You are on page 1of 21

Available online at www.sciencedirect.

com

ScienceDirect

Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209


www.elsevier.com/locate/cma

An SPH modeling of bubble rising and coalescing in three


dimensions
Aman Zhang a,∗ , Pengnan Sun a,b , Furen Ming a
a College of Shipbuilding Engineering, Harbin Engineering University, Harbin 150001, China
b CNR-INSEAN, Marine Technology Research Institute, Rome, Italy

Received 14 November 2014; received in revised form 15 May 2015; accepted 19 May 2015
Available online 11 June 2015

Highlights
• A robust three-dimensional multiphase SPH model is applied taking care on combining the advantages of different SPH models.
• A suitable multiphase interface treatment is applied and the time integration method is modified for better stability.
• Background pressure is shown to play an important role in keeping the multiphase interface stable when the Bond number is
relatively small.
• Bubble rising and coalescing are simulated in three dimensions.
• A wide validation is performed with both experimental data and other numerical results.

Abstract

The numerical simulation of bubbly flows is challenging due to the unstable multiphase interfaces with large density ratios and
viscous ratios. The multiphase Smoothed Particle Hydrodynamics (SPH) method is applied in this paper to simulate the phenomena
of bubbles rising and coalescing in three dimensions. Firstly, the multiphase SPH model is introduced in detail, including the
derivation of the discretized governing equations based on the principle of virtual work, the viscous force, the multiphase interface
treatment, the time-stepping scheme, the boundary implementation, etc. Considering the expensive computational cost in three-
dimensional (3-D) SPH simulations, the effects of the scale of the computational domain and the density ratio on the multiphase
interface are numerically investigated in order to decrease the amount of calculation. Afterwards, several cases of single bubbles
rising through viscous fluids are tested and the SPH results are validated by both the experimental data and other numerical results
in the literature. Furthermore, the phenomena of bubbles coalescing in both vertical and horizontal directions are simulated and
the results agree well with the experimental data. It is found that the background pressure in the equation of state is essential to
keep the multiphase interface smooth and stable when the Bond number is relatively small. The fair agreements between the results
of SPH and other reference results demonstrate that the present multiphase SPH model is robust and stable enough to accurately
simulate the dynamic phenomena of rising bubbles in different conditions, which can give a reference for engineering applications.
⃝c 2015 Elsevier B.V. All rights reserved.

Keywords: SPH; Surface tension; Rising bubble; Bubble coalescence; Multiphase flow

∗ Correspondence to: College of Shipbuilding Engineering, Harbin Engineering University, Room 513, Chuanhai Building, No. 145 Nantong
Street, Nangang District, Harbin 150001, China. Tel.: +86 13766831003; fax: +86 451 82518296.
E-mail addresses: zhangaman@hrbeu.edu.cn (A. Zhang), sunpengnan@yeah.net (P. Sun), mingfuren@gmail.com (F. Ming).

http://dx.doi.org/10.1016/j.cma.2015.05.014
0045-7825/⃝ c 2015 Elsevier B.V. All rights reserved.
190 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

1. Introduction

The research of bubble dynamics has potential applications in the exploitation of marine resources (e.g. oil and gas
resources, combustible ice, etc.). During the exploitation and processing of these resources, considerable bubbles may
grow and rise up in the fluid. Therefore, the mechanism research about the dynamic behavior of these rising bubbles
is essential [1,2].
There are unstable multiphase interfaces with large density ratios and viscous ratios in bubbly flows. Moreover,
these unstable interfaces may break and recombine during the flow evolution. Therefore, bubbly flows are extremely
complex and hard to be predicted. Generally, there are three kinds of approaches that are usually adopted by the
researchers to simulate these problems [3]. The approaches are respectively theoretical analysis, experimental obser-
vation and numerical simulation. However, the theoretical analysis is only restricted to extremely ideal conditions.
Therefore, it is essential for the mechanism explanation for some simple cases, but not so suitable for practical ap-
plications with complex conditions. Regarding the experimental observation, the cost is usually very expensive and
the experimental period is very long. In addition, after the experimental equipment is completely constructed, it is
difficult to extend or add new functions while it is convenient for numerical tools.
There are some CFD (Computational Fluid Dynamics) methods showing strong abilities in the simulation of rising
bubbles and the interaction between them, e.g. Volume of Fluid (VOF) method [4], Level Set (LS) method [5,6], Front
Tracking Method (FTM) [7–9], Lattice Boltzmann Method (LBM) [10–13], etc. The advantage and disadvantage
of these mesh-based methods are analyzed (see e.g. Annaland et al. [4]). Mesh-free methods, e.g. SPH, Moving
Particle Semi-implicit (MPS), are considered as CFD methods of the next generation since they are very suitable
and convenient for the simulation of dynamic phenomena with large deformation, moving boundary and multiphase
mixing [14,15]. SPH method was first proposed by Gingold and Monaghan [16] and Lucy [17] to simulate the
astrophysical phenomena and then widely used in the simulation of hydrodynamics [18,19]. Up to now, different SPH
variants [20] have been widely applied in free surface flows [21,22], violent fluid–solid interactions [19], wave–body
interactions [23,24] and multiphase flows [21,25–28], etc. However, to our knowledge, there are few papers discussing
the ability of SPH in the simulation of bubbles rising [29] and coalescing [30,31] in three dimensions.
In the early literature, Colagrossi and Landrini [21] and Grenier et al. [27] simulated the rising and splitting of
a single two-dimensional (2-D) bubble in the viscous fluid and good agreements were achieved between the results
of SPH and LS method [5]. However, in their works, a numerical surface tension was adopted instead of the real
one. Afterwards, Grenier et al. [1] added real surface tension into a novel Hamiltonian SPH solver, and then rising
and coalescing of 2-D bubbles were simulated, while 3-D results were not given in that paper. Bubble necking at a
submerged orifice was simulated by Das and Das [32–35]. In their works, both 2-D and 3-D results were presented.
Das and Das [32–35] mainly focus on the dynamic phenomena of bubbles detaching from the submerged orifice,
while the rising process of the simulated bubbles was not emphasized or compared with the experimental data. Szewc
et al. [36] made a detailed description of the simulation of 3-D rising bubbles and the results agree well with the
experimental data. However, in their model, the changing of the values of interface sharpness force may cause the
terminal bubble shapes completely different from the reference results. In Szewc et al. [37], another kind of interface
treatment was proposed to improve the interface sharpness, but it still depends on an empirical coefficient which may
completely change the final results.
In the present work, a multiphase SPH model, which is built taking care on combining the advantages of different
SPH models [1,25,27,38], is applied to simulate the phenomena of bubbles rising and coalescing. The numerical
results are validated by the experimental data in the literature [29–31]. The problem that the final results are greatly
affected by the interface sharpness force mentioned in Szewc et al. [36,37] is solved by using a density-weighted
surface tension model presented in Adami et al. [38]. Besides, the interface sharpness force proposed in Grenier
et al. [27], which causes negligible effects on the final results, is applied to avoid the interface penetration. It is worth
underlining that, besides the surface tension, the viscous force is also crucial for an accurate simulation of bubbly
flows. Grenier et al. [1] compared several kinds of viscous force formulas. The conclusion is that the viscous formula
of Morris et al. [39] gives the most accurate results for the simulation of multiphase flows. In the present work, the
viscous term in Hu and Adams [25], which is similar to the one proposed by Morris et al. [39], is applied even though
the angular momentum conservation is not so accurately satisfied. Regarding the other numerical techniques, for the
cases with small Bond numbers, the background pressure should be added in the equation of state to improve the
stability of the multiphase interface. Since the interface sharpness force may cause serious pressure oscillations in the
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 191

bubble, a modified prediction–correction time-stepping scheme is used, in which the velocity is updated firstly and
then the updating of the particle position is based on the recently updated velocity, which is very important in keeping
the simulation stable.
The paper is organized as follows: in Section 2, the multiphase SPH model is introduced in detail; In Section 3,
the effects of the scale of the computational domain and the density ratio are numerically investigated. Some useful
conclusions are drawn for reducing the computational cost; In Section 4, cases of rising and deforming of a single
bubble are simulated firstly and the results are compared with some reference results including experimental data and
other numerical results. After that, the phenomena of bubbles coalescing in different directions are simulated and the
mechanism of these interesting dynamic behaviors is analyzed.

2. Multiphase SPH model

2.1. Governing equations

In SPH, the fluid domain is discretized into a number of moving fluid particles which carry general physical values
such as density, velocity, acceleration, etc. Therefore, the Navier–Stokes equations in Lagrangian view are adopted to
describe the flow system [15]. The mass conservation equation is as follows:

= −ρ∇ · u (1)
Dt
where D denotes the material derivative. ρ, u are the density and velocity of the particles, respectively. These particles
are driven by the pressure gradient ∇ p, viscous stress FV , body force F B and surface tension F S (F S is calculated only
for the particles adjacent to the multiphase interface). Therefore, according to Newton’s second law, the momentum
conservation equation is as follows:
Du
ρ = −∇ p + FV + F B + F S . (2)
Dt
In the simulation of freely rising bubbles, one may consider the fluid as incompressible, so the viscous stress FV
can be calculated as follows [25]:

FV = η∇ 2 u (3)
where η is the dynamic viscous coefficient. According to the Continuum Surface Force (CSF) model in Brackbill
et al. [40], the surface tension is considered as body force as follows:

F S = −βκ n̂λ (4)


where β, κ, n̂ denote the surface tension coefficient, the multiphase interface curvature and the unit normal,
respectively. λ is a weight function denoting the distribution of the magnitude of the surface tension at both sides of
the multiphase interface. Due to the difference of particle density across the multiphase interface, the weight function
should be proportional to the density in order to keep the continuity of acceleration across the interface (see Section 2.4
for more details).
The Navier–Stokes equations are not closed at this stage. It is common to consider the fluid as barotropic and using
an equation of state to uncouple the mass and momentum conservation equations [18]. The equations of state for the
denser phase denoted by subscript l and the lighter phase denoted by subscript g are as follows:
 γl
c2 ρ0l ρ

pl = l − 1 + pb
γl ρ
 0l γg (5)
cg2 ρ0g ρ

pg = − 1 + pb
γg ρ0g
where γl = 7 and γg = 1.4 [21]. ρ0l and ρ0g are the reference density when the pressure p is equal to pb . pb
is the background pressure used to prevent the SPH specific numerical instability namely tensile instability and
keep the particles distributing uniformly [41]. However, too large background pressure would induce extra particle
192 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

resettlements and make the numerical results unreasonable (see Colagrossi et al. [42] for more details). Therefore, the
background pressure pb should be restricted to the minimum value as long as the simulation is stable. Note that pb is
set to zero if it is not mentioned specifically in the present work.
The artificialsoundspeed cl of the denser phase is determined through allowing the highest pressure variation ∆ p
satisfying ∆ p/ ρ0l cl2 ≪ 1, and ∆ p can be estimated by the velocity, the reference water depth and the surface
tension (see Grenier et al. [1] for more details). Finally, the following criteria are applied:

u max 2
 
gH 2β
≈ 0.01, ≈ 0.01, ≈ 0.01 (6)
cl 2
cl ρ0l dcl2
where u max , H and d are the maximum velocity, the undisturbed fluid depth and the bubble diameter at the initial
stage, respectively. The above criteria ensure the density variation inside the fluiddomain is within 1% [1,18]. The
artificial sound speed cg of the lighter phase is calculated as cg = cl2 γg ρ0l / γl ρ0g [21]. Note that, cg is much larger
than cl when the density ratio ρ0l /ρ0g is considerable, which makes the time step small and therefore the simulation
becomes inefficient (for the detailed analysis, see Section 3.2).

2.2. Discretized governing equations

Up to now, a number of SPH variants were proposed to simulate multiphase flows with large density ratios and
viscous ratios in quite a few published papers, e.g. [21,25–28]. The multiphase SPH model proposed by Hu and
Adams [25] is applied in this paper. The free surface cannot be considered in this SPH variant since the kernel
function adjacent to the free surface is truncated and it may cause serious density singularity. However, one may add
gas particles above the fluid surface to explore the surface effects. The interpolating function used in the multiphase
SPH variant is as follows:
Wi j Wi j
χi j =  = (7)
Wi j σ (ri )
j

where subscripts i and j are the particle index. r is a vector denoting the particle position. Wi j is an abbreviated
notation for the kernel function W (r, h), in which r = ri − r j  denotes the distance between the neighboring particles
ri and r j and h is the smoothing length. In the present work, the renormalized Gaussian kernel function [27,43] is
used due to its preferable properties analyzed in Jin and Ding [44]. This kernel function is as follows:
−(r/ h)2 − C
  
α e
 0
if r ≤ 3h
n
W (r, h) = 1 − C1

 (8)
0 otherwise
C0 = e−9 ; C1 = 10C0 .
Considering the computational cost in 3-D cases, h = ∆x is used  present work and ∆x denotes the initial
in the
particle spacing in Cartesian coordinate system. αn equals to 1/ π n/2 h n , where n denotes the spatial dimension. As

the smoothing length h approaches zero, the renormalized Gaussian kernel becomes a Dirac function [43].
The volume approximation of particle i is as follows [25]:

Wi j V j
  Wi j j 1
Vi ≈ χi j V j = Vj = ≃ . (9)
j j
σ (ri ) σ (ri ) σ (ri)

From the derivation above, one can see clearly the volume approximation of particle i equals to 1/σ (ri ), which is
independent from the density of the adjacent particles. Therefore, the volume approximation is applicative for those
simulations with large density ratios at multiphase interfaces. This is the key idea of the present multiphase SPH
model. Based on ρi = m i /Vi , the density approximation is as follows [25]:

ρi = m i σi = m i Wi j . (10)
j
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 193

With the density approximation equation, one can obtain the corresponding discretized form of ∇ · (δw), where δw
is the virtual displacement inside the continuous medium. The detailed derivation of ∇ · (δw) is shown in Appendix A.
Based on ∇ · (δw), one may obtain the discretized form of ∇ p based on the principle of virtual work (hereinafter
abbreviated as PVW), which ensures the conservations of both the linear and angular momenta.
When the work of the stress tensor on the boundary is zero, Colagrossi et al. [45] gave the following formula
derived through PVW:
 
⟨∇ p⟩i · δwi Vi = − pi ⟨∇ · (δw)⟩i Vi . (11)
i i

Eq. (11) means that the work of the stress tensor inside the fluid domain is equal to the variation of the internal
energy [45], which ensures an accurate conservation of energy. Substituting Eq. (A.7) into Eq. (11), one gets
  
⟨∇ p⟩i · δwi Vi = pi Vi2 δwi − δw j · ∇i Wi j .

(12)
i i j

Eq. (12) can be rewritten as


    
⟨∇ p⟩i · δwi Vi = δwi · pi Vi2 ∇i Wi j − pi Vi2 δw j · ∇i Wi j . (13)
i i j i j

According to the symmetrical characteristic of the kernel gradient, one has ∇i Wi j = −∇ j Wi j , based on which, the
second term on the right-hand side of Eq. (13) is transformed as
   
− pi Vi2 δw j · ∇i Wi j = δwi · p j V j2 ∇i Wi j . (14)
i j i j

Substituting Eq. (14) into Eq. (13), and considering the randomicity of δwi , one gets
1  
⟨∇ p⟩i = pi Vi2 + p j V j2 ∇i Wi j . (15)
Vi j

In summary, the governing equations are as follows:



ρi = m i Wi j



 j 

1 1 

 Dui 
2 2 V B S
= − pi Vi + p j V j ∇i Wi j + Fi + Fi + Fi (16)
 Dt ρi Vi j
  γ
ρ

 

− 1 + pb .

 p = p0
ρ0
The above equations are the same as the governing equations in the multiphase SPH model of Hu and Adams [25],
but the derivation in the present work is based on PVW, which proves the property of energy conservation. The
capability of the multiphase SPH model in the simulation of rising bubbles will be shown in the numerical results in
Sections 3 and 4.

2.3. Viscous force

The dynamic behavior of freely rising bubbles is greatly influenced by the viscosity of the surrounding liquid [29].
Therefore, the accuracy of the numerical result is directly related to the calculation of viscous stress. There are mainly
two kinds of discretized viscous equations in SPH, which have been widely validated and used by SPH researchers
[1,27,41,46]. The first one is the formula proposed by Monaghan and Gingold [47] (hereinafter MGF). When the
viscous coefficient is constant in the whole flow domain, the viscous force is
 (ui − u j ) · (ri − r j )
V
FMGF (ri ) = ξη  ∇i Wi j V j (17)
ri − r j 2 + (εh)2

j
194 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

Fig. 1. Sketch of the transitional domain near the multiphase interface.

where ξ is equal to 2 (n + 2), which depends on the spatial dimension n [46] and ε is set to 0.01 in order to keep
the denominator non-vanishing if two particles are too close. Generally, in the simulation of free surface flow, MGF
is widely used as a viscous damping term due to its preferable consistency near the free surface (see Colagrossi
et al. [46] for more details). In addition, MFG conserves both linear and angular momenta since the interacting forces
are all along the connecting lines between the particle pairs. In the simulation of multiphase flows, the dynamic
viscous coefficient needs an adjustment on the multiphase interface. Here, ηi j = 2ηi η j / ηi + η j proposed by Hu
 

and Adams [25] is used, where i and j denote the particles of different phases, and then we have
 2ηi η j (ui − u j ) · (ri − r j )
V
FMGF (ri ) = ξ ∇i Wi j V j . (18)
ηi + η j ri − r j 2 + (εh)2
 
j

A formula similar to Eq. (18) was used in Grenier et al. [27]. Nevertheless, through the numerical tests in Grenier
et al. [1], the viscous formula originally proposed by Morris et al. [39] gives relatively more accurate results. In Hu
and Adams [25], a formula similar to Morris et al. [39] was proposed (hereinafter HEA) as follows:
 2ηi η j  2  (r − r ) · ∇ W (u − u )
i j i ij i j
V
FHEA (ri ) = Vi + V j2  . (19)
(ηi + η j ) ri − r j 2 + (εh)2 Vi

j

Eq. (19) is more suitable for the discretized equation system in the present work. Therefore, it is used in all the test
cases. HEA does not strictly conserve the angular momentum, but since the computational time here is not very long,
the simulation is stable and the results are acceptable.

2.4. The multiphase interface treatment

Due to the difference of Van der Waals’ force in different fluid phases, there exists surface tension at the multiphase
interface. The surface tension should not be neglected since it plays a crucial role in the deformation and motion of
rising bubbles [29]. Based on the continuous surface tension model [40], there have been several kinds of surface
tension models in SPH, e.g. Morris [48], Adami et al. [38], Szewc et al. [36], Breinlinger et al. [49], Zainali
et al. [50], etc. The surface tension model [38] which is suitable for the interface with considerable density ratio is
applied here. Besides, an interface sharpness force is added into the surface tension model in order to avoid interface
penetration.
Firstly, as is plotted in Fig. 1, a transitional domain near the multiphase interface is defined using a weight function
which determines the distribution of the magnitude of the surface tension (it is a kind of body force according to
the CSF model in Brackbill et al. [40]). Due to the considerable density difference, the weight function needs some
adjustments in order to keep the continuity of acceleration across the interface. According to Adami et al. [38], the
j
weight function and the normal direction n is calculated by defining an index number ci for particle i, while j is the
j
particle adjacent to the particle i. Inspired by Adami et al. [38], we directly define ci is as follows:

 2ρi if particle i and j belong to different phases
j
ci = ρi + ρ j (20)
0 if particle i and j belong to the same phase.

A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 195

By calculating the gradient of ci , the unit normal n̂ is achieved as follows [38]:


ni ∇ci
n̂i = = (21)
|ni | |∇ci |
 
j
where ∇ci can be calculated similar to Adami et al. [38] using an inter-particle-averaged value cii + ci /2 as
follows:
j
1  2 ci + ci
∇ci = (Vi + V j2 ) i ∇i Wi j (22)
Vi j 2
j
where cii = 0 and ci is calculated by Eq. (20). The modulus of ∇ci can serve as the weight function [38]. Through
calculating the divergence of n̂, one obtains the curvature of the multiphase interface. The reproducing divergence
approximation formula in Adami et al. [38] is applied here to improve the accuracy at the boundary of the transitional
j
domain where the kernel function is truncated. Besides, from the definition of ci , it can be deduced that for one fluid
phase, the unit normal n̂ always points from itself to the other phase, as is shown in Fig. 1. Therefore, in the calculation
j
of the curvature κi , we add a coefficient ϕi to invert the direction of the unit normal n̂ j in the case that particle i and
j
particle j belong to different phases. ϕi is defined as follows:

j −1 if particle i and j belong to different phases
ϕi = (23)
1 if particle i and j belong to the same phase.
Finally, the curvature κi is calculated as follows:
 j

n̂i − ϕi n̂ j · ∇i Wi j V j
j
κi = −∇ · n̂i = −n      (24)
ri − r j  · ∇i Wi j  V j
j

where n is the spatial dimension. Substituting the interface curvature κ, unit normal n̂ and weight function |∇ci | into
Eq. (4), the surface tension can be calculated as follows [38]:

FiS = −βκi |∇ci | n̂i = β(∇ · n̂i )∇ci . (25)


In order to keep the interface smooth, the interface sharpness force, which was proposed in Grenier et al. [27] and
proven to cause negligible effects on the accuracy of the surface tension model in Szewc et al. [36], is added into
the momentum equation. The interface sharpness force acts like a repulsive force perpendicular to the multiphase
interface, and then there will be a narrow spacing on the interface, which makes the interface clear and smooth. Here
the interface sharpness coefficient is 0.08, and finally the momentum equation is as follows:
 
Dui 1 1  2 2

V B S
= − pi Vi + p j V j ∇i Wi j + Fi + Fi + Fi
Dt ρi Vi j
0.08     
− | pi | Vi2 +  p j  V j2 ∇i Wi j . (26)
ρi Vi j

2.5. The solid wall boundary and time-stepping scheme

2.5.1. The solid wall boundary


Since the solid wall boundary in the present work simply consists of plane walls and right angles, the mirroring
ghost particle boundary is implemented. Any real fluid particle getting close to the wall boundary will be reflected
by a corresponding ghost particle on the other side (see Colagrossi et al. [21] for more details). On the solid wall, a
free-slip boundary condition is applied for all the cases in Sections 3 and 4 since the effect of the boundary layer on
the bubble motion is inessential.
196 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

2.5.2. Time-stepping scheme


Before integrating the equations of momentum and motion, the magnitude of the time step ∆t of the nth step is
calculated based on the surface tension coefficient β, the dynamic viscous coefficient η, the artificial sound speed c
and the maximal velocity u max as follows [1]:
  3 1/2
 ρh

 ∆t st = C F L st
2πβ



ρh 2

∆t η = C F L η
(27)


 η

 h
∆tc = C F L c

c + u max
where C F L st = 0.5, C F L η = 0.125, C F L c = 1.0 are used. The final time step ∆t is determined by

∆t = min ∆tst , ∆tη , ∆tc .


 
(28)
In the present work, a modified prediction–correction time-stepping scheme is applied, which is similar to the
combination of the velocity-Verlet scheme applied in Adami et al. [51] and the standard prediction–correction time-
stepping scheme in [18,52]. Also, the present time integration scheme is similar to the modified Verlet scheme applied
in Molteni and Colagrossi [43] which allows the adoption of a larger CFL factor. The motivation of modifying the
integration method is that, during our numerical test, the initial stage of the simulation may become quite unstable
due to some pressure oscillations caused by the interface sharpness force. This problem has also been mentioned in
Monaghan et al. [28] where the solution is to add some artificial damping at the initial stage. While in the present
work, we firstly update the velocity similar to the velocity-Verlet scheme [51], but the updating of the particle position
is based on the recently updated velocity, which is very important in keeping the calculation stable. The modified
prediction–correction time-stepping scheme is described as below.
Firstly, the prediction step is as follows:
 n+1/2  n

 ρi = mi Wi j
j



∆t Du n
  
n+1/2 n (29)
u i = u i +

 2 Dt i
rn+1/2 = rn + ∆t un+1/2 .



i i
2 i
Secondly, the correction step is as follows:
 n+1/2
ρ
 n+1

 i = m i Wi j

 j

Du n+1/2
 
n+1 n (30)
 u i = u i + ∆t


 Dt i
ri = ri + ∆tuin+1 .
 n+1 n

Similar to the standard prediction–correction scheme [18,52], the present integration scheme is also second-order
accurate. In the present work, the calculation code is programmed using FORTRAN and a parallel computation is
realized using OpenMP method based on the Intel (R) Core (TM) i7-3770 processor with 8 threads.

2.5.3. Convergence study


In this part, a convergence study is carried out with the 2-D rising bubble case introduced in Hysing et al. [53].
The results of TP2D code in [53] are used as the reference data. The initial condition for this test case is shown in
Fig. 2(a) where the radius of the bubble is R = 0.25 m. On the top and bottom of the solid wall, a no-slip boundary
condition is applied while on the other two sides, a free-slip boundary condition is applied. The density and viscosity
of the liquid are ρl = 1000 kg/m3 and ηl = 10 Pa s, respectively while the ones of the bubble are ρg = 100 kg/m3
and ηg = 1 Pa s, respectively. The surface tension coefficient is set to be β = 24.5 N/m and the gravity acceleration
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 197

a b

Fig. 2. (a) The initial condition for the 2-D rising bubble case; (b) The bubble shape and distribution of velocity with the particle number of
100 × 200 at t = 3 s.

a b

Fig. 3. (a) The time evolution of the center of mass; (b) The time evolution of the rising velocity.

is g = 0.98 m/s2 . The same flow field is discretized into three particle numbers: 25 × 50, 50 × 100 and 100 × 200.
The bubble shape and velocity distribution with the particle number of 100 × 200 at t = 3 s are shown in Fig. 2(b).
In order to compare the SPH results with the reference data [53], the center of mass of the bubble is defined as
follows:
Nbubble

yi
i
ybubble = (31)
Nbubble
where yi is the vertical coordinate of the gas particle i. The rising velocity of the bubble is defined as follows:
Nbubble

uyi
i
u bubble = (32)
Nbubble
where u yi is the vertical velocity of the gas particle i. The time evolution of the center of mass is shown in Fig. 3(a),
where the SPH results of all the three particle resolutions (i.e. 25 × 50, 50 × 100 and 100 × 200) agree well with
the reference data [53]. Fig. 3(b) shows the time history of the rising velocity. Though there are some oscillations
observed, the overall trend is in accordance with the reference data [53]. The convergence rate for the rising velocity
is calculated as log(ε32 /ε21 )/ log (2) = 1.23, where ε32 indicates the absolute error between the coarsest resolution
and the medium one, and similarly ε21 indicates the absolute error between the medium resolution and the finest one
(ε32 and ε21 are evaluated in L 2 norm).
198 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

Table 1
Parameters for the investigation on the effects of the solid wall boundary.
Particle spacing ∆x 0.0005 m
0.02 × 0.02 × 0.1 (a)
Computational domain 0.03 × 0.03 × 0.1 (b)
m
D×D×H 0.04 × 0.04 × 0.1 (c)
0.05 × 0.05 × 0.1 (d)
Bubble diameter d 0.01 m
Initial bubble position
(0, 0, 0.02) (a–d) m
(x0 , y0 , z 0 )
Liquid density ρl 1000 kg/m3
Bubble density ρg 100 kg/m3
Liquid viscosity ηl 0.1 kg/ (m s)
Bubble viscosity ηg 0.001 kg/ (m s)
Surface tension coefficient β 0.1 N/m

3. Numerical investigation

For 3-D simulations, the particle quantity is so enormous, which brings very expensive computational cost. In the
first part of this section, the effects of the solid wall boundary on the terminal rising state (including terminal rising
velocity and terminal bubble shape) of the freely rising bubble are tested to obtain a suitable physical scale for the
computational domain.
In the present multiphase SPH model, the time step is restricted by the density ratio at the multiphase interface.
The larger the density ratio is, the smaller the time step will be. In Annaland et al. [4], the rising gas bubble is
simulated with a density ratio of ρl /ρg = 100 in the VOF method, and Annaland et al. [4] state that this density
ratio is sufficiently high to mimic gas–liquid systems. In Hua et al. [9], with the Front Tracking Method, they get the
conclusion that the density ratio has negligible effects on final shape and terminal velocity of the rising bubble when it
is larger than 50. A similar conclusion is also drawn in Amaya-Bower and Lee [13] with Lattice Boltzmann Method.
The effects of the density ratio in the simulation of bubble dynamics using SPH method are not very clear. Therefore,
a numerical investigation on this is carried out in the second part.

3.1. The effects of the solid wall boundary

Firstly, we choose a case from Annaland et al. [4] to investigate the relation between the terminal rising velocity
and the scale of the computational domain. The investigation is characterized by the ratio D/d, where D is the width
of the computational domain and d is the diameter of the initial spherical bubble. The parameters for this test case are
shown in Table 1.
Time history of the rising velocity with different widths of the computational domain is shown in Fig. 4. One may
find that, after t = 0.05 s, the rising velocity enters a steady state, namely the terminal rising velocity. Regarding
the effects of D/d, one may find that, as D/d is increased from 2 to 5, the terminal rising velocity is increased but
the growth rate is decreased. The two curves of D/d = 4 and D/d = 5 almost coincide. Therefore, we may draw
a conclusion that, for the rising velocity, D/d = 4 is an acceptable value for the computational domain. The bubble
shapes at t = 0.2 s are shown in Fig. 5, from which one may find that, as D/d is increased larger than 3, the terminal
bubble shapes are similar. Note that, in all the numerical results of the present work, the colors on the bubble surface
denote the magnitude of the surface tension: the color red denotes the largest value and blue denotes the null. In
conclusion, considering both the terminal rising velocity and the terminal bubble shape, D/d = 4 will be used in the
following numerical examples.

3.2. The effects of the density ratio

In this part, the effects of the density ratio on the terminal rising state are numerically investigated with the
parameters shown in Table 2. Three density ratios, ρl /ρg = 10, ρl /ρg = 100 and ρl /ρg = 1000, are tested and the
results are compared with the numerical results of Front Tracking Method (FTM) in Hua et al. [7] and the experimental
results in Bhaga et al. [29].
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 199

Fig. 4. Time evolution of rising velocity with different widths of the computational domain.

Fig. 5. The terminal bubble shapes at t = 0.2 s with different widths of the computational domain; On the top shows the whole bubble and on the
bottom shows half section of the bubble; The colors on the bubble surface show the magnitude of the surface tension: the color red denotes the
largest value and blue denotes the null. (For interpretation of the references to color in this figure legend, the reader is referred to the web version
of this article.)

Table 2
Parameters for the investigation on the effects of the density ratio.
Particle spacing ∆x 0.0005 m
Computational domain
0.04 × 0.04 × 0.09 m
D×D×H
Bubble diameter d 0.01 m
Initial bubble position
(0, 0, 0.015) m
(x0 , y0 , z 0 )
Liquid density ρl 1000 kg/m3
100 (a)
Bubble density ρg 10 (b) kg/m3
1 (c)
Liquid viscosity ηl 2.15 × 10−2 kg/ (m s)
Bubble viscosity ηg 2.15 × 10−4 kg/ (m s)
Surface tension coefficient β 7.69 × 10−3 N/m

The bubble shapes of ρl /ρg = 1000 at different non-dimensional time instants are shown in Fig. 6, where the SPH
results agree well with the results of FTM [7]. This preliminarily verifies the effectiveness of the present multiphase
SPH model in the simulation of freely rising bubbles. In Fig. 7(a), the distribution of the direction of the surface tension
on the bubble surface is shown. One may find that, the directions of the surface tension are all perpendicular to the
bubble surface and at the edge of the bubble skirt, the magnitude of the surface tension is the largest due to the largest
curvature there. The distribution of the magnitude of the surface tension on the bubble surface is shown more clearly
in Fig. 7(b). Since the particle resolution is not very fine, the contour of the surface tension shows some oscillations,
but the overall distribution is reasonable, which makes the bubble shapes agree well with the reference results (see
Fig. 6). The velocity distribution around the bubble at different time instants is shown in Fig. 8, from which one may
observe that, as the bubble rising up, the wake flow of the bubble becomes stronger. Behind the bubble, an annular
200 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

Fig. 6. The SPH snapshots of the bubble shape compared with the results of FTM [7] at different time instants.

a b

Fig. 7. The distribution of surface tension on the bubble surface: (a) the direction of the force; (b) the magnitude of the force.

Fig. 8. The velocity distribution around the bubble at different time instants.

vortex forms after t (g/d)1/2 = 4. The flow wake may be the main reason for the interaction of multiple bubbles since
it creates an absorption effect behind the bubbles, which will be discussed in detail in Section 4.
Time history of the rising velocity with different density ratios is shown in Fig. 9, where most parts of the velocity
curves agree well with the results of FTM [7]. However, during t (g/d)1/2 = 0 to t (g/d)1/2 = 3, the velocity curves
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 201

Fig. 9. Time history of rising velocity with different density ratios.

Fig. 10. The terminal bubble shapes at t (g/d)1/2 = 10 for three density ratios.

of SPH show some oscillations while the velocity curve of FTM [7] is almost monotonous. It may be caused by the
definition of the bubble velocity as is shown in Eq. (32), where the arithmetic mean value of the vertical velocities of
all the gas particles is used. From Fig. 6, one may find that, at t (g/d)1/2 = 2, the bubble becomes a skirted shape (the
definition of bubble shapes refers to the bubble diagram in Grace [54]). At this stage, velocities of some gas particles
are downward, i.e. these velocities are negative in the summation of Eq. (32), and therefore, the velocity curve of SPH
shows a trough.
In Fig. 9, the time evolutions of the velocities of ρl /ρg = 100 and ρl /ρg = 1000 are practically identical but the
velocity curve of ρl /ρg = 10 diverges a little from them. Besides, the terminal rising velocities of ρl /ρg = 100 and
ρl /ρg = 1000 agree well with that predicted by FTM [7], but the terminal rising velocity of ρl /ρg = 10 is a little
lower. The discrepancy is within 10%. The reason is that, the total force in the vertical direction is affected by the
gravity of the bubble. The gravity of ρl /ρg = 10, ρl /ρg = 100 and ρl /ρg = 1000 is respectively 10%, 1% and 0.1%
of the buoyancy force which is identical for the three cases [13]. As the density ratio is increased, the effect of such a
dependence becomes smaller and smaller. Therefore, in SPH, ρl /ρg = 100 is large enough to simulate the dynamic
behaviors of rising gas bubbles [4].
From Fig. 10, one may find that, the terminal bubble shapes of the three density ratios are similar. This is because
the deformation of the bubble mainly depends on the value of the surface tension, which remains identical for all three
density ratios [13]. This conclusion also agrees with the one pointed out in Hua et al. [9] that the effect of density ratio
on the bubble rising velocity is more significant than on the terminal bubble shape. In conclusion, for the numerical
investigation of rising bubbles, if one focuses on the terminal bubble shape, the density ratio of ρl /ρg = 10 is enough;
while if one focuses on the dynamic behaviors of the whole rising process, ρl /ρg = 100 should be chosen. Please
note that, the present SPH model is capable for the simulation of the density ratio up to 1000, but it will need a long
computing time since the time step will be very small. In Section 4, in order to obtain a fast simulation, we will use a
decreased density ratio to mimic the gas–liquid system.
202 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

4. 3-D bubble rising cases

4.1. Single bubble rising test

Based on the conclusions of the numerical investigation in Section 3, the phenomena of a single bubble rising up in
different viscous fluids are simulated and validated. Before the detailed description of the SPH results, it is needed to
give three non-dimensional characteristic numbers which are usually used in the research of bubble dynamics. They
are Reynolds number Re, Bond number Bo (also known as Eötvös number in some literature) and Morton number
Mo defined as follows:
ρl g 1/2 d 3/2

 Re =
ηl




ρl gd 2


Bo = (33)

 β
4
 3
 Mo = gηl = Bo .



ρl β 3 Re4
Generally, experimental results are prescribed by another kind of Reynolds number Re∗ which is calculated by
replacing the g 1/2 d 1/2 in Re of Eq. (33) with the terminal rising velocity u ∞ . Re∗ is shown as follows:
ρl du ∞
Re∗ = . (34)
ηl
The bubble diagram of Grace [54] is characterized by Bo and Mo, which determines the terminal bubble shape
and the terminal Reynolds number Re∗ .
In this part, the simulations are conducted with a density ratio ρl /ρg = 10 since the terminal bubble shape is mainly
dependent on the surface tension and a decreased density ratio causes negligible effects [9,13]. Six cases are tested in
this part and the results are validated by the FTM results [7,9], the LBM results [13] and the experimental data [29].
The physical parameters of the test cases are similar to those in Table 2, except the density ratio, the viscous coefficient
and the surface tension coefficient. The density ratio is ρl /ρg = 10 and the liquid density is still 1000 kg/m3 . The
viscous ratio is ηl /ηg = 100. The viscous coefficient ηl and the surface tension coefficient β are calculated based on
the non-dimensional characteristic numbers Bo and Mo which are shown in the first column in Fig. 11.
From the terminal bubble shapes in Fig. 11, one may find the good agreements between SPH and other reference
results. However, some discrepancies still exist. In the results of LBM [13], a jet is formed in the case (a), while it is
not observed in the SPH simulation and the experimental data. For the 3-D simulation in Szewc et al. [36], a similar
jet happened when the sharpness force coefficient was set to be smaller than 0.5. In the present work, the sharpness
force coefficient is only set to be 0.08 in Eq. (26), which is much smaller than 0.5 while the terminal bubble shape
still agree well with the experimental data [29]. Therefore, one can see clearly the effectiveness of the multiphase
interface treatment in Section 2.4. For the cases from (b) to (e) in Fig. 11, the numerical results and the experimental
data all agree well, but for the case (f), both the SPH result and the LBM result give different bubble shapes compared
with the experimental data. It may be caused by the resolution which is not sufficiently fine, so the skirt of the bubble
cannot be well reproduced. From the terminal Reynolds number Re∗ in Fig. 11, one may find that most of the SPH
results are a little lower than those of the experiment [29]. According to the conclusion in Section 3.2, the discrepancy
can be resolved by improving the density ratio.

4.2. The co-axial and oblique coalescences

Starting from this part, the phenomena of bubbles coalescing in different directions will be simulated with the
multiphase SPH model. In these problems, the simulation of breaking and recombination of the multiphase interface is
very challenging for the traditional mesh-based methods, while it is relatively convenient for the mesh-free methods.
In this test case, the Bond number Bo and Morton number Mo are Bo = 16 and Mo = 2 × 10−4 , respectively.
According to the bubble diagram of Grace [54], the terminal Reynolds number Re∗ is 50 for a single rising bubble.
The experiment in Brereton et al. [30] gives the terminal Reynolds number Re∗ as 43. The VOF method in Annaland
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 203

Fig. 11. The terminal bubble shapes at t (g/d)1/2 = 10. The results of SPH are compared with the results of FTM [7,9], LBM [13] and the
experimental data [29].

Table 3
Parameters for the simulation of the co-axial and oblique coalescences.
Particle spacing ∆x 0.0005 m
Computational domain
0.04 × 0.04 × 0.08 m
D×D×H
Bubble diameter d 0.01 m
(x0 , y0 , z 0 ) = (0, 0, 0.025)

(co-axis)
Initial bubble position (x1 , y1 , z 1 ) = (0, 0, 0.015) m
(x0 , y0 , z 0 ) = (0, 0, 0.025)
(oblique)
(x1 , y1 , z 1 ) = (0.008, 0, 0.015)
Liquid density ρl 1000 kg/m3
Bubble density ρg 10 kg/m3
Liquid viscosity ηl 4.63 × 10−2 kg/ (m s)
Bubble viscosity ηg 4.63 × 10−4 kg/ (m s)
Surface tension coefficient β 0.0606 N/m

et al. [13] gives the terminal Reynolds number Re∗ as 40. Here the terminal Reynolds number Re∗ by SPH is about
43.19, which agrees well with the experimental data [30]. Through adjusting the initial relative positions of the two
spherical bubbles, the co-axial and oblique coalescences after the pair of bubbles rising up are simulated [4]. The
physical parameters are shown in Table 3.
In the SPH results, at t = 0.096 s, the trailing bubble touches the leading bubble. At this instant, we match the
snapshot of SPH with the fourth photograph in the experiment [30] since the phenomena are similar (see the top
204 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

Fig. 12. The SPH snapshots of the co-axial coalescence at some characteristic time instants (top); the experimental snapshots in Brereton et al. [30]
(middle); the velocity field around the bubbles (bottom).

and middle of Fig. 12). Since the time difference between the photographs in the experiment [30] is 0.03 s, we can
calculate the corresponding time instants of the other snapshots in SPH (see the top of Fig. 12). Finally, the whole
rising process at different time instants is compared between SPH method and the experiment data [30] in Fig. 12 and
a good agreement is achieved. One may find that at the first time instant, the SPH result seems quite different from the
experiment data but after that, the evolution is quite in accordance. This is because the initial bubble shape in the SPH
method is spherical, which is different from the initial condition of ellipsoid in the experiment [30]. However, due
to the pressure difference between the top and bottom of the bubble, the bubble deforms quickly into the ellipsoidal
shape and therefore the following evolution is quite in accordance. Thanks to the Lagrangian characteristic of SPH
method, it is easy to track each particle’s trajectory. The velocity field around the bubbles from rising up to coalescing
is shown on the bottom of Fig. 12, from which one may find that, the leading bubble deforms into the ellipsoidal shape
quickly due to the considerable resistance force from the above liquid, while the trailing bubble deforms into a ‘bullet’
shape due to the absorption effect from the flow wake of the leading bubble. At t = 0.096 s, the two bubbles touch
each other and merge into one bigger bubble quickly.
Fig. 13 shows the oblique coalescence of the bubble pair, from which one may see more clearly the ab-
sorption effect of the leading bubble. Since vertically below the leading bubble, the absorption effect is the
strongest while it is weakened in other directions. Therefore, the oblique coalescence ends later than the co-axial
coalescence.
Time histories of the rising velocity of the single bubble, the leading bubble and the trailing bubble are shown in
Fig. 14(a) and Fig. 14(b) for co-axis and oblique coalescence, respectively. Each bubble in the bubble pair rises more
quickly than the single bubble. Especially for the trailing bubble, due to the absorption effect of the leading bubble,
its rising velocity is always increasing. Just after the coalescence happens, at t = 0.096 s for co-axial coalescence and
at t = 0.15 s for oblique coalescence, the rising velocity reaches the largest. The reason is after the two bubbles touch
each other, a gas passage forms joining the two bubbles together. The curvature of the surface of the trailing bubble
is much larger than that of the leading bubble. Therefore, the gas particles are pushed upward quickly by the surface
tension. As time goes on, the velocities of both the two bubbles decrease a lot. The reason is after the coalescence
ends, the merged bubble becomes a skirted shape and therefore the resistance force increases a lot. From the analysis
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 205

Fig. 13. The SPH snapshots of the oblique coalescence at some characteristic time instants (top); the experimental snapshots in Brereton et al. [30]
(middle); the velocity field around the bubbles (bottom).

a b

Fig. 14. Time history of the rising velocity of the bubbles in the co-axial coalescence (a) and the oblique coalescence (b).

above, one may find that, the wake flow and the surface tension play an important role in the phenomena of bubbles
chasing and coalescing.

4.3. The horizontal coalescence

The horizontal coalescence is simulated in this part. The SPH results are compared with the experimental data in
Duineveld et al. [31]. When two spherical bubbles rise side by side, the flow velocity between the two bubbles is faster
than that on the other two sides, therefore, the pressure in the middle becomes smaller according to Bernoulli’s law.
Finally, the two bubbles approach to each other and coalesce. Note that for two ellipsoidal bubbles, the repelling effect
of the flow wakes behind the bubbles are much stronger, which may cause them depart from each other (see e.g. Yu
et al. [11]). The horizontal repelling of two ellipsoidal bubbles is out of the scope of this part.
The Bond number Bo and Morton number Mo in this test case are Bo = 0.436 and Mo = 2.5 × 10−11 ,
respectively. The data used for the simulation here is shown in Table 4. Since the Bond number is extremely small
here, a background pressure pb = 0.5 p0 is added in the equation of state to make the particles distribute uniformly
206 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

Table 4
Parameters for the simulation of the horizontal coalescence.
Particle spacing ∆x 0.0005 m
Computational domain
0.05 × 0.05 × 0.07 m
D×D×H
Bubble diameter d 0.01 m
(x0 , y0 , z 0 ) = (0.006, 0, 0.02)

Initial bubble position m
(x1 , y1 , z 1 ) = (−0.006, 0, 0.02)
Liquid density ρl 1000 kg/m3
Bubble density ρg 10 kg/m3
Liquid viscosity ηl 1.31 × 10−2 kg/ (m s)
Bubble viscosity ηg 1.31 × 10−4 kg/ (m s)
Surface tension coefficient β 2.25 N/m
Background pressure pb 5880 Pa

Fig. 15. The SPH snapshots of the horizontal coalescence compared with the experimental data [31] at some characteristic time instants.

and keep the multiphase interface stable. The snapshots of the horizontal bubble coalescence at some characteristic
time instants are shown in Fig. 15.
At t (g/R)1/2 = 0.140 (R is the radius of the bubble), the two bubbles touch each other and a gas passage forms
between them. Then the surface tension expands the gas passage and both ends of the coalesced bubbles approach to
the center. Due to the effects of buoyancy, the approaching directions are obliquely upward. At t (g/R)1/2 = 0.358,
the bubble becomes a ‘fusiform’ shape. The curvature on the bottom end of the merged bubble is larger than that on the
top end. Therefore the surface tension pushes the merged bubble to rise up quickly. At t (g/R)1/2 = 0.770, the bubble
becomes a ‘bullet’ shape. As the resistance force increases, the bubble becomes ellipsoidal at t (g/R)1/2 = 1.131. In
Chen et al. [55], 2-D bubble coalescence was well simulated within the framework of MPS which is also a mesh-free
method similar to SPH, however in two dimensions, the bubble is more like an infinite long gas tube rather than an
ellipsoid. Under this consideration, the 3-D results agree better with the experimental data [31], as is shown in Fig. 15.
In conclusion, with the multiphase SPH model, the horizontal coalescence of two spherical bubbles in the condition
of low Bond number is simulated. Background pressure pb is added in the equation of state, which plays an important
role in keeping the multiphase interface smooth and making the simulation stable.

5. Conclusions

In this paper, a 3-D multiphase SPH model is applied to simulate the dynamic behaviors of rising bubbles in viscous
fluids. Several conclusions are drawn as follows:
(i) The discretized governing equations are derived by the principle of virtual work, which proves the property of
energy conservation. A suitable viscous formula for multiphase flow is applied to achieve more accurate results. The
surface tension is calculated accurately when the density ratio is large. The interface penetration is well avoided by
applying the interface sharpness force. Besides, it is shown that the background pressure plays an important role in
keeping the multiphase interface stable when the Bond number is relatively small.
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 207

(ii) In order to resolve the problem of expensive computational cost in 3-D bubble rising simulations, a numerical
investigation is carried out for the analyses of the scale of the computational domain and the effects of the density
ratio. Regarding the scale of the computational domain, D/d = 4 is recommended. The density ratio ρl /ρg causes
negligible effects on the terminal bubble shape, but it has large effects on the dynamic behavior during the rising
process. Therefore, for the prediction of the terminal bubble shape which mainly depends on the surface tension,
ρl /ρg = 10 is acceptable. If one concentrates on the whole rising process, ρl /ρg = 100 is recommended. However it
is worth mentioning that, the present SPH model is capable for the simulation of the density ratio up to 1000, but the
time step will become very small.
(iii) A single bubble rising and multiple bubbles coalescing in different directions are simulated and the SPH results
agree well with the experimental data in the literature. The Lagrangian characteristic of SPH makes it very suitable
for the simulation of bubble coalescence, where the multiphase interfaces are broken and recombined naturally. In the
future, more complicated 3-D bubbly flow may be simulated by using a more efficient parallel computing program,
e.g. the GPU program.

Acknowledgments
The authors highly appreciate the valuable comments and constructive suggestions from the reviewers. This work is
supported by the Excellent Young Scientists Fund of China under Grant No. 51222904, the National Natural Science
Foundation of China under Grant No. 51379039, the National Program for Support of Top-notch Young Professionals
and CNR-INSEAN within the Project PANdA: PArticle methods for Naval Applications, protocol number N. 3263,
21 October 2014. Great appreciation is also given to Dr. Huang Xiao, who helped us in correcting and improving the
English text.

Appendix A. The derivation of the discretized form of divergence of the virtual displacement
The time derivative on both sides of Eq. (10) is as follows [56]:
Dρi  DWi j
= mi . (A.1)
Dt j
Dt

Transforming the time derivative on the right-hand side of Eq. (A.1), one gets

∂ Wi j Dri j ∂ Wi j ui − u j · ri − r j
   
DWi j
= = . (A.2)
Dt ∂ri j Dt ∂ri j ri j
We have
ri − r j ∂ Wi j
 
∇i Wi j = . (A.3)
ri j ∂ri j
Substituting Eq. (A.3) into Eq. (A.2), one gets
DWi j
= ui − u j · ∇i Wi j .
 
(A.4)
Dt
Substituting Eq. (A.4) into Eq. (A.1), one obtains
Dρi 
ui − u j · ∇i Wi j .

= mi (A.5)
Dt j

Substituting Eq. (1) into Eq. (A.5), one gets



ui − u j · ∇i Wi j .

⟨∇ · u⟩i = −Vi (A.6)
j

Replace u of Eq. (A.6) with the virtual displacement δw, one gets

⟨∇ · (δw)⟩i = −Vi δwi − δw j · ∇i Wi j .

(A.7)
j
208 A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209

Appendix B. Supplementary data

Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.cma.2015.05.014.

References

[1] N. Grenier, D. Le Touze, A. Colagrossi, M. Antuono, G. Colicchio, Viscous bubbly flows simulation with an interface SPH model, Ocean
Eng. 69 (2013) 88–102.
[2] T. Frising, C. Noik, C. Dalmazzone, The liquid/liquid sedimentation process: from droplet coalescence to technologically enhanced water/oil
emulsion gravity separators: a review, J. Dispersion Sci. Technol. 27 (2006) 1035–1057.
[3] R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles, Academic Press, 1978.
[4] M.V. Annaland, N.G. Deen, J.A.M. Kuipers, Numerical simulation of gas bubbles behaviour using a three-dimensional volume of fluid
method, Chem. Eng. Sci. 60 (2005) 2999–3011.
[5] 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.
[6] R. Croce, M. Griebel, M.A. Schweitzer, Numerical simulation of bubble and droplet deformation by a level set approach with surface tension
in three dimensions, Internat. J. Numer. Methods Fluids 62 (2010) 963–993.
[7] J. Hua, J.F. Stene, P. Lin, Numerical simulation of 3D bubbles rising in viscous liquids using a front tracking method, J. Comput. Phys. 227
(2008) 3358–3382.
[8] M.V. Annaland, W. Dijkhuizen, N.G. Deen, J.A.M. Kuipers, Numerical simulation of behavior of gas bubbles using a 3-D front-tracking
method, AIChE J. 52 (2006) 99–110.
[9] J. Hua, J. Lou, Numerical simulation of bubble rising in viscous liquid, J. Comput. Phys. 222 (2007) 769–795.
[10] Z. Yu, L.S. Fan, An interaction potential based lattice Boltzmann method with adaptive mesh refinement (AMR) for two-phase flow simulation,
J. Comput. Phys. 228 (2009) 6456–6478.
[11] Z. Yu, H. Yang, L.-S. Fan, Numerical simulation of bubble interactions using an adaptive lattice Boltzmann method, Chem. Eng. Sci. 66
(2011) 3441–3451.
[12] Z. Yu, H. Yang, L.S. Fan, Numerical simulation of bubble interactions using an adaptive lattice Boltzmann method, Chem. Eng. Sci. 66 (2011)
3441–3451.
[13] L. Amaya-Bower, T. Lee, Single bubble rising dynamics for moderate Reynolds number using Lattice Boltzmann Method, Comput. & Fluids
39 (2010) 1191–1207.
[14] M. Liu, G. Liu, Smoothed particle hydrodynamics (SPH): an overview and recent developments, Arch. Comput. Methods Eng. 17 (2010)
25–76.
[15] G.R. Liu, M. Liu, Smoothed Particle Hydrodynamics: A Meshfree Particle Method, World Scientific, 2003.
[16] R.A. Gingold, J.J. Monaghan, Smoothed particle hydrodynamics-theory and application to non-spherical stars, Mon. Not. R. Astron. Soc. 181
(1977) 375–389.
[17] L.B. Lucy, A numerical approach to the testing of the fission hypothesis, Astron. J. 82 (1977) 1013–1024.
[18] J.J. Monaghan, Simulating free surface flows with SPH, J. Comput. Phys. 110 (1994) 399–406.
[19] S. Marrone, M. Antuono, A. Colagrossi, G. Colicchio, D. Le Touzé, G. Graziani, δ-SPH model for simulating violent impact flows, Comput.
Methods Appl. Mech. Engrg. 200 (2011) 1526–1542.
[20] D.A. Barcarolo, Improvement of the precision and the efficiency of the SPH method: theoretical and numerical study (Ph.D. thesis of Ecole
Centrale de Nantes), 2013.
[21] A. Colagrossi, M. Landrini, Numerical simulation of interfacial flows by smoothed particle hydrodynamics, J. Comput. Phys. 191 (2003)
448–475.
[22] M. Antuono, A. Colagrossi, S. Marrone, C. Lugni, Propagation of gravity waves through an SPH scheme with numerical diffusive terms,
Comput. Phys. Comm. 182 (2011) 866–877.
[23] B. Bouscasse, A. Colagrossi, S. Marrone, M. Antuono, Nonlinear water wave interaction with floating bodies in SPH, J. Fluids Struct. 42
(2013) 112–129.
[24] X.Y. Cao, F.R. Ming, A.M. Zhang, Sloshing in a rectangular tank based on SPH simulation, Appl. Ocean Res. 47 (2014) 241–254.
[25] X. Hu, N. Adams, A multi-phase SPH method for macroscopic and mesoscopic flows, J. Comput. Phys. 213 (2006) 844–861.
[26] X. Hu, N. Adams, An incompressible multi-phase SPH method, J. Comput. Phys. 227 (2007) 264–278.
[27] N. Grenier, M. Antuono, A. Colagrossi, D. Le Touzé, B. Alessandrini, An Hamiltonian interface SPH formulation for multi-fluid and free
surface flows, J. Comput. Phys. 228 (2009) 8380–8393.
[28] J.J. Monaghan, A. Rafiee, A simple SPH algorithm for multi-fluid flow with high density ratios, Internat. J. Numer. Methods Fluids 71 (2013)
537–561.
[29] D. Bhaga, M. Weber, Bubbles in viscous liquids: shapes, wakes and velocities, J. Fluid Mech. 105 (1981) 61–85.
[30] G. Brereton, D. Korotney, Coaxial and oblique coalescence of two rising bubbles, in: Proceedings of AMD-Vol0, vol. 119, 1991.
[31] P. Duineveld, Bouncing and coalescence of bubble pairs rising at high Reynolds number in pure water or aqueous surfactant solutions,
in: Fascination of Fluid Dynamics, Springer, 1998, pp. 409–439.
[32] A.K. Das, P.K. Das, Bubble evolution and necking at a submerged orifice for the complete range of orifice tilt, AIChE J. 59 (2013) 630–642.
[33] A.K. Das, P.K. Das, Bubble evolution through a submerged orifice using smoothed particle hydrodynamics: effect of different thermophysical
properties, Ind. Eng. Chem. Res. 48 (2009) 8726–8735.
[34] A.K. Das, P.K. Das, Bubble evolution through submerged orifice using smoothed particle hydrodynamics: basic formulation and model
validation, Chem. Eng. Sci. 64 (2009) 2281–2290.
A. Zhang et al. / Comput. Methods Appl. Mech. Engrg. 294 (2015) 189–209 209

[35] A.K. Das, P.K. Das, Incorporation of diffuse interface in smoothed particle hydrodynamics: Implementation of the scheme and case studies,
Internat. J. Numer. Methods Fluids 67 (2011) 671–699.
[36] K. Szewc, J. Pozorski, J.P. Minier, Simulations of single bubbles rising through viscous liquids using smoothed particle hydrodynamics, Int.
J. Multiph. Flow 50 (2013) 98–105.
[37] K. Szewc, J. Pozorski, J.-P. Minier, On the issue of interface spurious fragmentations in multiphase SPH, in: Proceedings of the 9th
International SPHERIC Workshop, 03–05 June 2014, Paris, France.
[38] S. Adami, X. Hu, N. Adams, A new surface-tension formulation for multi-phase SPH using a reproducing divergence approximation,
J. Comput. Phys. 229 (2010) 5011–5021.
[39] J.P. Morris, P.J. Fox, Y. Zhu, Modeling low Reynolds number incompressible flows using SPH, J. Comput. Phys. 136 (1997) 214–226.
[40] J. Brackbill, D.B. Kothe, C. Zemach, A continuum method for modeling surface tension, J. Comput. Phys. 100 (1992) 335–354.
[41] S. Marrone, A. Colagrossi, M. Antuono, G. Colicchio, G. Graziani, An accurate SPH modeling of viscous flows around bodies at low and
moderate Reynolds numbers, J. Comput. Phys. 245 (2013) 456–475.
[42] A. Colagrossi, B. Bouscasse, M. Antuono, S. Marrone, Particle packing algorithm for SPH schemes, Comput. Phys. Commun. 183 (2012)
1641–1653.
[43] D. Molteni, A. Colagrossi, A simple procedure to improve the pressure evaluation in hydrodynamic context using the SPH, Comput. Phys.
Commun. 180 (2009) 861–872.
[44] J. Hongbin, D. Xin, On criterions for smoothed particle hydrodynamics kernels in stable field, J. Comput. Phys. 202 (2005) 699–709.
[45] A. Colagrossi, M. Antuono, D. Le Touzé, Theoretical considerations on the free-surface role in the smoothed-particle-hydrodynamics model,
Phys. Rev. E 79 (2009) 056701.
[46] A. Colagrossi, M. Antuono, A. Souto-Iglesias, D. Le Touzé, Theoretical analysis and numerical verification of the consistency of viscous
smoothed-particle-hydrodynamics formulations in simulating free-surface flows, Phys. Rev. E 84 (2011) 026705.
[47] J. Monaghan, R. Gingold, Shock simulation by the particle method SPH, J. Comput. Phys. 52 (1983) 374–389.
[48] J.P. Morris, Simulating surface tension with smoothed particle hydrodynamics, Internat. J. Numer. Methods Fluids 33 (2000) 333–353.
[49] T. Breinlinger, P. Polfer, A. Hashibon, T. Kraft, Surface tension and wetting effects with smoothed particle hydrodynamics, J. Comput. Phys.
243 (2013) 14–27.
[50] A. Zainali, N. Tofighi, M. Shadloo, M. Yildiz, Numerical investigation of Newtonian and non-Newtonian multiphase flows using ISPH
method, Comput. Methods Appl. Mech. Engrg. 254 (2013) 99–113.
[51] S. Adami, X. Hu, N. Adams, A generalized wall boundary condition for smoothed particle hydrodynamics, J. Comput. Phys. 231 (21) (2012)
7057–7075.
[52] Z. Chen, Z. Zong, M. Liu, L. Zou, H. Li, C. Shu, An SPH model for multiphase flows with complex interfaces and large density differences,
J. Comput. Phys. 283 (2015) 169–188.
[53] S. Hysing, S. Turek, D. Kuzmin, N. Parolini, E. Burman, S. Ganesan, L. Tobiska, Quantitative benchmark computations of two-dimensional
bubble dynamics, Internat. J. Numer. Methods Fluids 60 (2009) 1259–1288.
[54] J. Grace, Shapes and velocities of bubbles rising in infinite liquids, Trans. Inst. Chem. Eng. 51 (1973) 116–120.
[55] R. Chen, W. Tian, G. Su, S. Qiu, Y. Ishiwatari, Y. Oka, Numerical investigation on coalescence of bubble pairs rising in a stagnant liquid,
Chem. Eng. Sci. 66 (2011) 5055–5063.
[56] S. Marrone, Enhanced SPH modeling of free-surface flows with large deformations (Ph.D. thesis), Università di Roma Sapienza, 2012.

You might also like