suitable to simulate the multiphase flow problems,
in which the surface tension is negligible. The ratio of artificial sound speed between the light and dense fluids is 3 times
smaller than that used in Grenier et al. [154]. The computational cost can be reduced to some extent because the time-step
is dependent of the max artificial sound speed for the multiphase flow. However, the artificial sound speed of light fluid is
large than that of dense fluid, which has been discussed above.
Shadloo et al. [157] used particle number density combining CSPH to simulate the incompressible multiphase flow. The
gradient of the CSPH discretization can be realized by Eq. (75), and the CSPH discretization scheme for the Laplacian of an
arbitrary function can be written as Eqs. (76) and (77).

∂ fiα  βγ −1  N
1 α ∂ Wi j
= a f ji , (75)
∂ riβ i
j=1
σ i ∂ riγ


 αγ −1  α
∂ ∂ fiα N
2 ζi ζ j α ri j ∂ Wi j
ζ = 8 a f , (76)
∂ riβ
i
∂ riβ
i
j=1
σ j ζi + ζ j ri2j ∂ riγ
i j

 γ
∂ ∂ fiα 8  N
2 ζi ζ j α ri j ∂ Wi j
ζ =   f , (77)
∂ rβ
i
∂ rβ 1 + ai
χχ
j=1
σ j ζi + ζ j i j ri2j ∂ riγ
i i
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9645

N
m j α ∂ Wi j
αiαβ = r , (78)
j=1
ρ j i j ∂ rβ
i

where ζ denotes the fluid properties, such as μ, and ρ −1 depending on the discretized equations. The superscripts of α ,
β , γ and χ are the dimension indexes ranged from 1 to the number of dimensions. fiαj = fiα − f jα , and fiα represents the
α index of vector-function f of particle i. The term in the left side of Eq. (78) is a corrective second-rank tensor. Eq. (76) is
used for the discretization of the Laplacian of velocity field in the linear momentum equation while Eq. (77) is utilized for
the Laplacian of the pressure in the pressure Poisson equation. It should be noted that in the work of Shaldoo et al. [157],
the particle number density σ was used and a weighted harmonic mean interpolation was used to smooth the transport
2ζ ζ
parameter, namely, ζi = ζ +i ζ j . Such manner can improve the accuracy and robustness. Although the model can be used to
i j
simulate the density ratio up to 10 0 0, it is required to calculate the 2×2 corrected and inverse matrix for two-dimensional
problem and 3×3 matrix for three-dimensional problem so that the computational cost is not low.

9.2. Other ways

In addition to the above methods, Ritchie and Thomas [158] suggested a summation of particle averaged pressure instead
of a density to handle this problem, but their method does not satisfy the mass conservation and requires the gradient
of pressure to be small. Based on the fact that the mass difference between two type fluidic particles is the cause of the
velocity increase and the changing rate, Valizadeh et al. [146] concentrated on changing the mass of particles which interacts
with other type fluidic particle artificially. To this end, two correction factors were defined to change the mass, namely,
md new =cmd md and ml new =cml ml , where the subscript symbol d and l represent the dense and light fluids, respectively.
This method is simple and easy, and can treat the density ratio as high as 10 0 0. However, this method has not given the
theoretical basis about this amendment. Besides, Zhang and Deng [138] treated material discontinuity across the interface
with the ghost fluid method, and the interface status is calculated by applying the jump condition and then extended to the
corresponding ghost fluid particles.

9.3. Diffuse interface

The ways described above all are on the basis of the sharp interface. Xu et al. [159] proposed an algorithm where the
diffuse interface (DI) is incorporated into SPH to solve the problem. In this method, the interface is treated as a narrow
region characterized by a smooth but rapid variation of physical properties between the bulk values of two matters. More
detail can be found from the reference [160].
In fact, DI is better suited for simulating complex interface structure and more immune to stress singularities [161].
Despite promising, the incorporation of DI into SPH was firstly done in 2009 because its Lagrangian nature makes the
incorporation difficult. Xu et al. [159] used mass density as ordered parameter to combine the DI concept with SPH. Das and
Das [161–163] adopted the phase color code as ordered parameter and introduced separated transport equations. This work
greatly promotes the development and application of a hybrid model incorporating the DI concept into the particle-based
methodologies
To show more detailed information about the DI, both sharp interface and DI are compared in Fig. 9, in which two
different phases are represented by particles with white color (phase I) and black color (phase II). As a physical thickness
is assigned to the interface following the concept of DI, the momentum equation will change accordingly. With the com-
bination of the source term to account for the interfacial free energy, the momentum equation can be written as follow
[161],
dv 1 1 C
= − ∇ p + ∇ (μ∇ · v ) + f − ∇ φ, (79)
dt ρ ρ C apC n
where C is the color code of phase, having +1 value for the liquid and −1 for the gas, Cap is the particular capillary number,
Cn is the Cahn number, the ratio of mean interfacial thickness and characteristic length, φ is the chemical potential of fluid
which is calculated based on the color code. φ is calculated for each phase as follows,
 
φ = ∇ (C + 1)2 (C − 1 )2 − C n2 ∇ 2C. (80)
Das and Das [161–163] derived the transport equation of C by approximating the interfacial diffusion fluxes proportional
to the chemical potential gradients:
dC
= k∇ 2 φ . (81)
dt
With this method, Das and Das adopted the hybrid DI-SPH method to simulate the formation, growth and departure of
a bubble from a submerged orifice [161] and an inclined orifice [164], and the drop movement over an inclined surface
[163,165]. As an example shown in Fig. 10 [164], the results of the hybrid DI-SPH method are in good agreement with the
experiment data. Their works have demonstrated that the hybrid DI-SPH method is an effective tool in the simulation of the
multiphase flow problems with evolving interfaces.
9646 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Fig. 9. Scheme of the diffuse interface concept in SPH [161].

10. Applications of SPH for multiphase flow

As a matter of fact, although the SPH method shows great potential on the simulation of the multiphase flow, the appli-
cation of the SPH just started from the beginning of the 1990s attempted by Monaghan [66] for the free surface flow. Mon-
aghan and his coworkers also used the SPH method to study the behavior of the gravity flow and solitary waves [67,166],
the wave arrival at a beach [67] and a Scott Russell’s wave generator [167] and two-phase flow [168]. At present, SPH has
been successfully applied in many fields of fluidic dynamics. In this section, we will introduce some typical applications of
SPH for the multiphase flow, including the droplet behaviors, the Rayleigh–Taylor instability and the bubble dynamics.

10.1. Droplet behaviors

The behavior of droplets simulated by SPH for the first time is to study the coalescence of colliding van der Waals liquid
droplets [119], in which cohesive pressure model was used to simulate the surface tension. Similar works have also been
carried out by many researchers, including the collision of droplets [134,135,140], droplet deformation [169–172], droplet
movement [163,173], droplet impacting liquid surface [111,174,175] and solid surface [122,176].
For example, the droplet collision is a complex process and affected by many parameters like Reynolds number, Weber
number, droplet size and so on. And there are many results about droplet collisions. Only for the binary collision of same
liquid droplets, the results show permanent coalescence, reflexive separation and stretching separation. In this case, the
process of droplet collisions involving surface deformation and movement has been simulated using SPH. Melean et al.
[119] also presented a two-dimensional SPH simulation of head-on and off-center binary collision dynamics of equal-sized,
van der Waals liquid droplets. Their results reveal that the outcome of coalescence is affected by the Reynolds and Weber
numbers for low energy collision. Fig. 11 presents the evolution of the off-center collision model with Weber number of 2
[119]. Jiang et al. [140] simulated the dynamic collision process between two miscible or immiscible micro-droplets using
CSPH. The outcomes of two immiscible micro-droplets off-center collision with Reynolds number equal to 75 are shown
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9647

Fig. 10. Bubble evolution and its rising in liquid phase with and without DI method: (a) 0.018 s, (b) 0.041 s, (c) 0.0645 s [161].

in Fig. 12. In these two works, the droplets have equal size with two dimensions in vacuum. Acevedo and his co-works
[177–184] simulated the collision of liquid droplets in a vacuum environment with three dimensions. Two equal-sized or
unequal-sized droplets, three and many droplets interactions of multiple colliding are explored. Qiang et al. [185] simulated
two equal-sized droplets colliding including head-on and off-head colliding in air using two dimensional SPH.

10.2. Rayleigh–Taylor instability

The Rayleigh–Taylor instability (RTI) is an instability that is developing and evolving at the interface between two hori-
zontal parallel fluids with different densities and viscosities. The instability occurs when a multi-fluid system with different
densities is affected by the gravity force. An unstable disturbance tends to grow in the direction of gravitational field, thereby
liberating and lessening the potential energy of the system. Since the RTI is a significant phenomenon in many fields of en-
gineering and science, it has been widely studied. Because of the intrinsic advantages offered by SPH, some researchers have
also tried to use the SPH method to study the RTI.
9648 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Fig. 11. Time dependent evolution of the off-center collision model with Weber number of 2 [119].

Fig. 12. Time dependent evolution of the off-center collision model with Reynolds number of 75 [140].

The first work on the simulation of RTI by SPH was reported by Cummins and Rudman [71]. They used WCSPH and ISPH
with an approximate projection to enforce incompressibility. Fig. 13 compares the results of ISPH and WCSPH with an exact
projection method performed on a staggered finite-difference MAC grid. It can be concluded that the results of ISPH is more
accurate than the approach of WCSPH. A more accurate projection needs to be enforced because the use of an approximate
projection to enforce incompressibility leads to the error accumulation in the density field, which finally affects the results.
Tartakovsky et al. [152] also modeled the RTI problem with a multiphase and multi-component mixture by the WCSPH
method through solving momentum and mass balance equations concurrently, where the particle number density is used
to discretize the governing equation. Hu and Adams [77] combined the projection methods proposed and implemented by
Cummins et al. [71] and Shao et al. [20] to solve the RTI, in which the permitted maximum density error is 0.5%. And then
Hu and Adams [153] proposed a constant-density approach to simulate the RTI. Grenier et al. [154] presented a new WCSPH
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9649

Fig. 13. Interface descriptions of the RTI. (a) Initialization (b) Exact Projection, (c) ISPH, (d) WCSPH [71].

Fig. 14. Comparison of simultated results, (a) ISPH with 0.5% density error [77], (b) Constant density model [153], (c) Hamiltonian interface SPH formulation
[154].

formulation that has been discussed previously to model the RTI for validating their numerical scheme. The results are
compared in Fig. 14, which indicates the numerical results are acceptable. Shadloo et al. [157] employed CSPH to simulate
the RTI through solving Poisson equation and study the long time evolution of the RTI. To date, vast works on the RTI
simulated by SPH are for the purpose of validating the numerical scheme. However, the density ratio is usually lower than
2. Consequently, further research on the RTI by SPH with reasonable density ratio is needed.
9650 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

Fig. 15. Initialization of a single bubble rising in liquid phase.

10.3. Bubble dynamics

The bubble dynamics is of importance in engineering, including heat exchanger, boiling and so on. To investigate the
bubble dynamics, extensive efforts have been devoted to study the bubble formation, growth and departure at the orifice.
As the bubble dynamics shows great interfacial effect, SPH has also been applied to the simulation of bubble dynamics.
One classic case is that a single gas bubble upward motion in the liquid. Some researchers [147,154,186,187] have simulated
this case using SPH, The initialization of this case is shown in Fig. 15 and their results are compared in Fig. 16. Overall,
their results agreed with the solution of Level Set method [188]. However, there are some differences, such as the threshold
of detachment. The result by Grenier et al. [154] showed that the detachment was too hasty and bubble is completely
dichotomized in comparison with other methods.
In addition, Valizadeh et al. [146] used the method with modified particles mass within the interface to simulate the
upward movement of a single bubble in the dense fluid, which can ensure the interface stabilization. The simulation results
show that the interface particle distribution is improved as compared to the results of Colagrossi et al. [147]. Szewc et al.
[189] employed the modified SPH method with particle number density to simulate the rising motion of a single bubble
in viscous fluid, where the repulsive term was introduced to the pressure gradient term. Different regimes and the pre-
dicted topological changes as well as the terminal velocity and drag coefficients of bubbles were simulated to validate their
modified SPH method. Grenier et al. [190] used a Hamiltonian interface SPH formulation to simulate viscous incompressible
bubbly flows with increased complexity, including the isolated bubble evolution, the two bubbles merging and the sepa-
ration process in a bubbly flow. In their work, the influence of the Bond number was deeply investigated and analyzed.
Furthermore, the gas bubble formation at a submerged orifice was simulated by Das and Das [191,192], where CSPH and
changing smoothing length were employed. Recently, Das and Das [161,163] incorporated diffuse interface into SPH to sim-
ulate the gas bubble formation at a submerged orifice, the evolution and departure from the orifice. Besides, the bubble
formation at an inclined orifice was also simulated by Das and Das [163]. In this SPH simulation, the entire range of the
orifice inclination under the air flow of 30 cm3 /s was explored. With these works, the proposed numerical algorithms have
been validated, indicating that the SPH method can be used to study the bubble dynamics.

11. Summary

The interfacial phenomena in the multiphase flow bring a grand challenge to the simulation work. SPH, as a meshfree
particle method, has shown great potential in the simulation of the multiphase flow. However, due to the complexity of the
multiphase flow, SPH still face some critical issues for simulating the multiphase flow. This article has given a comprehen-
sive review of past studies on the SPH methods to simulate the multiphase flow with significant interfacial effect. Emphasis
is placed on the development of the SPH algorithm for the purpose of the multiphase flow, including the discretization
of governing equations, the pressure treatment about incompressible flow, the boundary condition treatment, the kernel
and improvement of SPH, the modeling of the free surface flow, the simulation of the surface tension and the treatment
of high density ratio. Some successful applications of the SPH to the multiphase flow have also been introduced. The past
efforts have laid a solid foundation to apply the SPH method for the multiphase flow. However, future research on the de-
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9651

Fig. 16. Comparison of simultated results on a single bubble rising in liquid phase, (a) Results by Colagrossi et al. [147], (b) Results by Grenier et al. [154],
(c) Results by Vahabi et al. [186] and (d) Results by Grenier et al. [187].

velopment of SPH toward the fluidic dynamics with the interfacial phenomena is still needed with addressing the following
critical issues. (i) Presently, the accuracy of SPH is still unacceptable for widespread application as compared to grid based
methods. Although existing investigations have improved the accuracy by modifying SPH, the modified algorithms is rather
complex and time costly in general. The new algorithms may be needed to improve the accuracy and in the meantime with
low time expense. (ii) The computational cost of existing SPH methods is critically high, especially for the modified SPH.
Efficient algorithm is necessary to aid this puzzle. (iii) The boundary treatment is currently puzzle. Although there many
different ways to treat boundaries, they are usually effective in a single fluid system. It is essential to extend the work
to a system with two or more fluids. (vi) Surface tension treatment is critically important in the simulation of the multi-
phase flow using SPH. More accurate description of the surface tension is essential for the application of SPH to the multi-
phase flow. (v) High density ratio is usually encountered in most multiphase flows. How to treat high density ratio is also
necessary for the development of SPH to spread its application. In particular, it is urgent to guarantee the continuity of force
when dealing with high-density-ratio case.

Acknowledgments

The authors gratefully acknowledge the financial supports of National Natural Science Funds (No. 51222603, No. 51276208
and No. 51325602) and Program for New Century Excellent Talents in University (NCET-12-0591).

Reference

[1] L.J. Guo, B.F. Bai, L. Zhao, X. Wang, H.Y. Gu, Online recognition of the multiphase flow regime and study of slug flow in pipeline, J. Phys.: Conf. Ser.
147 (2009) 012047.
[2] O. Hungr, S. Leroueil, L. Picarelli, The Varnes classification of landslide types, an update, Landslides 11 (2014) 167–194.
[3] P. Lollino, D. Giordan, P. Allasia, The Montaguto earthflow: a back-analysis of the process of landslide propagation, Eng. Geol. 170 (2014) 66–79.
[4] E. Delnoij, J.A.M. Kuipers, W.P.M.V. Swaaij, A three-dimensional CFD model for gas–liquid bubble columns, Chem. Eng. Sci. 54 (13) (1999) 2217–2226.
[5] D.D. Mcclure, J.M. Kavanagh, D.F. Fletcher, G.W. Barton, Development of a CFD model of bubble column bioreactors: part two–comparison of experi-
mental data and CFD predictions, Chem. Eng. Technol. 37 (1) (2014) 131–140.
9652 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

[6] Q.G. Xiong, E. Madadi-Kandjani, G. Lorenzini, A LBM–DEM solver for fast discrete particle simulation of particle–fluid flows, Continuum Mech. Ther-
modyn. 26 (6) (2014) 907–917.
[7] M. Robinson, S. Luding, M. Ramaioli, Fluid-particle flow and validation using two-way-coupled mesoscale SPH-DEM, Int. J. Multiph. Flow. 59 (2014)
121–134.
[8] D.A. DiCarlo, Stability of gravity-driven multiphase flow in porous media: 40 years of advancements, Water Resour. Res. 49 (8) (2013) 4531–4544.
[9] S. Subramaniam, Lagrangian–Eulerian methods for multiphase flows, Prog. Energy Combust. Sci. 39 (2) (2013) 215–245.
[10] B. Solenthaler, R. Pajarola, Density contrast SPH interfaces, in: Proceedings of the 2008 ACM SIGGRAPH/Eurographics symposium on computer ani-
mation, Dublin, Ireland, 2008, pp. 211–218.
[11] K. Matyash, R. Schneider, F. Taccogna, A. Hatayama, S. Longo, M. Capitelli, D. Tskhakaya, F.X. Bronold, Particle in cell simulation of low temperature
laboratory plasmas, Contrib. Plasma Phys. 47 (8-9) (2007) 595–634.
[12] A. Pukhov, J. Meyer-Ter-Vehn, Relativistic magnetic, Self-channeling of light in near-critical plasma: three-dimensional particle-in-cell simulation,
Phys. Rev. Lett. 76 (21) (1996) 3976–3978.
[13] F.M. Tome, S. McKee, GENSMAC: A computational marker and cell method for free surface flows in general domains, J. Comput. Phys 110 (1) (1994)
171–186.
[14] C.W. Hirt, B.D. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries, J. Comput. Phys. 39 (1981) 201–225.
[15] D.P Peng, B. Merriman, Stanley Osher, Hongkai Zhao, Myungjoo Kang, A PDE-based fast local level set method, J. Comput. Phys. 155 (1999) 410–438.
[16] F. Losasso, T. Shinar, A. Selle, R. Fedkiw, Multiple interacting liquids, ACM Trans. Graph. 25 (3) (2006) 812–819.
[17] F. Losasso, J. Talton, N. Kwatra, R. Fedkiw, Two-way coupled SPH and particle level set fluid simulation, IEEE Trans. Vis. Comput. Graph. 14 (4) (2008)
797–804.
[18] P.G. Jiao, Y.Q. Zhou, Z.R. Li, L. Chen, Simulation of two-phase flow using smoothed particle hydrodynamics, in: IEEE International Symposium on
Knowledge Acquisition and Modeling Workshop (KAM 2008 Workshop), Wuhan, China, 2008, pp. 296–300.
[19] S. Li, W.K. Liu, Meshfree and particle methods and their applications, Appl. Mech. Rev. 55 (2002) 1–34.
[20] T. Belytschko, Y. Krongauz, D. Organ, M. Fleming, P. Krysl, Meshless methods: an overview and recent developments, Comput. Methods Appl. Mech.
Eng. 139 (1) (1996) 3–47.
[21] G.R. Liu, M.B. Liu, Smoothed Particle Hydrodynamics: A Meshfree Particle Method, World Scientific, 2003.
[22] D. Violeau, R. Issa, Numerical modelling of complex turbulent free-surface flows with the SPH method: an overview, Int. J. Numer. Methods Fl. 53 (2)
(2007) 277–304.
[23] M.B. Liu, G.R. Liu, Particle Methods for Multi-Scale and Multi-Physics, World Scientific, 2015.
[24] L.B. Lucy, A numerical approach to the testing of fusion processes, Astron. J 82 (1977) 1013–1024.
[25] R.A. Gingold, J.J. Monaghan, Smoothed particle hydrodynamics-theory and application to non-spherical stars, Mon. Not. Roy. Astron. Soc. 181 (1977)
375–389.
[26] O. Kum, W.G. Hoover, H.A. Posch, Viscous conducting flows with smooth-particle applied mechanics, Phys. Rev. E. 109 (1995) 67–75.
[27] W.G. Hoover, C.G. Hoover, E.C. Merritt, Smooth-particle applied mechanics: conservation of angular momentum with tensile stability and velocity
averaging, Phys. Rev. E. 69 (1) (2004) 016702.
[28] Q.T. Ho, J. Carmeliet, A.K. Datta, T. Defraeye, M.A. Delele, E. Herremans, L. Opara, H. Ramon, E Tijskens, Ruud van der Sman, P.V. Liedekerke, P. Ver-
boven, B.M. Nicolaï, Multiscale modeling in food engineering, J. Food Eng. 114 (3) (2013) 279–291.
[29] B. Nayroles, G. Touzot, P. Villon, Generalizing the finite element method: diffuse approximation and diffuse elements, Comput. Mech. 10 (5) (1992)
307–318.
[30] G. Yagawa, T. Yamada, Free mesh method: a new meshless finite element method, Comput. Mech. 18 (5) (1996) 383–386.
[31] W.K. Liu, S. Jun, Y.F. Zhang, Reproducing kernel particle methods, Int. J. Numer. Methods Fluids 20 (8-9) (1995) 1081–1106.
[32] H. Wang, G.Y. Li, X. Han, Z.H. Zhong, Development of parallel 3D RKPM meshless bulk forming simulation system, Adv. Eng. Softw. 38 (2) (2007)
87–101.
[33] J. Dolbow, T. Belytschko, An introduction to programming the meshless element free Galerkin method, Arch. Comput. Method Eng. 5 (3) (1998)
207–241.
[34] E. Oñate, S. Idelsohn, A mesh-free finite point method for advective-diffusive transport and fluid flow problems, Comput. Mech. 21 (4-5) (1998)
283–292.
[35] J.J. Monaghan, Smoothed particle hydrodynamics and its diverse applications, Annu. Rev. Fluid Mech. 44 (2012) 323–346.
[36] J.J. Monaghan, Smoothed particle hydrodynamics, Annu. Rev. Astron. Astrophys. 30 (1992) 543–574.
[37] J.J. Monaghan, Smoothed particle hydrodynamics, Rep. Prog. Phys. 68 (8) (2005) 1703–1759.
[38] P.W. Randles, L.D. Libersky, Smoothed particle hydrodynamics: some recent improvements and applications, Comput. Meth. Appl. Mech. Eng. 139 (1)
(1996) 375–408.
[39] M.B. Liu, G.R. Liu, Smoothed particle hydrodynamics (SPH): an overview and recent developments, Arch. Comput. Method Eng. 17 (1) (2010) 25–76.
[40] J.J. Monaghan, A. Kocharyan, SPH simulation of multi-phase flow, Comput. Phys. Commun. 87 (1) (1995) 225–235.
[41] S. Adami, X.Y. Hu, N.A. Adams, A new surface-tension formulation for multi-phase SPH using a reproducing divergence approximation, J. Comput.
Phys. 229 (13) (2010) 5011–5021.
[42] M. Lahooti, A. Pishevar, M.S. Saidi, A novel 2D algorithm for fluid solid interaction based on the smoothed particle hydrodynamics (SPH) method, Sci.
Iran. 18 (3) (2011) 358–367.
[43] Y. Amini, H. Emdad, M. Farid, A new model to solve fluid–hypo-elastic solid interaction using the smoothed particle hydrodynamics (SPH) method,
Eur. J. Mech. B-Fluids. 30 (2) (2011) 184–194.
[44] M.B. Liu, J.R. Shao, H.Q. Li, A SPH model for free surface flows with moving rigid objects, Int. J. Numer. Methods Fluids 74 (9) (2014) 684–697.
[45] Q.G. Xiong, L.J. Deng, W. Wang, W. Ge, SPH method for two-fluid modeling of particle–fluid fluidization, Chem. Eng. Sci. 66 (9) (2011) 1859–1865.
[46] L.J. Deng, Y.N. Liu, W. Wang, W. Ge, J.H. Li, A two-fluid smoothed particle hydrodynamics (TF-SPH) method for gas–solid fluidization, Chem. Eng. Sci.
99 (2013) 89–101.
[47] M.G. Omang, J.K. Trulsen, Multi-phase shock simulations with smoothed particle hydrodynamics (SPH), Shock Waves 24 (5) (2014) 1–16.
[48] M. Müller, B. Solenthaler, R. Keiser, M. Gross, Particle-based fluid-fluid interaction, in: Proceedings of the ACM SIGGRAPH/Eurographics symposium
on Computer animation, Angeles, America, 2005, pp. 237–244.
[49] J.J. Monaghan, Why particle methods work, SIAM J. Sci. and Stat. Comput. 3 (4) (1982) 422–433.
[50] H.B. Jin, X. Ding, On criterions for smoothed particle hydrodynamics kernels in stable field, J. Comput. Phys. 202 (2) (2005) 699–709.
[51] O. Flebbe, S. Muenzel, H. Herold, H. Riffert, H. Ruder, Smoothed particle hydrodynamics: physical viscosity and the simulation of accretion disks,
Astrophys. J. 431 (1994) 754–760.
[52] S.J. Watkins, A.S. Bhattal, N. Francis, J.A. Turner, A.P. Whitworth, A new prescription for viscosity in smoothed particle hydrodynamics, Astron. Astro-
phys. Suppl. Ser. 119 (1996) 177–187.
[53] H. Takeda, S.M. Miyama, M. Sekiya, Numerical simulation of viscous flow by smoothed particle hydrodynamics, Prog. Theor. Phys. 92 (5) (1994)
939–960.
[54] L. Brookshaw, A method of calculating radiative heat diffusion in particle simulations, Proc. Astron. Soc. Aust. 6 (1985) 207–210.
[55] J.P. Morris, P.J. Fox, Y. Zhu, Modeling low Reynolds number incompressible flows using SPH, J. Comput. Phys. 136 (1) (1997) 214–226.
[56] J.J. Monaghan, Heat conduction with discontinuous conductivity, Appl. Math. Rep. 95 (18) (1995) 7.1.
[57] P.W. Cleary, J.J. Monaghan, Conduction modelling using smoothed particle hydrodynamics, J. Comput. Phys. 148 (1) (1999) 227–264.
[58] A. Tartakovsky, P. Meakin, Modeling of surface tension and contact angles with smoothed particle hydrodynamics, Phys. Rev. E. 72 (2) (2005) 026301.
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9653

[59] A.M. Tartakovsky, K.F. Ferris, P. Meakin, Lagrangian particle model for multiphase flows, Comput. Phys. Commun. 180 (10) (2009) 1874–1881.
[60] J.K. Chen, J.E. Beraun, T.C. Carney, A corrective smoothed particle method for boundary value problems in heat conduction, Int. J. Numer. Methods
Eng. 46 (2) (1999) 231–252.
[61] J.K. Chen, J.E. Beraun, A generalized smoothed particle hydrodynamics method for nonlinear dynamic problems, Comput. Meth. Appl. Mech. Eng. 190
(1) (20 0 0) 225–239.
[62] G.A. Dilts, Moving-least-squares-particle hydrodynamics—I. Consistency and stability, Int. J. Numer. Methods Eng. 44 (8) (1999) 1115–1155.
[63] J. Bonet, T.S.L. Lok, Variational and momentum preservation aspects of smooth particle hydrodynamic formulations, Comput. Methods Appl. Mech.
Eng. 180 (1) (1999) 97–115.
[64] P.S.R Fredini, A.C. Limache, Evaluation of weakly compressible SPH variants using derived analytical solutions of Taylor–Couette flows, Comput. Math.
Appl. 66 (3) (2013) 304–317.
[65] P.W. Cleary, M. Prakash, J. Ha, N. Stokes, C. Scott, Smooth particle hydrodynamics: status and future potential, Prog. Comput. Fluid Dyn. 7 (2) (2007)
70–90.
[66] J.J. Monaghan, Simulating free surface flows with SPH, J. Comput. Phys. 110 (2) (1994) 399–406.
[67] J.J. Monaghan, A. Kos, Solitary waves on a Cretan beach, J. Waterw. Port Coast. Ocean Eng. 125 (3) (1999) 145–155.
[68] M.S. Shadloo, A. Zainali, M. Yildiz, A. Suleman, A robust weakly compressible SPH method and its comparison with an incompressible SPH, Int. J.
Numer. Methods Eng. 89 (8) (2012) 939–956.
[69] A.J. Chorin, Numerical solutions of the Navier–Stokes equations, Math. Comput. 22 (104) (1968) 745–762.
[70] A.J. Chorin, On the convergence of discrete approximations to the Navier–Stokes equations, Math. Comput. 23 (106) (1969) 341–353.
[71] S.J. Cummins, M. Rudman, An SPH projection method, J. Comput. Phys. 152 (2) (1999) 584–607.
[72] B. Solenthaler, R. Pajarola, Predictive-corrective incompressible SPH, ACM Trans. Graph (TOG) 28 (3) (2009) 40.
[73] A. Khayyer, H. Gotoh, S.D. Shao, Corrected incompressible SPH method for accurate water-surface tracking in breaking waves, Coast. Eng. 55 (3)
(2008) 236–250.
[74] X. Liu, H. Xu, S. Shao, P. Lin, An improved incompressible SPH model for simulation of wave–structure interaction, Comput. Fluids. 71 (2013) 113–123.
[75] M.M. Tong, D.J. Browne, An incompressible multi-phase smoothed particle hydrodynamics (SPH) method for modelling thermocapillary flow, Int. J.
Heat Mass Transf. 73 (2014) 284–292.
[76] S.D Shao, E.Y.M. Lo, Incompressible SPH method for simulating Newtonian and non-Newtonian flows with a free surface, Adv. Water Resour. 26 (7)
(20 03) 787–80 0.
[77] X.Y. Hu, N.A. Adams, An incompressible multi-phase SPH method, J. Comput. Phys. 227 (1) (2007) 264–278.
[78] R. Xu, P. Stansby, D. Laurence, Accuracy and stability in incompressible SPH (ISPH) based on the projection method and a new approach, J. Comput.
Phys. 228 (18) (2009) 6703–6725.
[79] M. Gomez-Gesteira, B.D. Rogers, R.A. Dalrymple, A.J.C. Crespo, State-of-the-art of classical SPH for free-surface flows, J. Hydraul. Res. 48 (S1) (2010)
6–27.
[80] E.S. Lee, C. Moulinec, R. Xu, D. Violeau, D. Laurence, P. Stansby, Comparisons of weakly compressible and truly incompressible algorithms for the SPH
meshfree particle method, J. Comput. Phys. 227 (18) (2008) 8417–8436.
[81] E.S. Lee, D. Violeau, R. Issa, S. Ploix, Application of weakly compressible and truly incompressible SPH to 3-D water collapse in waterworks, J. Hydraul.
Res. 48 (S1) (2010) 50–60.
[82] M. Asai, A.M. Aly, Y. Sonoda, Y. Sakai, A stabilized incompressible SPH method by relaxing the density invariance condition, J. Appl. Math. 11 (2012)
2607–2645.
[83] R. Issa, E.S. Lee, D. Violeau, D.R. Laurence, Incompressible separated flows simulations with the smoothed particle hydrodynamics gridless method,
Int. J. Numer. Methods Fluids. 47 (10–11) (2005) 1101–1106.
[84] A. Valizadeh, J.J. Monaghan, A study of solid wall models for weakly compressible SPH, J. Comput. Phys. 300 (2015) 5–19.
[85] G. Oger, M. Doring, B. Alessandrini, P. Ferrant, Two-dimensional SPH simulations of wedge water entries, J. Comput. Phys. 213 (2) (2006) 803–822.
[86] S. Kulasegaram, J. Bonet, R.W. Lewis, M. Profit, A variational formulation based contact algorithm for rigid boundaries in two-dimensional SPH appli-
cations, Comput. Mech. 33 (33) (2004) 316–325.
[87] J. Feldman, J. Bonet, Dynamic refinement and boundary contact forces in SPH with applications in fluid flow problems, Int. J. Numer. Methods Eng.
72 (2007) 295–324.
[88] Y.W. Han, H.F. Qiang, J.L. Zhao, W.R. Gao, A new repulsive model for solid boundary condition in smoothed particle hydrodynamics, Acta Phys. Sin.
62 (4) (2013) 221–229.
[89] J.J. Monaghan, J.B. Kajtar, SPH particle boundary forces for arbitrary boundaries, Comput. Phys. Commun. 180 (10) (2009) 1811–1820.
[90] H.F. Qiang, Y.W. Han, K.P. Wang, W.R. Gao, Numerical simulation of water filling process based on new method of penalty function SPH, Eng. Mech.
28 (1) (2011) 245–250.
[91] B. Rogers, R. Dalrymple, SPH modeling of tsunami waves, Adv. Num. Model Simul. Tsun. Wave Runup. 10 (2007) 75–101.
[92] M.B. Liu, J.R. Shao, J.Z. Chang, On the treatment of solid boundary in smoothed particle hydrodynamics, Sci. China-Technol. Sci. 55 (1) (2012) 244–254.
[93] F. Bierbrauer, P.C. Bollada, T.N. Phillips, A consistent reflected image particle approach to the treatment of boundary conditions in smoothed particle
hydrodynamics, Comput. Meth. Appl. Mech. Eng. 198 (41) (2009) 3400–3410.
[94] J. Fang, R.G. Owens, L. Tacher, A. Parriaux, A numerical study of the SPH method for simulating transient viscoelastic free surface flows, J. Non
Newton. Fluid Mech. 139 (2006) 68–84.
[95] K. Gong, H. Liu, B.L. Wang, Water entry of a wedge based on SPH model with an improved boundary treatment, J. Hydrodyn. 21 (6) (2009) 750–757.
[96] M. Gómez-Gesteira, R.A. Dalrymple, Using a three-dimensional smoothed particle hydrodynamics method for wave impact on a tall structure, J.
Waterw. Port Coast. Ocean Eng. 130 (2) (2004) 63–69.
[97] M. Ferrand, D.R. Laurence, B.D. Rogers, D. Violeau, C. Kassiotis, Unified semi-analytical wall boundary conditions for inviscid, laminar or turbulent
flows in the meshless SPH method, Int. J. Numer. Methods Fluids 71 (4) (2013) 446–472.
[98] A. Mayrhofer, M. Ferrand, C. Kassiotis, D. Violeau, F.X. Morel, Unified semi-analytical wall boundary conditions in SPH: analytical extension to 3-D,
Numer, Algorithms 68 (1) (2014) 15–34.
[99] L.M. De, D.L. Touzé, B. Alessandrini, Normal flux method at the boundary for SPH, in: 4th SPHERIC Workshop, Nantes, France, 2009, pp. 149–156.
[100] J.C. Marongiu, F. Leboeuf, E. Parkinson, Numerical simulation of the flow in a Pelton turbine using the meshless method SPH and a new simple solid
boundary treatment, in: Proceedings of the 7th European Conference on Turbomachinery Fluid Dynamics and Thermodynamics, Athens, Greece, 2007,
pp. 849–856.
[101] A. Valizadeh, J.J. Monaghan, A study of solid wall models for weakly compressible SPH, J. Comput. Phys. 300 (2015) 5–19.
[102] M.B. Liu, G.R. Liu, K.Y. Lam, Constructing smoothing functions in smoothed particle hydrodynamics with applications, J. Comput. Appl. Math. 155 (2)
(2003) 263–284.
[103] X.F. Yang, M.B. Liu, Improvement on stress instability in smoothed particle hydrodynamics, Acta Phys. Sin. 61 (22) (2012) pp. 224701–224379.
[104] X.F. Yang, S.L. Peng, M.B. Liu, A new kernel function for SPH with applications to free surface flows, Appl. Math. Model. 38 (s15–16) (2014) 3822–3833.
[105] M.B. Liu, G.R. Liu, K.Y. Lam, A one-dimensional meshfree particle formulation for simulating shock waves, Shock Waves 13 (3) (2003) 201–211.
[106] F. Xu, Y. Zhao, R. Yan, T. Furukawa, Multidimensional discontinuous SPH method and its application to metal penetration analysis, Int. J. Numer. Meth.
Eng. 93 (2013) 1125–1146.
[107] M. Tatari, M. Shahriari, M. Raoofi, Numerical modeling of magneto-hydrodynamics flows using reproducing kernel particle method, Int. J. Numer.
Model.-Electron. Netw. Device Fields 29 (4) (2015) 548–564.
[108] M.B. Liu, W.P. Xie, G.R. Liu, Modeling incompressible flows using a finite particle method, Appl. Math. Model. 29 (2005) 1252–1270.
9654 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655

[109] M.B. Liu, G.R. Liu, Restoring particle consistency in smoothed particle hydrodynamics, Appl. Numer. Math. 56 (1) (2006) 19–36.
[110] G.A. Dilts, Moving least-squares particle hydrodynamics II: conservation and boundaries, Int. J. Numer. Methods Eng. 48 (10) (20 0 0) 1503–1524.
[111] D.M. Li, X.Y. Li, Y. Lin, Numerical simulation of droplet impacting liquid surface by SPH, Sci. China-Technol. Sci. 54 (7) (2011) 1873–1880.
[112] C. Andrea, A. Matteo, L.T. David, Theoretical considerations on the free-surface role in the smoothed-particle-hydrodynamics model, Phys. Rev. E. 79
(5Pt 2) (2009) 711–715.
[113] A. Kiara, K. Hendrickson, D.K.P. Yue, SPH for incompressible free-surface flows. Part I: Error analysis of the basic assumptions, Comput. Fluids 86
(2013) 611–624.
[114] A. Kiara, K. Hendrickson, D.K.P. Yue, SPH for incompressible free-surface flows. Part II: performance of a modified SPH method, Comput. Fluids 86
(2013) 510–536.
[115] G. Oger, M. Doring, B. Alessandrini, P. Ferrant, Two-Dimensional SPH simulations of wedge water entries, J. Comput. Phys. 213 (2) (2006) 803–822.
[116] L.D.G. Sigalotti, H. López, A Donoso, E Sira, J. Klapp, A shock-capturing SPH scheme based on adaptive kernel estimation, J. Comput. Phys. 212 (1)
(2006) 124–149.
[117] A. Colagrossi, G. Colicchio, C. Lugni, M. Brocchini, A study of violent sloshing wave impacts using an improved SPH method, J. Hydraul. Res. 48 (2010)
94–104.
[118] S. Nugent, H.A. Posch, Liquid drops and surface tension with smoothed particle applied mechanics, Phys. Rev. E. 62 (4) (20 0 0) 4968.
[119] Y. Meleán, L.D.G. Sigalotti, Coalescence of colliding van der Waals liquid drops, Int. J. Heat Mass Transf. 48 (19) (2005) 4041–4061.
[120] L.D.G. Sigalotti, J. Daza, A. Donoso, Modelling free surface flows with smoothed particle hydrodynamics, Condens. Matter Phys. 9 (2) (2006) 46.
[121] A.M. Aly, M. Asai, Y. Sonda, Modelling of surface tension force for free surface flows in ISPH method, Int. J. Numer. Methods Heat Fluid Flow. 23 (3)
(2013) 479–498.
[122] T.X. Su, L.Q. Ma, M.B. Liu, J.Z. Chang, A numerical analysis of drop impact on solid surfaces by using smoothed particle hydrodynamics method, Acta
Phys. Sin. 62 (6) (2013) 064702.
[123] N. Akinci, G. Akinci, M. Teschner, Versatile surface tension and adhesion for SPH fluids, ACM Trans. Graph (TOG) 32 (6) (2013) 182.
[124] J. Kordilla, A.M. Tartakovsky, T. Geyer, A smoothed particle hydrodynamics model for droplet and film flow on smooth and rough fracture surfaces,
Adv. Water Resour. 59 (2013) 1–14.
[125] M. Becker, M. Teschner, Weakly compressible SPH for free surface flows, in: Proceedings of the 2007 ACM SIGGRAPH/Eurographics symposium on
Computer animation, San Diego, CA, USA, 2007, pp. 209–217.
[126] M.Y. Zhang, H. Zhang, L.L. Zheng, Simulation of droplet spreading, splashing and solidification using smoothed particle hydrodynamics method, Int. J.
Heat Mass Transf. 51 (13) (2008) 3410–3419.
[127] G.Z. Zhou, W. Ge, J.H. Li, A revised surface tension model for macro-scale particle methods, Powder Technol 183 (1) (2008) 21–26.
[128] G.Z. Zhou, W. Ge, B. Li, X.P. Li, P. Wang, J.W. Wang, J.H. Li, SPH simulation of selective withdrawal from microcavity, Microfluid. Nanofluid. 15 (4)
(2013) 481–490.
[129] J.U. Brackbill, D.B. Kothe, C. Zemach, A continuum method for modeling surface tension, J. Comput. Phys. 100 (2) (1992) 335–354.
[130] J.P. Morris, Simulating surface tension with smoothed particle hydrodynamics, Int. J. Numer. Methods Fluids. 33 (3) (20 0 0) 333–353.
[131] L.D.G. Sigalotti, J. Daza, A. Donoso, Modelling free surface flows with smoothed particle hydrodynamics, Condens. Matter Phys. 9 (2) (2006) 359–366.
[132] X.Y. Hu, N.A. Adams, A multi-phase SPH method for macroscopic and mesoscopic flows, J. Comput. Phys. 213 (2) (2006) 844–861.
[133] S. Adami, X.Y. Hu, N.A. Adams, A new surface-tension formulation for multi-phase SPH using a reproducing divergence approximation, J. Comput.
Phys. 229 (13) (2010) 5011–5021.
[134] M.Y. Zhang, Simulation of surface tension in 2D and 3D with smoothed particle hydrodynamics method, J. Comput. Phys. 229 (19) (2010) 7238–7259.
[135] M.Y. Zhang, S.D. Zhang, H. Zhang, L.L. Zheng, Simulation of surface-tension-driven interfacial flow with smoothed particle hydrodynamics method,
Comput. Fluids 59 (2012) 61–71.
[136] H.F. Qiang, K. Liu, F.Z. Chen, Numerical implementation of deformation and motion of droplet at the interface between vapor and solid surface with
smoothed particle hydrodynamics methodology, Acta Phys. Sin. 61 (20) (2012) 204701.
[137] S.E. Hieber, P. Koumoutsakos, A Lagrangian particle level set method, J. Comput. Phys. 210 (1) (2005) 342–367.
[138] M.Y. Zhang, X.L. Deng, A sharp interface method for SPH, J. Comput. Phys. 302 (C) (2015) 469–484.
[139] G.T. Duan, S. Koshizuka, B. Chen, A contoured continuum surface force model for particle methods, J. Comput. Phys. 298 (2015) 280–304.
[140] T. Jiang, L.G. Lu, W.G. Lu, Numerical study of collision process between two equal diameter liquid micro-droplets using a modified smoothed particle
hydrodynamics method, Acta Phys. Sin. 62 (22) (2013) 224701.
[141] T. Jiang, J. Ouyang, X.K. Zhao, J.L. Ren, The deformation process of viscous liquid drop studied by using kernel gradient corrected SPH method, Acta
Phys. Sin. 60 (5) (2011) 054701.
[142] N. Akinci, G. Akinci, M. Teschner, Versatile surface tension and adhesion for SPH fluids, ACM Trans. Graph (TOG). 32 (6) (2013) 182.
[143] D. Liu, Y.C. Guo, W.Y. Lin, Droplet deformation and surface tension modeling using the smoothed particle hydrodynamics method, J. Tsinghua Univ.
53 (3) (2013) 384–388.
[144] J.J. Monaghan, A. Rafiee, A simple SPH algorithm for multi-fluid flow with high density ratios, Int. J. Numer. Methods Fluids. 71 (5) (2013) 537–561.
[145] H.P. Gröbelbauer, T.K. Fanneløp, R.E. Britter, The propagation of intrusion fronts of high density ratios, J. Fluid Mech. 250 (1993) 669–687.
[146] A. Valizadeh, M. Shafieefar, J.J. Monaghan, S.A.A.S. Neyshaboori, Modeling two-phase flows using SPH method, J. Appl. Sci. 8 (21) (2008) 3817–3826.
[147] A. Colagrossi, M. Landrini, Numerical simulation of interfacial flows by smoothed particle hydrodynamics, J. Comput. Phys. 191 (2) (2003) 448–475.
[148] J.J. Monaghan, On the problem of penetration in particle methods, J. Comput. Phys. 82 (1) (1989) 1–15.
[149] T. Belytschko, Y. Krongauz, J. Dolbow, C. Gerlach, On the completeness of meshfree particle methods, Int. J. Numer. Methods Eng. 43 (5) (1998)
785–819.
[150] Z. Chen, Z. Zong, M.B. Liu, L. Zou, H.T. Li, C. Shu, An SPH model for multiphase flows with complex interfaces and large density differences, J. Comput.
Phys. 283 (2015) 169–188.
[151] F. Ott, E. Schnetter, A modified SPH approach for fluids with large density differences, Arxiv Physics E-prints. (2003) 3112.
[152] A.M. Tartakovsky, P. Meakin, A smoothed particle hydrodynamics model for miscible flow in three-dimensional fractures and the two-dimensional
Rayleigh–Taylor instability, J. Comput. Phys. 207 (2) (2005) 610–624.
[153] X.Y. Hu, N.A. Adams, A constant-density approach for incompressible multi-phase, SPH. J. Comput. Phys. 228 (6) (2009) 2082–2091.
[154] N. Grenier, M. Antuono, A. Colagrossi, D.L. Touzé, B. Alessandrini, An Hamiltonian interface SPH formulation for multi-fluid and free surface flows, J.
Comput. Phys. 228 (22) (2009) 8380–8393.
[155] J.J. Monaghan, SPH without a tensile instability, J. Comput. Phys. 159 (2) (20 0 0) 290–311.
[156] H. Wendland, Piecewise polynomial, positive definite and compactly supported radial functions of minimal degree, Adv. Comput. Math. 4 (1995)
389–396.
[157] M.S. Shadloo, A. Zainali, M. Yildiz, Simulation of single mode Rayleigh–Taylor instability by SPH method, Comput. Mech. 51 (5) (2013) 699–715.
[158] B.W. Ritchie, P.A. Thomas, Multiphase smoothed-particle hydrodynamics, Mon. Not. R. Astron. Soc. 323 (3) (2001) 743–756.
[159] Z. Xu, P. Meakin, A.M. Tartakovsky, Diffuse-interface model for smoothed particle hydrodynamics, Phys. Rev. E 79 (3) (2009) 036702.
[160] D.M. Anderson, G.B. McFaddenr, A.A. Wheeler, Diffuse-interface methods in fluid mechanics, Annu. Rev. Fluid Mech. 30 (1) (1998) 139–165.
[161] A.K. Das, P.K. Das, Incorporation of diffuse interface in smoothed particle hydrodynamics: Implementation of the scheme and case studies, Int. J.
Numer. Methods Fluids 67 (6) (2011) 671–699.
[162] A.K. Das, P.K. Das, Equilibrium shape and contact angle of sessile drops of different volumes—Computation by SPH and its further improvement by
DI, Chem. Eng. Sci. 65 (13) (2010) 4027–4037.
 Z.-B. Wang et al. / Applied Mathematical Modelling 40 (2016) 9625–9655 9655

[163] A.K. Das, P.K. Das, Simulation of drop movement over an inclined surface using smoothed particle hydrodynamics, Langmuir 25 (19) (2009)
11459–11466.
[164] A.K. Das, P.K. Das, Bubble evolution and necking at a submerged orifice for the complete range of orifice tilt, AICHE J. 59 (2) (2013) 630–642.
[165] A.K. Das, P.K. Das, Multimode dynamics of a liquid drop over an inclined surface with a wettability gradient, Langmuir 26 (12) (2010) 9547–9555.
[166] J.J. Monaghan, Gravity currents and solitary waves, Physica D 98 (2) (1996) 523–533.
[167] J.J. Monaghan, A. Kos, Scott Russell’s wave generator, Phys. Fluids. 12 (3) (20 0 0) 622–630.
[168] J.J. Monaghan, Implicit SPH drag and dusty gas dynamics, J. Comput. Phys. 138 (2) (1997) 801–820.
[169] L.C. Qiu, Numerical simulation of deformation process of viscous liquid drop based on the incompressible smoothed particle hydrodynamics, Acta
Phys. Sin. 62 (12) (2013) 124702.
[170] S. Adami, X.Y. Hu, N.A. Adams, A conservative SPH method for surfactant dynamics, J. Comput. Phys. 229 (5) (2010) 1909–1926.
[171] M.S. Shadloo, A. Rahmat, M. Yildiz, A smoothed particle hydrodynamics study on the electrohydrodynamic deformation of a droplet suspended in a
neutrally buoyant Newtonian fluid, Comput. Mech. 52 (3) (2013) 693–707.
[172] H.B. Xiong, J. Zhu, Study of droplet deformation, heat-conduction and solidification using incompressible smoothed particle hydrodynamics method,
in: Proceedings of the 9th International Conference on Hydrodynamics, Beijing, China, 2010, pp. 150–153.
[173] T. Breinlinger, P. Polfer, A. Hashibon, T. Kraft, Surface tension and wetting effects with smoothed particle hydrodynamics, J. Comput. Phys. 243 (2013)
14–27.
[174] L.Q. Ma, J.Z. Chang, H.T. Liu, M.B. Liu, Numerical simulation of droplet impact on liquid with smoothed particle hydrodynamics method, Acta Phys.
Sin. 61 (5) (2012) 054701.
[175] N. Nishio, K. Yamana, Y. Yamaguchi T. Inaba, K. Kuroda, T. Nakajima, K. Ohno, H. Fujimura, Large-scale SPH simulations of droplet impact onto a liquid
surface up to the consequent formation of Worthington jet, Int. J. Numer. Methods Fluids 63 (12) (2010) 1435–1447.
[176] D.M. Li, Z.C. Wang, L. Bai, X. Wang, Investigations on the process of droplet impact on an orifice plate, Acta Phys. Sin. 62 (19) (2013) 194704.
[177] A. Acevedo-Malavé, M. García-Sucre, Coalescence collision of liquid drops I: off-center collisions of equal-size drops, AIP Adv. 1 (3) (2011) 032117.
[178] A. Acevedo-Malavé, M. García-Sucre, 3D Coalescence Collision of Liquid Drops using Smoothed Particle Hydrodynamics, INTECH Publishers, Rijeka,
2011.
[179] A. Acevedo-Malavé, M. García-Sucre, Coalescence collision of liquid drops II: off-center collisions of unequal-size drops, AIP Adv. (3) (2011) 032118.
[180] A. Acevedo-Malavé, M. García-Sucre, Many drops interactions I: Simulation of coalescence, flocculation and fragmentation of multiple colliding drops
with smoothed particle hydrodynamics, J. Comput. Multiphas. Flow. 4 (2) (2012) 121–134.
[181] A. Acevedo-Malavé, M. García-Sucre, Many drops interactions II: Simulation of coalescence, flocculation and fragmentation of multiple colliding drops
with smoothed particle hydrodynamics, J. Comput. Multiphas. Flow. 4 (2) (2012) 135–146.
[182] A. Acevedo-Malavé, M. García-Sucre, Clusters formation of drops from many droplets collisions: a 3D smoothed particle hydrodynamics approach, J.
Comput. Multiphas. Flow 4 (2) (2012) 147–158.
[183] A. Acevedo-Malavé, A three-dimensional SPH approach for modelling the collision process between liquid drops: The formation of clusters of un-
equal-sized drops, Computational and experimental fluid mechanics with applications to physics, Engineering and the Environment, Springer Inter-
national Publishing, Berlin, 2014.
[184] A. Acevedo-Malavé, Numerical modeling of unequal-size droplets collisions using a lagrangian mesh-free particle method, CFD Lett. 5 (1/2) (2013)
32–42.
[185] H.F. Qiang, C. Shi, F.Z. Chen, Y.W. Han, Simulation of two-dimensionaldroplet collisions based on SPH method of multi-phase flows with large density
differences, Acta Phys. Sin. 62 (21) (2013) 214701.
[186] M. Vahabi, K. Sadeghy, Simulating bubble shape during its rise in Carreau-Yasuda fluids using WC-SPH method, Nihon Reoroji Gakk 41 (5) (2014)
319–329.
[187] N. Grenier, D.L. Touze, M. Antuono, A. Colagrossi, An improved SPH method for multi-phase simulations, in: Proceedings of the 8th International
Conference on Hydrodynamics (ICHD2008), Nantes, France, 2008.
[188] M. Sussman, P. Smereka, S. Osher, A level set approach for computing solutions to incompressible two-phase flow, J. Comput. Phys. 114 (1994)
146–159.
[189] K. Szewc, J. Pozorski, J.P. Minier, Simulations of single bubbles rising through viscous liquids using smoothed particle hydrodynamics, Int. J. Multiphas.
Flow. 50 (2013) 98–105.
[190] N. Grenier, D.L. Touzé, A. Colagrossi, M. Antuono, G. Colicchio, Viscous bubbly flows simulation with an interface SPH model, Ocean Eng 69 (2013)
88–102.
[191] A.K. Das, P.K. Das, Bubble evolution through submerged orifice using smoothed particle hydrodynamics: basic formulation and model validation,
Chem. Eng. Sci. 64 (10) (2009) 2281–2290.
[192] A.K. Das, P.K. Das, Bubble evolution through a submerged orifice using smoothed particle hydrodynamics: effect of different thermophysical proper-
ties, Ind. Eng. Chem. Res. 48 (18) (2009) 8726–8735.