Smoothed particle hydrodynamics

(SPH) for complex fluid flows: Recent

developments in methodology and
applications
Cite as: Phys. Fluids 31, 011301 (2019); https://doi.org/10.1063/1.5068697
Submitted: 18 October 2018 . Accepted: 04 December 2018 . Published Online: 10 January 2019

Ting Ye (叶挺), Dingyi Pan (潘定一) , Can Huang (黄灿) , and Moubin Liu (刘谋斌)

COLLECTIONS

This paper was selected as an Editor’s Pick

Phys. Fluids 31, 011301 (2019); https://doi.org/10.1063/1.5068697 31, 011301

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

Smoothed particle hydrodynamics (SPH) for

complex fluid flows: Recent developments in
methodology and applications
Cite as: Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697
Submitted: 18 October 2018 • Accepted: 4 December 2018 •
Published Online: 10 January 2019

Ting Ye (叶挺),1,a) Dingyi Pan (潘定一),2,a) Can Huang (黄灿),3,4,5,a) and Moubin Liu (刘谋斌)3,4,5,b)

AFFILIATIONS
1
Department of Computational Mathematics, Jilin University, Jilin Province 130012, China
2
Department of Engineering Mechanics, Zhejiang University, Hangzhou 310027, China
3
BIC-ESAT, College of Engineering, Peking University, Beijing 100871, China
4
Institute of Ocean Research, Peking University, Beijing 100871, China
5
State Key Laboratory for Turbulence and Complex Systems, Peking University, Beijing 100871, China

a)
T. Ye, D. Pan, and C. Huang contributed equally to this work.
b)
Author to whom correspondence should be addressed: [email protected]. Tel.: +86-10-62766982.

ABSTRACT
Computer modeling of complex fluid flows usually presents great challenges for conventional grid-based numerical methods.
Smoothed particle hydrodynamics (SPH) is a meshfree Lagrangian particle method and has special advantages in modeling com-
plex fluid flows, especially those with large fluid deformations, fluid-structure interactions, and multi-scale physics. In this paper,
we review the recent developments of SPH in methodology and applications for modeling complex fluid flows. Specifically, in
methodology, some important issues including modified SPH particle approximation schemes for improving discretization accu-
racy, different particle regularization techniques, and various boundary treatment algorithms for solid boundary, free surface,
or multiphase interface are described. More importantly, the SPH method with ideas from the dissipative particle dynamics
for complex fluids in macro- or meso-scales is discussed. In applications, different complex fluid flows, including biological
flows, microfluidics and droplet dynamics, non-Newtonian fluid flows, free surface flows, multiphase flows, and flows with
fluid-structure interaction, are reviewed. Some concluding remarks in SPH modeling of complex fluid flows are provided.
Published under license by AIP Publishing. https://doi.org/10.1063/1.5068697

I. INTRODUCTION rheological responses of the cells are usually investigated for

the practical purpose of cell classification, separation, and dis-
Complex fluid flows widely exist in nature, and the study
ease diagnosis. The deformation of the cell and its interac-
of complex fluid flows is of great importance in engineering
tion with ambient fluids may happen at different scales with
and sciences. For example, in ocean engineering and coastal
evolving rheology.1
hydrodynamics, viscous incompressible flows including wave
dynamics (wave generation, wave breaking, and wave inter- Computational fluid dynamics (CFD) is a powerful tool
action with structures), dam breaking, water filling and water that helps us to understand and predict flow phenomena, and
discharge (to and from a water tank or reservoir), entry and grid- or mesh-based numerical methods such as the finite dif-
exit of water, and liquid sloshing can be frequently observed. ference methods (FDM), the finite volume methods (FVM), and
These flow problems usually involve changing and breakup of the finite element methods (FEM) have been widely used for
free surfaces, strong turbulence, vortices, and violent fluid- domain discretization in most CFD applications during the last
solid interactions. In bioengineering, a typical complex fluid century. In these grid-based numerical methods, individual
flow is cells passing a micro-channel, where the dynamic and grid points (or nodes) are connected together in a pre-defined

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-1

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

manner by a topological map, which is termed as a mesh (or excellent conservation properties in aspects of energy, lin-
grid). The meshing results in, conventionally, elements in FEM, ear momentum, angular momentum, mass, and entropy if no
cells in FVM, and grids in FDM. A mesh or grid system con- artificial viscosity operates.12
sisting of nodes, cells, or elements must be defined to pro-
SPH has some special advantages over the traditional
vide the relationship between the nodes before the approx-
grid-based numerical methods in modeling complex fluid
imation process for the differential equations or partial dif-
flows. By properly deploying particles at specific positions
ferential equations (PDEs). Based on a properly pre-defined
at the initial stage, free surfaces and moving boundaries can
mesh system, governing equations can be converted to a
all be traced naturally in the process of simulation regard-
set of algebraic equations with nodal unknowns for the field
less of the complexity of the movement of the particles,
variables.
while this has been challenging the Eulerian grid-based meth-
Due to the pre-defined mesh, conventional grid-based ods continuously. SPH naturally satisfies the interface track-
methods suffer from special difficulties which limit their ing in simulations of multiphase flows13 —the motion of the
applications in modeling complex fluid flows with free sur- fluid is represented by the motion of particles, and the fluid
faces, deformable boundaries, moving interfaces, extremely surfaces or fluid-fluid interfaces move with particles repre-
large deformations, and fluid-structure interactions. For senting their phase defined at the initial stage. Moreover,
Lagrangian grid-based methods such as FEM, a grid is there are no advection errors in SPH, and the scheme is fully
attached on, moves and deforms with the moving objects. It is Galilean invariant, unlike grid-based methods. The readers
therefore easy to obtain time history of movement and is con- may refer to some excellent reviews including the studies of
venient to treat or track moving features such as free surfaces Monaghan,14,15 Liu and Liu,8 Violeau and Rogers,16 Gotoh and
and deformable interfaces. However, it is very difficult to treat Khayyer,17 Zhang et al.,18 Shadloo et al.,19 and Wang et al.20
large deformations due to possible mesh entanglement. Spe-
Another attractive feature is that, as a particle-based
cial techniques like mesh rezoning are usually required, and
method, SPH is very much similar to the classic molecular
these techniques are often tedious and time-consuming, and
dynamics21 (MD) for micro-scale applications and dissipa-
may even introduce additional inaccuracy into the solution.2,3
tive particle dynamics1,22,23 (DPD) for meso-scale applications.
In contrast, for Eulerian grid-based methods such as FDM and These three particle methods share the same features with
FVM, a computational grid system is fixed on the computa- purely meshfree and Lagrangian nature. For example, in MD,
tional domain such that there is no problem to deal with large DPD, and SPH, individual particles are used to represent the
deformations. But it is quite difficult to treat or track mov- state of a system and the evolution of the particles is gov-
ing features and special algorithms such as volume-of-fluid4 erned simply by the Newton’s second law through using a
(VOF) and level-set5 (LS) are usually necessary. weight function, i.e., potential function in MD, weight func-
During the last decades, an increasing interest has been tion in DPD, or smoothing function in SPH. A particle in MD,
focused on the development of the next generation computa- DPD, and SPH acts as both a material point and as an approx-
tional methods, meshfree particle methods, such as the Mov- imation point, that is, the particle is regarded as a single atom
ing Particle Semi-implicit (MPS) method6,7 and the Smoothed or molecule in MD in atomistic scales, a small cluster of atoms
Particle Hydrodynamics (SPH) method.8,9 In the MPS method, or molecules in DPD in meso-scales, and a very small region
the governing equations are transformed into interactions in SPH in macro-scales. MD, DPD, and SPH are manifestly
among moving particles, and a semi-implicit algorithm is used Galilean invariant (unlike some lattice Boltzmann models) such
to model incompressible flows via solving the pressure Pois- that they can be combined together using some kind of hand-
shaking algorithms through modeling complex fluid flows at
son equation (PPE), while the other terms are explicitly calcu-
different scales. Some features of MD or DPD can also be
lated. So far the MPS method has been widely applied to mod-
integrated into SPH to allow its extension to meso- or even
eling complex fluid flows especially free surface flows with
micro-scale applications. For example, Español described a
fluid-structure interactions. Smoothed Particle Hydrodynam-
fluid particle dynamics model (FPM),24 which is the synthesis
ics (SPH) is a truly Lagrangian particle method for modeling
of DPD and SPH, and Español and Revenga25 combined fea-
complex fluid flows. SPH was originally developed for astro-
tures from DPD and SPH to develop the smoothed dissipative
physical problems by Lucy10 and Gingold and Monaghan11
particle dynamics (SDPD) model in which the Navier-Stokes
approximately 40 years ago, as the collective movement of
(N-S) equation governing the fluid flow is discretized using
those particles is similar to the movement of a liquid or gas
SPH approximations while thermal fluctuations are included
flow. Since its invention, the SPH method has been extended
in a consistent way.
to a vast range of problems in both fluid and solid mechanics
because of the superior ability to incorporate complex physics In this paper, the latest development of SPH in method-
into the SPH formulations. In SPH, field variables (such as den- ology and applications for modeling complex fluid flows is
sity, velocity, and acceleration) can be obtained by approx- reviewed. The paper is organized as follows: In Sec. II, the basic
imating the governing equations on a set of particles. The concept of SPH, some modified SPH particle approximation
connectivity between particles is not generated as part of the schemes and SPH formulation for incompressible fluid flows
computation and can evolve with time. Therefore, SPH allows are described. In Secs. III and IV, different particle regular-
a straightforward handling of very large deformation and has ization techniques and various boundary treatment algorithms

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-2

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

for solid boundary, free surface, or material interface are dis- be determined from the values of the function and the deriva-
cussed, respectively. In Sec. V, the SPH method with ideas tives of the smoothing function, rather than from the deriva-
from DPD for complex fluids in micro- or meso-scales is given. tives of the function itself. It is pointed out by Liu et al. that the
In Sec. VI, different complex fluid flows, including biologi- SPH kernel approximations for a field function and its deriva-
cal flows, microfluidics and droplet dynamics, non-Newtonian tive are of 2nd order accuracy for the interior domain.14,15 In
fluid flows, free surface and multiphase flows, and flows with contrast, for boundary or interface areas, the above SPH ker-
fluid-structure interaction, are reviewed. This paper ends in nel approximations are less than 1st order accuracy due to the
Sec. VII with some remarks in SPH modeling of complex fluid truncated approximation domain, and this is usually referred
flows. to as boundary deficiency14,15 of the conventional SPH.
The SPH particle approximation converts the continuous
II. METHODOLOGY form of the kernel approximation of a field function and its
A. Discretization schemes derivatives to the discrete summations over a set of particles.
The use of particle summations to approximate the integral
1. Basic concept of conventional SPH is, in fact, a key approximation that makes the SPH method
In the SPH method, the governing partial differential simple without using a background mesh for numerical inte-
equations are approximated in two steps, the kernel approx- gration, and it is also the key factor influencing the solution
imation and the particle approximation. The kernel approxi- accuracy of the SPH method. When the particles are arbitrarily
mation of a function f at a certain point or particle i can be distributed and change at each time step, an accurate anal-
obtained by multiplying the function with a smoothing func- ysis of the SPH particle approximation is very difficult. Note
tion W and then integrating over the computational domain as that the 2nd order accuracy mentioned above is valid just for
follows:3  regular particle distributions. This is an ideal case because
fi  f(r)Wi (r)dr, (1) SPH particles are often irregularly distributed during simula-
tions even when the initial particle distribution is regular. For
where r is the position vector with x, y, z as its three com- boundary or interface areas, or irregular particle distributions,
ponents in a 3D space, dr is the volume element of the mate- the SPH particle approximations can be less than the 1st order
rial, and W i (r) = W(r − ri , h) is the smoothing function whose accuracy because it is not able to reproduce a linear function
influence is limited to a support domain characterized by the or even a constant and this is denoted as particle inconsis-
smoothing length h multiplying a possible scalar factor κ. tency.14,15 This is why the conventional SPH method was often
regarded as a low order accuracy method, and this leads to the
The corresponding particle approximation of f in the
development of various modified or improved versions of SPH
original SPH can be written as
particle approximation schemes.
N
X mj
fi  fj Wij , (2) 2. Early improved SPH approximation schemes
ρj
j=1
One early improved approach14,26 is based on the anti-
where W ij = W(rj − ri , h). N is the total number of particles symmetric property of the derivative of a smoothing function
within the support domain of the concerned particle i. mj as
and ρj are the mass and density of a neighboring particle j, N
X mj
respectively. Wi,α = 0. (5)
ρj
j=1
The kernel approximation of a spatial derivative fi can
be obtained by substituting the function into Eq. (1) with Therefore, combining with Eq. (4), when approximating the
the corresponding derivative, using integration by parts and derivative of a function f, the particle approximation can be
neglecting the boundary terms as follows: rewritten as
 N
X mj  
fi,α  f(r)Wi,α dr. (3) fi,α = fj − fi Wi,α (6)
ρj
j=1
The corresponding particle approximation of the spatial or
derivative is then written as
N
X mj  
N
mj fi,α = fj + fi Wi,α . (7)
ρj
X
fi,α  fj Wi,α , (4) j=1
ρj
j=1
Equations (6) and (7) actually use the particle inconsistency in
where W i, α = ∂W i (r)/∂rα is the spatial derivative of the parti- approximating the derivative of the smoothing function to off-
cle i in the direction α and α is the dimension index repeated set or balance the particle inconsistency in approximating the
from 1 to the number of dimensions dim . derivative of the concerned function f.
It should be noted that the SPH kernel approximation of Randles and Libersky26 derived a normalization formula-
the derivative of a field function allows the spatial gradient to tion for the density approximation as

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-3

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

N
P
mj Wij is a good improvement over the conventional SPH method
j=1 which may not have 1st order accuracy28,29 for problems with
ρi = (8) irregular particle distributions.
N mj
ρj Wij
P
j=1
4. Kernel gradient correction (KGC)
and a normalization for the divergence of the stress tensor Another popular approach to improve the accuracy of
σ as SPH is the kernel gradient correction technique.30 It is known
N mj
(σj − σi )∇i Wij that the approximation accuracy of the derivatives is closely
P
j=1 ρj related to the calculation of the gradient of the smoothing
(∇ · σ)i = . (9)
N mj function (or kernel gradient). It is possible to use a correc-
(rj − ri )∇i Wij
P
j=1 ρj tive kernel gradient rather than the conventional kernel gradi-
ent to obtain higher approximation accuracy. For example, in
Note that Eqs. (8) and (9) use the same way to improve a two-dimensional space, based on the Taylor series expan-
the accuracy of the approximations in view of particle sion on the SPH approximation of a function, it is possible
inconsistency. to restore the accuracy for general cases with the following
3. Corrective smoothed particle method (CSPM) correction on the kernel gradient:
Based on Taylor series expansion on the SPH approxima-
∇new
i Wij = L(ri )∇i Wij , (12)
tion of a function, Chen and Beraun27 developed a corrective −1
smoothed particle method. In one-dimensional space, the par- X mj xji Wi,x yji Wi,x +
ticle approximation for function f(r) at particle i can be written L(ri ) = *. * +/ , (13)
ρj , xji Wi,y yji Wi,y -
j
as , -
N mj
fj Wij
P
where xji = xj − xi , yji = yj − yi . It is noted that there is no
j=1 ρj
fi = . (10) need to significantly change the structure of SPH computer
N mj
P
Wij programs and procedure of SPH simulations because only the
j=1 ρj kernel gradient is corrected in the kernel gradient correction.
Similarly in CSPM, a corrective particle approximation for It is therefore convenient to implement SPH equations, on the
the first derivative fi,α can be written as (in one-dimensional basis of the conventional SPH.
space)
N mj 5. Finite particle method (FPM)
( fj − fi )Wi,x
P
j=1 ρj A further improvement on CSPM is the finite particle
fi,α = . (11) method by Liu et al.,28,29 which can also be derived by simul-
N mj
(xj − xi )Wi,x
P
taneously applying the Taylor series expansion on the SPH
j=1 ρj
approximation of a function and its derivatives while retain-
It is found that the CSPM particle approximations for a field ing derivatives with necessary orders. For example, if only
function and its derivative are of 1st order accuracy (or retaining the first order derivative in the Taylor series expan-
0th order consistency) for problems with irregular particle sion, the FPM approximations for a field function f and its first
distributions (including interface and boundary areas). This derivative at particle i can be written as

N mj N mj  −1  P N mj
(rjα − riα )Wij  
 P 
Wij fj Wij 
P
j=1 ρj ρ ρ
" # 
fi j=1 j   j=1 j
  
=  N m N mj N mj
 , (14)
fi,α  P j W (rjα − riα )Wij,β  
 
fj Wij,β 
P P
 j=1 ρj ij,β j=1 ρj   j=1 ρj 

where W ij,β = W i,β , and rα denotes the component of the posi- disadvantages still exist. For example, the local matrix is
tion vector in the direction α. The above FPM approximations asymmetric, which can cause numerical instability. Also, the
have 2nd order accuracy and are not sensitive to irregular par- choice of the kernel function sometimes is limited when
ticle distributions that may produce considerable numerical solving the partial differential equations (PDEs) with higher
oscillations for the conventional SPH approximations. derivatives, and this may lead to some special requirements
on the kernel function.31 By replacing W ij,β in Eq. (14) with
6. Kernel gradient free SPH (KGF-SPH) (rj β − ri β )W ij , a field function f and its first derivative at
Although the improved methods mentioned above have particle i can be written as kernel gradient free (KGF) SPH
higher accuracy than the conventional SPH method, some approximation32

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-4

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

N mj N mj  −1  N mj
(rjα − riα )Wij
 
Wij fj Wij
P P P
j=1 ρj j=1 ρj j=1 ρj
" #
fi
   
=  N m
   .
N mj N mj (15)
fi,α  P j β β β β
  P
β β
 j=1 ρj (rj − ri )Wij (rjα − riα )(rj − ri )Wij   fj (rj − ri )Wij 
P 
j=1 ρj   j=1 ρj 

In KGF-SPH, the discrete scheme of the Laplacian opera- asymmetric form as

tor can be expressed as33
N mj
fj Wij

 P
N mj 
j=1 ρj
 


2 fj − fi Wij
P 
=

f
j=1 ρj

i

N mj


∇2 fi =

. (16)

Wij
 P
αd ηd

j=1 ρj





 . (18)
N mj
The parameters αd and η d in Eq. (16) are determined by a spe-


( f f )W
P
−

j i ij,α

j=1 ρj

cific kernel function and space dimensions. αd and η d for a



fi,α =


cubic spline function are 15/(7πh2 )3 and 31h4 π/210 in two-

P mj  α

 N 
rj − riα Wij,α



dimensional space, respectively. It is seen that no matter the j=1 ρj



kernel gradients exist or not, the numerical approximations

in Eqs. (15) and (16) can always be conducted. This is obvi- It is clear that the DFPM involves normalization on the approx-
ously different from the conventional SPH in which the ker- imation of the field function f and its derivatives with the
nel gradient should exist so as to conduct approximations smoothing kernel function and its derivatives, respectively.
for a field function and its derivatives. Also, the corrective On the one hand, DFPM inherits advantages from the con-
matrix derived by KGF-SPH is symmetric and this is differ- ventional SPH because only simple normalizations are used,
ent from FPM in which the corrective matrix is asymmetric. and thus it is flexible, cost-effective, and easy in coding. On
It is noted that Zhang and Batra proposed the symmetric SPH the other hand, as DFPM also possesses major features of
(SSPH) method34 which has similar ideas with KGF-SPH and FPM, it is more accurate than the conventional SPH and is
is mainly applied to elastic-dynamic problems in solid mate- not sensitive to the particle distribution or selection of the
rials. Different from SSPH with a 6 × 6 matrix in 2D or a smoothing function and smoothing length. More importantly,
10 × 10 matrix in 3D space, KGF-SPH only requires invert- different from FPM, CSPM, and KGC, no matrix inversion is
ing a 3 × 3 matrix in 2D or a 4 × 4 matrix in 3D spaces. required in DFPM. Therefore, the problem of ill-conditioned
Therefore, KGF-SPH can be easily extended to complex flow matrices can be effectively avoided, and this is very attrac-
problems.35 tive for modeling practical problems such as free surface
flows.
7. Decoupled finite particle method (DFPM)
As corrective matrices are used in FPM, CSPM, and KGC, B. SPH formulations for fluid flows
for modeling practical problems with extremely disordered By using the aforementioned SPH kernel and particle
particle distribution, the corrective matrices for some parti- approximations, a partial differential equation system can be
cles may be singular. The ill-conditioned corrective matrix can written into a corresponding SPH formulation. For exam-
lead to numerical errors and may even terminate the simula- ple, for viscous incompressible fluid flows, the mass and
tion. Based on such considerations, Zhang and Liu36,37 pro- momentum equations in a Lagrangian frame of reference are
posed a decoupled FPM. The corrective matrix L in FPM can described as follows:
be simplified as
dρ
N mj

 = −ρ∇ · u
 dt



Wij , 0 , (19)

 P 

j=1 ρj  ρ du = −∇p + τ + F


 

 
L=
  
, (17)

N mj  dt
α α
 
0, (rj − ri )Wij,α 

 P 
j=1 ρj

 

where t represents the time, ρ is the scalar density, u is the
 
velocity vector, p denotes the pressure, τ is the viscous stress,
by conducting Taylor series analyses with integration approx-
and F is the body force.
imation and assuming diagonal dominance of the resultant
corrective matrix. This new corrective matrix L is diagonally All aforementioned discretization schemes described in
dominant and decouples the approximations of the field func- Sec. II A can be used to approximate the fluid governing equa-
tion f and its derivatives. Therefore, correction with L leads tions (19). Different schemes can lead to different discretized
a decoupled finite particle method (DFPM). After some triv- equations with different advantages. The conventional SPH
ial transformations, the particle approximations of the field method is the earliest discretization scheme and can be easily
function f and its derivatives in DFPM can be written in the implemented in computer coding.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-5

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

It is noted that an extra equation is required to resolve the approximation schemes as discussed in Sec. II A, one other
pressure for closing the governing equation (19). When SPH is approach is particle regularization, i.e., re-distributing the
used to model compressible flows, the equation of state (EOS) irregular particles to make them more regular for better
is naturally included to describe the relation of pressure with accuracy. For such purposes, a rule is usually necessary to
density or energy (or temperature). When modeling incom- determine when and where to conduct the particle regu-
pressible flows, however, there are two main different ways larization and a procedure is also necessary for describing
to solve the pressure. In the early stage, an artificial com- how to conduct the particle regularization. As SPH parti-
pressibility equation is usually used to calculate the pressure, cles move according to the specific velocity that is obtained
and this is usually referred to as weakly compressible SPH from integration of acceleration related to stress, there
(WCSPH). A frequently used artificial EOS in WCSPH is38 are several particle regularization techniques, e.g., applying
artificial force or stress, adjusting velocity, simply moving
p = c2 (ρ − ρ0 ), (20)
particles, or combining them together. Frequently used par-
where ρ0 is a reference density and c is the sound speed, and ticle regularization techniques include XSPH, tensile correc-
it is a key factor deserving careful consideration.39 tion (TC), dynamics stabilization (DS), and the particle shifting
technique (PST).
One possible problem for the artificial EOS is that the
resultant pressure is usually rough and oscillatory. Some A. XSPH
researchers try to include the constraint of constant density
into the SPH method, somewhat similar to the moving particle In the XSPH model, a local average velocity is acquired by
semi-implicit method.6 This leads to projection-based incom- taking into account the neighbors’ velocities within the inter-
pressible smoothed particle hydrodynamics (ISPH) meth- ested particle support domain. Then, the particle movement
ods.40,41 In ISPH, the pressure is implicitly obtained from is modified based on the local average velocity to attain a
solving a pressure Poisson’s equation more isotropic particle distribution. In this technique, parti-
cles move according to the modified velocity as46
∇Pn+1 ∇ · (un + τ∆t + F∆t)
!
∇· = . (21) dri X mj 
ρ

∆t = ui − ε ui − uj Wij , (25)
dt ρ̄ij
j
Another approach to overcome spurious pressure fluctuations
is the δ-SPH model42 in which a diffusive term is added to where ρ̄ij = (ρi + ρj )/2 and ε is an adjustable parameter.
the continuity equation in order to remove the spurious high- Shahriari et al.47 pointed out that XSPH does not always pro-
frequency oscillations. The density diffusive term reads as duce significant improvements, and it may even generate non-
X mj physical solutions in the presence of real viscosity, especially
ρ δ = δhc ψ · ∇i Wij , (22) when the flow is subjected to high velocity gradients or when
ρj ij
j the particle resolution is limited.
 ri − rj f
B. Tensile correction (TC)
 g
ψij = 2 ρi − ρj 2
− h∇ρiLi + h∇ρiLj , (23)
r 
 ij  Monaghan48 added a repulsive force into the interaction
of inter-particles in order to remove the numerical insta-
X mj 
bility and avoid clustering or penetrating of inter-particles.


 h∇ρiLi = ρi − ρj Li · ∇i Wij
ρj

The repulsive force term is usually referred to as artificial


j



stress.49–51 Since it increases with the decrease in the dis-



 −1 , (24)
mj  tance between inter-particles, the repulsive force may be

 X

L =

rj − ri ∇i Wij 

  
 i  ρj constructed by the kernel function as



  j 
!
dui X
where the coefficient δ determines the intensity of the diffu- =− mi RSnij ∇i Wij , (26)
dt TC
sion of density and is usually set to be 0.1. ∇i W ij is the kernel j
gradient of particle i, and rij = |ri − rj |. In Eq. (24), the first term
Wij
is the discrete scheme of the Laplacian operator ∇ · ∇ρ. h∇ρiL Sij = , (27)
W(∆)
represents the renormalized density gradient.43
where ∆ denotes the average particle spacing in the neighbor-
hood of particle i, and n is usually taken equal as 4 for fluid
III. PARTICLE REGULARIZATION TECHNIQUES dynamic simulations. The factor R is expressed as follows:
As mentioned before, the accuracy of the conventional R = Ri + Rj , (28)
SPH particle approximation is closely related to the deploy-
ment of particles. Except for boundary or interface areas, where Ri depends on the pressure p and the density ρ
a regular particle distribution leads to 2nd order accuracy, ε | pi |
 Ri = ρi2 pi < 0

while an irregular particle distribution leads to accuracy

. (29)
less than 1st order.44,45 Apart from using higher order SPH  Ri = 0 pi > 0


Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-6

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

The parameter ε depends on smoothing length h, n, and the particles i and j. r̄i is the average particle spacing in
number of spatial dimensions, and it is usually taken as 0.2. the neighborhood of the particle i, i.e.,
1 X
C. Dynamic stabilization (DS) r̄i = rij . (36)
N
j
Inspired by Monaghan’s inter-particle repulsive force
concept48 for eliminating tensile instability, Tsuruta et al.
developed a dynamic stabilization scheme.52 It is also a par- Step 2: After the shifting vectors of all fluid particles are
ticle regularization scheme originally proposed for the MPS calculated in accordance with Eqs. (33)–(35), the
method. The aim of the DS scheme is to provide a mini- following equation is then applied to correct field
mum required stabilizing acceleration to keep particles uni- variables:
formly spaced. The DS-related acceleration is obtained as ϕi0 = ϕi + δri · (∇ϕ)i , (37)
follows:
dui
!
m X DS where i and i0 stand for the previous and mod-
= i Fij Wij , (30) ified variables, respectively. ϕ stands for a gen-
dt DS ρi
j eral field variable which represents the particle
position r, density ρ, velocity u, pressure p, and
0 r  ≥ φ temperature T.
 ij  ij





FDS = 2. Lind et al.’s algorithm

, (31)
r
ρi ρj

ij  − * φ 2 − rij,y 2 − rij,x + rij  r  < φ
 ∆t ( ρi +ρj )  ij  ij

 2 ij     rij  In Xu et al.’s algorithm, the direction and amount of the

shifting vector are determined from the arrangement of the
, -

φi + φj nearby particles. By using Fick’s law to control the direction of

φij = (1 − αφt ) , (32) the shifting vector, Lind et al.54 made a further improvement
2
on Xu et al.’s algorithm and associated numerical parameters
where α φt is the ratio of the time step to the Courant number, controlling the degree of particle shifting with a given time
φ signifies the particle diameter, and ∆t stands for the time step and particle resolution
step. rij (= ri − rj ) denotes the position vector, and rij,x and
rij,y are the horizontal and vertical components of rij , respec- h2 X  Wij 4  mj
 !
tively. The DS scheme can guarantee conservation of both δri = − 1 + 0.2  ∇i Wij , (38)
2
j 
 W(∆)  ρj
linear and angular momenta because it generates radial and
anti-symmetric inter-particle forces. where ∆ is the initial particle spacing. In order to reduce the
error caused by possible large shifting, Lind et al. set an upper
D. Particle shifting technology (PST) limit on the particle shifting distance, defined to be 0.2h. It can
1. Xu et al.’s algorithm be found that Eq. (38) is similar to Eq. (26). As pointed out by
Colagrossi et al.,55 the vector − ∇i W is always along the direc-
P
To enhance the accuracy and stability of particle meth-
tion of the maximum lack of “mass” and maximum anisotropy.
ods in the Lagrange framework, Xu et al.53 added a correct-
After the shifting vectors of all fluid particles are calculated,
ing velocity into fluid governing equations to weakly shift
Eq. (37) is still used to correct the field variables.
particle positions. Generally, PST contains the following two
steps: Furthermore, Lind et al. developed the shifting technology
for free surfaces54
Step 1: Solve the shifting vector of particle i by the follow-
ing equation: δr = δri · τ
 i,τ


δri = CζRi , (33)  δr = α(δr · n − β) . (39)
 i,n i

where δri is the shifting vector of particle i. C repre- Here, the shifting vector δri is calculated using Eq. (38). τ and
sents the diffusion coefficient and is usually chosen n denote the tangent and normal vectors to the free surface,
as 0.04.53 ζ and Ri are the shifting magnitude and respectively. Parameter α is chosen such that α ∈ [0, 1] and
direction, respectively, and they can be solved by controls the amount of diffusion in the normal direction. For
the following equations: high speed violent flows, such as dam-break, α = 0. For slow
flows of long durations, such as regular wave propagation, a
ζ = Umax ∆t, (34)
small degree of normal shifting is employed to prevent irreg-
ular particle distributions, thus α = 0.1. β is taken to be the
X r̄2
Ri = i
nij , (35) normal concentration gradient at the unperturbed plane free
j
rij2 surface.

where Umax is the maximum velocity of the fluid 3. Khayyer et al.’s algorithm
particles, rij is the distance between particles i and Khayyer et al.56 further developed a particle shifting
j, and nij is the unit displacement vector between algorithm of free surfaces with the aim of enhancing the

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-7

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

performance of particle shifting for fluid flows contain- Step 2: The position ri m+1 of the particle i at the m + 1th
ing interfaces. The PST algorithm for free-surface and its iteration step is calculated by
vicinity particles ignores the shifting normal to the interface
and is written as rm+1
i = rm m
i + δri . (45)

h2 X mj Step 3: Repeat steps 1 and 2 until the iterative calcula-

δri = − (I − ñi ñi ) · ∇i Wij , (40) tion achieves the given convergence criterion. The
2 ρj
j convergence criterion can be defined as
where ñ is a corrected unit normal vector, given by max  χm  − χ  ≤ ε,
0 (46)
 i
P mj
Bi · ∇i Wij X
j ρj χim = h2 Wijm , (47)
ñi =   , (41) j
 P mj
Bi · ∇i Wij 
j ρj  where χi is the sum of the kernel function corre-
lating particle i. χ i m denotes the sum of the kernel


X mj 
−1 function correlating particle i at the mth itera-
tion step. χ0 denotes the average value of the sum

Bi = *. ri − rj ∇i Wij +/ . (42)
ρj at the initial moment. ε is an adjustable parame-
j
ter. The smaller the parameter ε is, the more uni-
, -
4. Huang et al.’s algorithm form the particles obtained by IPST are. Numerical
Based on the studies by Xu et al.,53 Colagrossi et al.,55 and simulations show that the particles are sufficiently
Lind et al.,54 Huang et al.57 developed a modified formula to uniform when ε is less than or equal to 0.2.
calculate the particle shifting vector δr
Step 4: After step 3, the field variables must be corrected by
" 
Wij 4
 # the following equation:
X mj 1 + 0.24 W(∆)
δri = −Umax ∆t h ∇i Wij . (43) ϕi0 = ϕi + (ri0 − ri ) · (∇ϕ)i , (48)
ρj 2
j
where ϕ i and ϕ i0 represent the previous and modi-
Again Umax is the maximum velocity of fluid particles. Equa- fied field variables of particle i, respectively. ri and
tion (43) is similar to Eq. (38), while the main difference ri0 are the previous and modified positions of the
between Eqs. (38) and (43) is that the smoothing length h particle i, respectively. It should be noted that the
in Eq. (38) is replaced by the maximum convection distance previous field variables and particle positions are
Umax ∆t. As pointed out by Xu et al.,53 the shifting distances used to calculate the gradient (∇ϕ)i in Eq. (48).
should be large enough to prevent particle clustering, but
It is noted that all above-mentioned methods (XSPH,
much less than convection distances to ensure that particles
TC, DS, and PST) are able to regularize particle distribu-
still move along streamlines after a weakly correction of par-
tions to some extent. XSPH has been rarely used in SPH
ticles’ positions. In fact, the maximum convection distance
due to its low accuracy. TC and DS are somewhat simi-
used here is similar to the characteristic velocity of the flow
lar and have exact conservation of both linear and angular
considered in Arbitrary Lagrangian Eulerian (ALE) SPH.58
momenta. However, some latest results56 showed that the DS
When the smooth strength is larger than the average par- scheme does not always provide the regular particle distri-
ticle spacing, the conventional PST is usually able to effectively bution as PST. PST has become a widely used procedure to
improve the uniform degree of particle distribution in the pro- overcome irregular particle distributions. Different improved
cess of SPH simulation. It is not valid if the smooth strength versions of PST have their own features and advantages. The
is less than the average particle spacing. The condition that work of Xu et al. is the original contribution in particle shift-
the smooth strength is less than the average particle spacing ing, and it is demonstrated to be successful in many situa-
may be necessary while modeling flows around thin structures tions. The shifting algorithm of Lind et al. usually produces
such as airfoil.59 For this reason, Huang et al.60 further devel- more regular particle distributions, especially for simulations
oped an iterative PST (IPST). IPST is achieved by the following with a closed computing domain. The shifting algorithm of
four steps: Khayyer et al. improved the shifting accuracy for free-surface
and its vicinity particles. The shifting algorithm of Huang
Step 1: The shifting vector δri m of the particle i at the mth et al. has the same effect as the algorithm of Xu et al. but
iteration step is calculated by requires less computational cost. Generally, the iterative PST
!m does not lead to a serious non-conservation problem when
X mj
δrm = −U h∆t ∇ W , (44) a proper convergence criterion is used to limit the shifting
max i ij
i ρj distance.61 The iterative PST shows better performance than
j
the conventional PST when the smooth strength is smaller
where m (=1, 2, 3, . . .) is the number of iterations. than the average particle spacing. It should be noted that

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-8

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

! n1 ! n2 #
rij
" !
the high accuracy particle approximation schemes, such as  r0 r0 r0

 D − , ≤1
FPM, are more suitable to be combined with PST because rij rij 2 rij

rij


Fij = 

they are able to reduce the errors caused by the particle ! , (49)
 r0
>1

0

shifting.


 rij

where Fij is the repulsive force between a solid particle j and

IV. BOUNDARY TREATMENT ALGORITHMS an approaching fluid particle i, and rij is the magnitude of the
As pointed out by Liu et al.,8,62 boundaries are inevitably position vector rij . Parameters n1 and n2 are usually taken as
truncated when conducting the SPH approximations, leading 12 and 4, respectively. r0 is the cut-off distance and usually
to boundary deficiency with low accuracy near the bound- set to be the initial particle spacing. D is a problem dependent
aries. These boundaries may include fixed or moving solid parameter with the scale of the square of the largest velocity.
boundaries, moving material interfaces, and changing free It is found that the L–J force is highly repulsive and goes to
surfaces. Hence, SPH suffers from the difficulty of bound- infinite when the two particles approach each other, and it is
ary treatment since its invention, and this has been hinder- also sensitive to parameters such as n1 , n2 , r0 , and D. For exam-
ing its further development and application in engineering ple, if the cutoff distance r0 is too large, some fluid particles
and sciences because boundary conditions cannot be imple- may feel early and unnecessary repulsive force. This leads to
mented directly and rigorously as in the grid-based methods. the occurrence of initial disturbance and even blowup of fluid
In this subsection, we will review some treatment techniques particles. If r0 is too small, fluid particles may have already
on the solid boundary, free surface detection, and multiphase penetrated the boundary before experiencing repulsive
interface. force.
Liu et al.62 developed a repulsive force model in which a
A. Solid boundary treatment finite distance-dependent repulsive force is exerted on fluid
Like other particle-based methods such as MD and DPD, particles approaching solid boundaries
both the obstacle areas and flow regions can be discretized rij
with particles in SPH. The SPH particles in the boundary area Fij = 0.01c2 χf(η) , (50)
rij2
are sometimes referred to as virtual particles or ghost parti-
cles in different literature. The deployment of virtual boundary
rij
particles and the interaction with fluid particles are important 1− , 0 < rij ≤ 1.5∆



χij =  1.5∆

in implementing the solid boundary condition. Some stud-  , (51)
0 otherwise

ies on original SPH only place virtual particles right on the


boundary line (or the boundary surface in 3D). These virtual
particles on the one hand take part in the particle approx- 
 2/3, 0 < η ≤ 2/3
imations of fluid particles and, on the other hand, exert a


 (2η − 1.5η ) 2/3 < η ≤ 1


 2
repulsive (or penalty) force on approaching fluid particles, and


f(η) =  , (52)
therefore these virtual particles are sometimes referred to as 


 0.5(2 − η)2 1<η<2
repulsive particles. As the conventional SPH method does not




0 otherwise

have 1st order accuracy in the boundary area, in recent liter- 
ature, virtual particles are also placed in the boundary area,
where η = rij /0.75hij , hij = (hi + hj )/2, and ∆ is the initial
together with the high order particle approximation schemes
distance of two adjacent particles.
when estimating the interaction of fluid and solid particles, so
as to improve the computational accuracy in boundary areas. The boundary force model developed by Monaghan and
Actually, virtual particles can be placed both on the bound- Liu et al. did not consider the effect of particle velocity
ary and in the boundary area, just as in the coupled dynamic on the boundary force. When fluid particles move in par-
SBT algorithm.62 Therefore, the repulsive force model and the allel to the boundary, these forces can feel a non-uniform
deployment of boundary particles are two key issues in the normal force and a non-zero tangential force, which may
solid boundary treatment and will be reviewed in following cause a large disturbance to the flow near a boundary. An
context. alternative form of the repulsive boundary force that may
reduce the disturbance on the flow was proposed by Rogers
1. Repulsive force models and Dalrymple63 in modeling tsunami waves and run-up is
Since the particles continuously move with fluid flows written as
in Lagrangian methods, a nonphysical phenomenon may be F = nR(ψ)P(ξ)ε(z, u⊥ ), (53)
present that the particles can easily penetrate the solid where n is the unit normal direction of the solid boundary.Ψ is
walls. To avoid this unphysical penetration, Monaghan39 pro- the perpendicular distance of the particle from the wall. R(Ψ )
posed a force model where a repulsive force is exerted on is a repulsive function
the fluid particles adjacent to the boundary. The repulsive
force is similar to the Lennard–Jones (L–J) force in molecular 1
R(ψ) = A √ (1 − q), (54)
dynamics q

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-9

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

where q = Ψ/2h is the normalized distance between particles 


 0.02 z≥0
and the wall. The coefficient A is



ε(z) =  0 > z ≥ −h0 ,
 z/h0  + 0.02 (58)
1


A = 0.01c2 ,

(55)

h


1 z/h0  > 1

where c is the speed of sound. The function P(ξ) is chosen so
that a fluid particle experiences a constant repulsive force as 
 0 u⊥ > 0
it travels parallel to the wall




ε(u⊥ ) =  |20u⊥ |/c |20u⊥ | < c .

!! (59)
1 2πξ


P(ξ) =

1 + cos , (56) 
|20u⊥ | > c

2 1

∆b 

where ∆b is the distance between any two adjacent boundary u⊥ = (uf − ub ) · n is the normal velocity of the fluid particles,
particles. ξ is the tangent projection of the distance between where the subscripts f and b refer to fluid and boundary parti-
fluid and boundary particles. Finally, the function ε(z, u⊥ ) is cles, respectively. z is the elevation above the local still water
a modification to Monaghan and Kos’s64 original suggestion level.
and adjusts the magnitude of the force according to the local It can be seen that the above repulsive boundary force by
water depth and velocity of the water particle normal to the Rogers and Dalrymple63 includes many parameters which are
boundary inconvenient for SPH practitioners to translate this method
ε(z, u⊥ ) = ε(z) + ε(u⊥ ), (57) into the code. A more simple form was given by Han et al.65 as

dim
  

 *. 2 min ui − uj · nj , −1 Wij hij nj +/  
U ui − uj · nj < 0



 − max
Fij = 
 . r · n  /
  ij j , (60)

 , -

  
0 ui − uj · nj ≥ 0



where dim is the number of spatial dimensions. Fij represents applied into various particle methods. However, fixed virtual
the repulsive force of the boundary particle j acting on the particles are usually difficult to be arranged inside slender
fluid particle i. To keep the same dimension on two sides of bodies59 unless multiple resolutions are considered to make
Eq. (60), Zhang et al.66 made a modification on Eq. (60) and an uninform particle system.67 Therefore the dynamic virtual
obtained the following model: particles sometimes are necessary that are generated at each
  d
time step in order to adapt the neighbor fluid particles in sim-

 ui − uj · nj Wij hijim nj +   ulations of a slender body.68,69 The dynamic virtual particles
ui − uj · nj < 0
 *
 −.c

 . //
Fij =  r · n  , (61) are created by mirroring the fluid particles with respect to the

  ij j  corresponding boundary particles, as shown in Fig. 1.
 ,
 -  

 0 ui − uj · nj ≥ 0
Different approaches can be used to obtain the field vari-
where the threshold (i.e., −1) is removed, and the sound speed ables of virtual particles. For the fixed ones, they may be
c is used to replace U2max . associated with the boundary particles which are calculated
2. Virtual particle models according to the governing equations or associated with the
projecting (intercalated) points which are intercalated from
Virtual particles are generated to represent the solid
the interior fluid particles (see Fig. 2). However, the dynamic
boundary areas and used to calculate the fluid particles near
virtual particles are usually associated with the interior fluid
the boundary for removing the boundary deficiency problem,
particles (see Fig. 1).
further achieving the purpose of improving the SPH numerical
accuracy near boundary areas. The pressure of the virtual particle is calculated by
Generally, virtual particles are fixed in the boundary area
" #
Dub
and generated only once at the first time step. The thickness pvi = pas + ρ · nb − g · nb ε∆. (62)
Dt
of the virtual boundary particles is determined by the kernel
function and the initial particle spacing. The fixed virtual par- The subscript vi denotes the virtual particle, as denotes the
ticles are easily implemented and their field variables can be associated particle (boundary particles, projecting points, or
conveniently calculated, and therefore, they have been widely fluid particles), and b denotes the boundary particles. n is the

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-10

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

B. Free-surface particle detection algorithms

1. Kernel based model
Since the support of the kernel near the free surface
is truncated, the free surface particles may be identified by
using the density ratio. If the ratio of the density ρi associated
with a particle is less than a specified value ρ0 , the particle is
then identified as a free surface particle. Specifically, the fol-
lowing equation should be satisfied to identify a free surface
particle:70
ρi ≤ 0.99ρ0 , (63)
where the particle i is considered as a free surface particle.
The density ρi is calculated by
X
ρi = mj Wij . (64)
j

For the MPS method, the particle number density is usu-

FIG. 1. Diagram of dynamic virtual particles. ally used to detect the free surface particles. For example,
Koshizuka and Oka6 applied the formula

unit normal vector of the boundary particle. g is the accel- di < αd0 , (65)
eration of the gravity. ε∆ is the distance between the virtual where d0 is the initial number density and di is the particle
particle and the associated particle. number density of particle i, calculated by
For a fixed solid boundary, there are different meth- X
di = Wij . (66)
ods to calculate the velocity of the virtual particles. The
j
velocity of the virtual particles is usually set as zero for
the no-slip wall boundary or equals to the tangential veloc- When di < αd0 , the particle i is considered as a free surface
ity of the associated particles for the free-slip wall bound- particle. α is problem-dependent and different values can be
ary. Sometimes, the velocity of the virtual particles may be used, e.g., 0.97 in the work of Koshizuka and Oka.6
set as the opposite velocity on the associated particles in
Another popular method is to use the divergence of a
order to prevent the fluid particles from penetrating the
particle position
solid wall. Also, the velocity of the virtual particles may be
set according to the ratio of virtual particles and associ- X mj
∇·r= rij · ∇i Wij . (67)
ated particles to boundary.38 For a moving solid boundary, ρj
j
the velocity of the virtual particles can be set as the veloc-
ity of the moving boundary or calculated by the dynamic This divergence is equal to 2 for 2D simulations in the core of
equations. the domain and much below 2 for surface particles. A criterion

FIG. 2. Diagram of fixed virtual particles

associated with (a) projecting points and
(b) the boundary particles.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-11

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

set to 1.5 is used to determine which particles belong to the

surface.71
2. Eigenvalue and geometry
Based on eigenvalue and geometry, Marrone et al.72 pro-
posed an algorithm to identify free surface particles. This algo-
rithm is composed of two steps. The first step is a selection
process to quickly perform the second step and acts as a tool
to detect particles belonging to separated fluids such as jets
and drops which could hardly be detected by the second step.
The renormalization matrix is considered in the first step, FIG. 3. Sketch of the scan region used in the algorithm.72
given by
 −1
mj 
X

Bi =  rj − ri ∇i Wij  . (68) particles far from the free-surface, characterized by high val-
 j ρj  ues of λ. B denotes the region composed of particles which are
close to the free-surface or are in regions of the domain where
The minimum eigenvalue λ of the matrix B−1 is used to deter-
particles are not uniformly spread.
mine the regions of the fluid domain where the free surface
can exist or not. The minimum eigenvalue λ depends on the The second step is to complement the first step, while
spatial organization of the particles j in the neighborhood of a more precise and reliable control is carried out on par-
the considered calculation point i. When going away from the ticles belonging to B in order to complete the free-surface
fluid domain, the minimum eigenvalue tends theoretically to detection. It is possible to get a more accurate evaluation by
0, while the minimum eigenvalue inside the fluid domain tends defining a normal direction
theoretically to 1. If the renormalized Gaussian kernel shape73 Pm
with a support radius 3h (where h = 1.33∆ is the smooth- −Bi · ρjj ∇i Wij
j
ing length and ∆ is the average particle spacing) is used, the ni = 
following relation is obtained: . (70)
 mj 
Bi · ρj ∇i Wij 
P
j
λ ≤ 0.2



 i∈E 
0.2 < λ ≤ 0.75 .

i ∈ B
 (69) From this normal direction ni , a region is defined, which
0.75 < λ is named scan region, as shown in Fig. 3. The point T is

i∈I

located at distance h from i in the normal direction and the
E denotes the region composed by some particles belonging unit vector τ is perpendicular to n. The conditions to assess
to separated fluids such as thin jets and drops, characterized whether the particle i belongs to or not to the free-surface are
by low values of λ. I denotes the region composed of interior described as

f √ r  < hg


 ∀j ∈ N rji  ≥ 2h,  jT  i < Free surface


 f √ g
 ∀j ∈ N r  < 2h ,
 r · n + rjT · τ < h i < Free surface . (71)


  ji   jT


otherwise i ∈ Free surface



N denotes all the particles in the domain. If the first condi- 3. Geometric model
tion is true, it means that the neighbor under examination Ma and Zhou74 proposed a geometric model to identify
is in the dark region (S1) in Fig. 3. If the second condition free surface particles, and later Huang et al.57,75 extended it
is satisfied, it means that the neighbor is in the gray region for treating free surface in 3D spaces.
(S2). These two regions together form the whole scan region.
If any neighbor is located in the scan region, it means that In a 2D space, two orthogonal coordinate systems are
there is no cavity in the normal direction or that this cav- established for each particle, as shown in Fig. 4. With the help
ity has a diameter less than 2h so that particles inside it are of these coordinate systems, the support domain of any par-
deeply interacting and it is not a true cavity. Therefore, this ticle is divided into eight regions, q1 , q2 , q3 , q4 , Q1 , Q2 , Q3 , and
algorithm may track the free-surface of cavities of diameter Q4 . For particles in the vicinity of free surface, there is at least
larger than 2h. one region containing no more than three particles if the cubic

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-12

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

FIG. 4. Illustration of free surface particle

detection for the 2D geometry model. (a)
Coordinate system I. (b) Coordinate sys-
tem II. Reproduced with permission from
C. Huang et al., Int. J. Numer. Methods
Eng. 113, 179 (2018). Copyright 2018
John Wiley and Sons.57

spline function with h = 4∆/3 is chosen as the kernel, such as too large, some particles in the vicinity of free surface cannot
particle A in Fig. 4. In contrast, for the interior particles (away be detected. If the specific number is too small, interior fluid
from the free surface), each region must contain more than or particles away from the free surface may be misjudged as free
equal to three particles such as the particle B in Fig. 4. surface particles. Hence, the specific number “three particles”
or “four particles” is an empirical parameter and needs a large
In 3D space, the support domain of each particle is divided
number of numerical tests for different kernel functions.
into eight regions, Q1 , Q2 , Q3 , Q4 , Q5 , Q6 , Q7 , and Q8 , as shown
in Fig. 5(a). If the x and y coordinate axes rotate ∠45◦ around
the z coordinate axis, as shown in Fig. 5(b), the support domain C. Multiphase interface treatment
of the particle can be again divided into other eight regions,
The treatment of interfaces between different materi-
q1 , q2 , q3 , q4 , q5 , q6 , q7 , and q8 . For a particle in the vicin-
als, fluids, and phases is a crucial step in solving prob-
ity of free surface, there is at least one region containing
lems with multi-materials, multi-fluids, and multi-phases.
no more than four particles if the cubic spline function with
Improper treatment of interfaces may not only result in
h = 4∆/3 is chosen as the kernel. On the contrary, for an inte-
numerical oscillations in the interface area but also cause
rior particle, each region must contain more than or equal to
unphysical particle penetration. The discontinuities of fluid
four particles.
parameters often occur on the interfaces, posing great chal-
In the geometric model, the specific number “three parti- lenges on accurately and stably solving of the continuity and
cles (2D)” or “four particles (3D)” depends on the type of kernel momentum equations. Thus, special treatments need to be
function and the dimension of space. If the specific number is made when solving the governing equations.

FIG. 5. Illustration of the method for

detecting free surface particles for the
3D geometry model. (a) Coordinate sys-
tem I. (b) Coordinate system II. Repro-
duced with permission from Huang et al.,
“Coupled finite particle method for simu-
lations of wave and structure interaction,”
Coastal Eng. 140, 147 (2018). Copyright
2018 Elsevier.75

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-13

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

For multiphase flows, the density is usually obtained non-ignorable influence on the light fluid. To prevent such
directly from the SPH summation of the mass of all particles drawbacks, an alternative formula is given by
in the support domain of a given particle, rather than from the
continuum equation which may cause the inconsistency of the 1 1 X
− ∇pi = − Vj pj ∇i Wij . (77)
density and velocity fields by the use of the continuity equa- ρi ρi
j
tion.76–81 According to Hu and Adams’s work,82,83 the density
is solved by Equation (77) is the application of the conventional SPH
X
ρi = mi Wij . (72) method [see Eq. (4)]. Therefore, Eq. (77) remains some char-
j acteristics of the conventional SPH method. A more popular
Using this approach, the sharp density discontinuities at the formula to solve the pressure gradient is proposed by Hu and
interface can be treated naturally. Adams,82,83 expressed as
When the continuity equation is used to update the den- 1 1 X 
sity, the density re-initialization is usually applied to avoid − ∇pi = − pi Vi2 + pj Vj2 ∇i Wij . (78)
ρi mi
density oscillations and pressure instabilities.84 To adapt to j
the mass-discontinuity and density discontinuity issues on
multiphase interfaces, Chen et al.85 proposed a corrected Considering the same volume of particles for different fluid
density re-initialization formula phases, i.e., V i = V j , Eq. (78) is simplified as

ρHj Vj Wij
P
1 1 X 
j − ∇pi = − pi + pj Vj ∇i Wij . (79)
ρinew = P , (73) ρi ρi
Vj Wij j
j

  Equation (79) was originally developed by Colagrossi and Lan-

pj − p0 drini73 for the simulations of interfacial flows. It can be found
ρHj = + ρi0 , (74) in Eqs. (77)–(79) that only the volume, position, velocity, and
c2i
pressure of the neighboring particles are used such that the
where V j is the volume of particle j. ρHj is the density obtained denser fluid has an adequate influence on the lighter fluid even
by the pressure of particle j and the equation of state for par- if the density ratio is large.
ticle i, ci is the sound speed of particle i, and ρi0 denotes the The dynamic viscous coefficient also needs an adjustment
reference density of particle i. To avoid the artificial diffusions on the multiphase interface. A widely used form was developed
and to lower computational cost, Eq. (73) is often applied every by Hu and Adams82,83 as
20 to 30 time steps. The corrections could also be regarded as
converting particles of different phases into the same phase 1 1 X 2µi µj  2  rij · ∇i Wij  
as the concerned particle. The converted particle occupies the − (∇ · µi ∇)ui = − V + Vj2 ui − uj .
ρi mi µi + µj i r 2
volume of the original particle, while its density is converted j ij

to another phase. (80)

Equation (80) is similar to the formula proposed by Morris
Adami et al.80 developed a scheme to solve the pressure et al.38 Hu and Adams made the adjustment of the viscous
gradient for multiphase flow simulations, and the pressure coefficient to Morris et al.’s formula, i.e., replacing the original
term in the momentum equation was approximated as 2µi µj
dynamic viscous coefficient µi by the formula µi +µj . This dis-
1 1 X 2  ρi pj + ρj pi crete form is more popular for the discretized equation system
− ∇pi = − Vi + Vj2 ∇i Wij . (75)
ρi mi ρi + ρj in simulations of multiphase flows.76–78,85
j
When the numerical example is in the scale of meter,
This approximation is reasonable for multiphase flows with
surface tension effects at such scale are negligible. However,
small density differences. However, for those with large den-
when the scale reduces to micrometer and even smaller, the
sity ratios, the influence of the pressure from neighboring par-
surface tension has a significant influence on the fluid flow
ticles may be improperly ignored.85 For example, when parti-
with free surface or multi-phase flow with a sharp inter-
cle i and particle j are in two different phases with a very large
face. In SPH, it is easy and straightforward to include sur-
density ratio (e.g., ρi  ρj ), it derives ρi pj  ρj pi under the
face tension models, either using the continuum surface force
assumption of the continuous pressure over interfaces. Thus,
(CSF) model86–90 or the inter particle interaction force model.8
the following formula can be obtained:
To describe the surface tension, there are different models
ρi pj + ρj pi ρj pi in the SPH method, including the Cohesive Pressure Model
≈ ≈ pi . (76) (CPM), Inter-Particle Force (IPF) model, and Continuum Sur-
ρi + ρj ρi + ρj
face Force (CSF) model. Wang et al.20 gave a detailed review
The above analysis shows that Eq. (75) will reduce the denser for these models, and the readers may refer to their work for
fluid’s influence on the lighter fluid when the density ratio learning more information about the surface tension models.
is large. But in practice, the pressure of the denser fluid has Here, a simple comment on the CSF model is given. According

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-14

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

to Hu and Adams,82,83 the acceleration caused by the surface in various complex flows, as presented in Sec. VI. The formu-
tension fs is solved by lation of DPD can be expressed as97
!s
du fs 1 dri
= ui ,
Y
= = ∇· , (81)


ρ ρ dt

dt





 (84)
 m dui = FC + FD + FR ,
! 

1 1
Y 

=α I|∇C | 2 − |∇C | |∇C | , (82)  dt
|∇C | dim
where FC , FD , and FR are the conservative, dissipative, and
where α is the surface-tension coefficient, dim is the spatial random forces, given by
dimension, I is the unit two order tension vector, and C is
the gradient of a color index which has a unit jump across the FC = αij ω C (rij )eij ,
P
interface. The readers may refer to Hu and Adams’s work76,77 j
FD = − γij ω D (rij )(eij · uij )eij ,
P
and Adami et al.’s work80 for learning the method to calculate (85)
j
the color index C. On the basis of the CSF model by Brackbill dWij
FR = σij ω (rij ) dt eij ,
R
P
et al.,91 Adami et al.80 developed a direct model into the SPH j
method, expressed as
!s and αij , γ ij , and σ ij are the user-defined coefficients to char-
fs
!
du −ακn 1 ∇ · ∇C acterize the force strengths; dW ij is an independent incre-
= = =− α ∇C, (83)
dt ρ ρ ρ |∇C | ment of the Wiener process; ω C (rij ), ω D (rij ), and ω R (rij ) are
where κ is the curvature of the interface and n is the unit nor- the weight functions. To produce the correct thermal fluctu-
mal direction of the interface. According to Lafaurie et al.’s ations, i.e., to satisfy the fluctuation dissipative theory (FDT),
derivation,92 Eq. (82) is equivalent to Eq. (83). Equation (82) is the coefficients and weight functions need to satisfy
simpler than Eq. (83) in terms of form, whereas Eq. (82) is eas- q
ier to be extended into cases with more than two phases in the σij2 = 2kB Tγij , ω R (rij ) = ω D (rij ). (86)
context of the SPH method.78
where kB T is the Boltzmann temperature. For more details
about the above DPD formulations, refer to the work of
V. FROM SPH TO SDPD FOR COMPLEX FLUIDS Español and Warren.97
A. Relations of SPH to DPD and SDPD The conservative force FC in Eq. (85) produces an EOS that
can be approximated as98
As we know, SPH was invented initially for the simu-
lations of astrophysical problems at large scales10,11,14 and P = dkB T + aαd2 , (87)
later modified to laboratory-scale situations like viscous flow
and thermal conduction.93–96 Recently, it is further extended where d is the number density of the DPD particles, α is the
to the meso-scale, deriving the smoothed dissipative particle conservative coefficient, and a is a fitting parameter equal to
dynamics (SDPD),25 although it is proposed for the purpose of 0.101 ± 0.001 in DPD units. It has quadratic density dependence
remedying some deficiencies of the original dissipative par- so that the conservative force between particles is shown as a
ticle dynamics (DPD). The meso-scale is roughly defined as soft repulsive force. The soft nature allows quite large time
space-time scales from 10 nm to 0.1 mm and 1 ns to 1 ms, steps in DPD, compared with MD that usually needs to deal
respectively.1 The lower limits are associated with micro- with steep repulsive potentials. This is one of the main advan-
scopic atoms so that a mesoscopic system should be sub- tages of DPD, i.e., much faster than MD, and the reason why
ject to quantum mechanics with stochastic features. The DPD was attractive for MD users in the mesoscopic simula-
upper limits are associated with macroscopic materials, so it tions. However, it is also the reason why DPD was attacked
also should obey the laws of classical mechanics with deter- for CFD users in the simulations of complex flows. The soft
ministic features. Therefore, a mesoscopic method should repulsive force makes the classical DPD well model the behav-
present both microscopic fluctuations and macroscopic ior of gas, but cannot be used to simulate real fluid, especially
properties. liquid-gas-solid co-existing systems, such as droplet dynam-
ics and blood cells. A direct solution of this issue is to intro-
1. Dissipative particle dynamics duce a long-range attractive component into the conservative
The original DPD was introduced by Hoogerbrugge and force. Like the repulsive component, the attractive compo-
Koelman22 based on lattice-gas automation (LGA) to remove nent should also be a soft interaction to retain the advan-
its lattice artifacts and was formulated by Español and tages of the DPD method, and at short particle separations,
Warren97 as a sound statistical mechanics model a few years the repulsive component should be strong enough, relative
later. Groot and Warren98 provided the ranges of DPD param- to the attractive component to prevent the particle density
eters to achieve a satisfactory compromise among speed, sta- from becoming too high.1 Based on such considerations, Liu
bility, compressibility, and viscosity. Since then, it has been et al. developed a DPD model with short-range repulsion and
accepted as a classical mesoscopic model and widely applied long distance attraction.23 The DPD model was shown to be

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-15

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

effective in modeling the kinetics of phase separation, com- shows an artificial crystallization as the number of packed
plex multiphase flows, and if the interactions between sta- atoms exceeds a critical value, for example, 20 atoms given
tionary wall particles and mobile fluid particles are assigned in the work of Pivkin and Karniadakis,109 due to the increased
a similar form, the dynamics of fluid-fluid-solid contact point conservative force.
of two dimensions or contact lines of three dimensions can
The random force FR in Eq. (85) describes thermal fluctua-
be simulated.99,100 Similarly, the many-body DPD (MDPD) was
tions, whose strength σ ij depends on the dissipative strength
also proposed, improved, and spread,101–106 where an attrac-
γ ij and system temperature kB T in DPD, as given in Eq. (86).
tive component was introduced into the DPD conservative
This implies that the thermal fluctuations will be still present
force. The side effect of introducing the attractive component
or dominant even in a macroscopic problem, where a dissipa-
may cause the DPD particles to be accumulated or dispersed
tive particle represents a very large volume element. This is
excessively. To avoid this, a number density di , the same as in
obviously unreasonable because, generally, one expects that
SPH, was also associated with each particle in MDPD, which
the thermal fluctuations should be smaller and smaller, as
ensures that the repulsive force is dominant at a high number
the dissipative particles represent larger and larger volume
density, but the attractive force at a low number density. Obvi-
elements. This issue is also attributed to the scale-invariant
ously, MDPD allows for more general EOS than in the classical
feature of DPD formulation, i.e., there is no length scale
DPD, for example,103
associated with each dissipative particle in DPD.108 As a result,
P = dkB T + Adk , (88) it is ambiguous for DPD users how large thermal fluctua-
tions coming into play is reasonable for a given problem, and
where A is a coefficient. Eq. (88) can be obtained by adjust-
the unreasonable fluctuations may lead to wrong physical
ing the forms of attractive and repulsive components and even
phenomena.
providing a large k value to simulate real fluid. In spite of this,
MDPD still embodies its EOS implicitly, like DPD. It cannot Finally, the DPD formulation in Eq. (84) has no energy
explicitly give an arbitrary EOS that one wants to adopt. equation such that the classical DPD is suitable to treat with
isothermal problems. This point is, in fact, not very important.
The dissipative force FD in Eq. (85) is related to the viscos-
The DPD formulation can couple the energy equation easily,
ity of fluid, and a common expression often used by DPD users
by including an internal energy for each DPD particle, which is
is estimated as98
named energy-conserving DPD (EDPD).110
45kB T 2πγdrc5
ηDPD = 4πγdrc3
+ 1575
, (89) 2. Smoothed dissipative particle dynamics
where rc is the cut-off radius and γ is the conservative coeffi- The SPH formulations for incompressible flow can be re-
cient. There are two issues encountered when the users would written in a simple form as
like to set the viscosity by Eq. (89). One is the physical scales, dri
for example, what is the value of η DPD corresponding to the


 = ui

 dt
fluid viscosity η = 10−3 Pa · s? The other is that the viscosity


 m dui = F + F



η DPD depends on the number density d such that the viscos- C D , (90)

 dt
ity is completely different when changing the particle reso-


 Ti dSi = ∆QV + ∆QC



lution. In other words, the fluid simulated is dependent on

 dt
the numerical resolution, which is quite different with com-
where the momentum equation is broken into the conserva-
mon CFD approaches. The reason for these two issues is that
tive force FC and dissipative force FD , expressed as
the DPD formulation is scale invariant.107 In order to establish
a link between the DPD results and the physical scales, DPD FC = αij rij ,
P
users often obtain the DPD simulation results first and then fit j
P f g (91)
them to a fluid problem. Unfortunately, it is extremely diffi- FD = − γij uij + (eij · uij ) eij ,
j
cult to fit the real fluid problem exactly; therefore, one often
fits some key features of the problem, at the price of neglect- with the coefficients
ing some unfocused or secondary properties. For example, Pj
P 5η Fij
the compressibility is often sacrificed to exchange the exact αij = *. 2i + 2 +/Fij , γij = , (92)
d dj 3 di dj
fitting of the density and viscosity. Based on this treatment, , i -
the first issue can be approximately solved. However, Español and the entropy equation is divided into the viscous contribu-
and Warren108 recently pointed out that “we advise caution tion ∆QV and conductive contribution ∆QC , given by
with this simple approach (the classical DPD) because, usu-
ally, many other properties of the system go wrong.” The sec- 1P  2 
∆QV = γij uij + (eij · uij )2 ,
ond issue is innate because there is no notion of resolution, 2
P j (93)
grid refinement, and convergence in DPD.108 There have been ∆QC = χij Tij ,
j
attempts to address this issue, by linking the DPD param-
eters with the level of coarse-grained, i.e., the number of with the coefficient
atoms packed by a dissipative particle.107,109 However, a dis- Fij
χij = −2κ . (94)
appointing conclusion has been drawn that the simulated fluid di dj

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-16

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

To extend SPH to meso-scale applications, SDPD intro- the simulation accuracy. However, it is difficult to ensure that
duces a random force FR into the momentum equation for the particles distributed uniformly at every simulation step,
incorporating microscopic fluctuations and correspondingly but instead often disordered even if the initial distribution is
adds its entropy contribution ∆QR into the entropy equation. uniform. To achieve the similar simulation accuracy with grid-
Hence, the SDPD formulations are given by25 based methods, therefore, the number of particles used in
particle-based methods is often quite larger than the number
dri


 = ui of meshes in the grid-based methods, resulting in its disadvan-



 dt tage of higher computational cost. In this respect, a high per-
du


m i = F + F +F

, (95) formance computing (HPC) technique may still be a preferable
C D R



 dt choice for particle-based methods currently.
 Ti dSi = ∆QV + ∆QC + ∆QR




 dt As mentioned above, SDPD just adds a random force
FR and its entropy contribution ∆QR into the SPH formula-
where
tions. However, this addition is done in a consistent manner,
X dW
M ij
FR = Aij · eij , (96) coupling the physics behind the fluctuation-dissipation the-
dt orem. By casting SPH formulations in the GENERIC (general
j
and equation for the non-equilibrium reversible-irreversible cou-
1 P dW M ij dVij pling) framework,25 the expressions of FR and ∆QR in SDPD
∆QR = − Aij : eij uij + Cij , (97)
P
2 j dt dt are derived in the form of Eqs. (96) and (97). The GENERIC
j
with the coefficients framework ensures that SDPD formulations are thermody-
# 1/2 namically consistent, can respect the first and second laws
40η Ti Tj Fij Fij 1/2
" " #
of thermodynamics at the discrete level, and produce the
Aij = , Cij = 4κTi Tj . (98)
3 Ti + Tj di dj di dj correct thermal fluctuations governed by the Einstein distri-
bution function.25 The level of the fluctuations is not-scale
dV ij is an independent increment of the Wiener process, dWij invariant and depends on the volume Vi of a SDPD particle,
is a matrix form of independent increments of the Wiener i.e., the fluctuating velocity has the magnitude
process, and dWM ij is its symmetric part
D E DkB T 1 1
M ij = 1 dWij + dWT . v2i0 = ∼ , (100)
ρ0 Vi
f g
dW ij
(99) Vi
2
More details about the above SDPD formulations can be where D is the system dimensionality, kB T is the Boltzmann
obtained from an excellent review given by Ellero and temperature, and ρ0 is the system density. To illustrate this
Español.111 feature, Ellero and Español111 gave a good example of a solid
sphere with radius R suspended in a fluid, where D = 3, ρ0
Since SDPD just adds a random force and its entropy con- = 1000 kg/m3 , and kB T ≈ 4 × 10−21 J. If the sphere is a macro-
tribution in SPH formulations, it has the SPH framework with scopic one with R = 1 m, the resolution of SDPD particles is
specific macroscopic properties, such as viscosity, pressure, set by rcSDPD = 0.1 m (rcSDPD is the radius of a SDPD particle) to
and density. It can be understood as a direct numerical dis- resolve the fluid flow around the sphere. This
cretization of the Navier-Stokes equations, so it is regarded D Egives the volume
Vi = 10−3 m3 , and the fluctuation is about v2i0 ≈ 6 × 10−11 m/s.
as a top-down approach from the macro- to meso-scale.111 In
Under macroscopic flow conditions, this fluctuating velocity
SDPD, the macroscopic properties can be introduced directly
can be neglected completely, and SDPD is equivalent to SPH
as input of the simulation code, as in any other classical CFD
in this sense. That is, SDPD can be regarded as a substitute
methods. It also retains the features of SPH on mass, linear
for SPH at the macro-scale. If the sphere is a mesoscopic
momentum, and total energy conservations at the discrete
one with R = 0.1 µm, the resolution of SDPD particles is set
level.111 Along with these benefits, SDPD inherits the short-
by rcSDPD = 0.01 µm to ensure that the resolution is 10 times
comings of SPH itself, including the treatment of the solid
D E than the sphere size. Thus, we have Vi = 10
less −24 m−3 and
boundary, and high computational cost required.108 It adopts 2
vi0 ≈ 2 m/s. Under mesoscopic flow conditions, this fluc-
the kernel function of SPH, whose support domain is trun-
tuating velocity is dominant and can lead to non-zero diffu-
cated by the solid boundary, resulting in the impairment of
sional dynamics. Hence, SDPD can also describe thermal fluc-
numerical accuracy near the solid boundary. So far, there have
tuations correctly at the meso-scale, beside having specific
been a lot of SBT methods for resolving this issue (discussed in
macroscopic properties.
Sec. IV A). However, SDPD and SPH are still facing some chal-
lenges in developing high-efficient and easily implemented
SBT methods. Both SDPD and SPH, as particle-based meth- B. Modeling complex fluids
ods, generally require higher computational cost to achieve A commonly accepted definition of complex fluids is that
the similar simulation accuracy, compared with grid-based they are materials intermediate between conventional liquids
methods. The accuracy of a particle-based method is highly and solids, displaying fluid-like and solid-like behavior as well
dependent on the particle distribution, and a common sense is as fluctuations, for example, colloid suspensions, blood, and
that the more uniform the particles are distributed, the higher polymer solutions.112,113 Modeling these complex fluids has

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-17

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

attracted increasing attention during the last decade, as com- where η s is the viscosity of the solvent. τp has the different
puter technology and even the HPC technique make it possible forms for different complex fluids, for Oldroyd-B having
to get desired results with high fidelity, previously available ηp
only to simple fluids. The standard SPH, DPD, and SDPD equa- τp = (c − I), (105)
λ
tions are considered to model simple fluids, and a question
is thus raised: “How to generalize these methods to model where λ = η/k, η = η s + η p , η p is the viscosity of the polymer,
complex fluids?” and c is the conformation tensor governed by
∂c 1
To answer the question above, first let us review how it + u · ∇c = c · ∇u + ∇uT · c + (I − c). (106)
is solved in the macroscopic methods. The behavior of com- ∂t λ
plex fluids is in general characterized by the presence of a Now we turn our attention to model complex fluids by
large range of different time and space scales, i.e., multi-scale SPH, DPD, and SDPD. SPH is a macroscopic method itself, and
in nature.108,114 It is too expensive to describe the larger SDPD can be regarded as both a mesoscopic and a macro-
scales at the micro-scale only, for example, molecular dynam- scopic method. Hence, SPH and SDPD can adopt the same
ics requiring a huge number of particles. If at the macro-scale modeling approach used in macroscopic methods, i.e., con-
only, the microscopic feature of complex fluids cannot be structing constitutive models. For example, Fang et al.120 and
described at all, such as fluctuations and structural changes. Rafiee et al.121 constructed the Oldroyd-B model by the con-
Therefore, the multi-scale methodology should be a desired tinuum approach in SPH to study the viscoelastic free sur-
and promising way to cope with this complexity, especially face flows. Ellero’s group122–124 also adopted the Oldroyd-B
particle-based methods.115 Unfortunately, their applications model but constructed by the microstructural approach in
have been largely limited by the complicated modeling and SPH and SDPD, respectively, to study the viscoelastic flow,
implementation114 so that the single-scale approaches are still where a conformation tensor is associated with each particle
prevailing, which not only conceals the structures below the reading
considered scale but also smoothens the structures above that (∇ui ) (∇ui )T
scale, causing severe distortion of the simulations results.115 z }| { z }| {
T
At the macro-scale, a common approach is developing consti- X uij X uij
tutive relationships, which usually connect the strain rate D ċi = ci · *.− Fij rij +/ + *.− Fij rij +/
dj dj
with extra stress τ , j - , j -
τ = f(D), (101) 1 dc̃i
· ci + (I − ci ) + , (107)
and then coupled it into Navier-Stokes equations λ dt
and the last term is the stochastic part describing the con-
σ = −P I + τ, (102) tribution of thermal fluctuations. Note that Eq. (107) is a
direct discretization of Eq. (106) for the Oldroyd-B consti-
where σ is the Cauchy stress tensor, P is the hydrostatic pres- tutive model.111 This approach is too complicated, involving
sure, and I is the unit tensor. There are various approaches one more variable ci and its governing equation, to be directly
for constructing constitutive equation (101), and they are applied into DPD. Moreover, it is also an issue in DPD to add
roughly classified into two types: continuum and microstruc- the stress contribution τp of polymers, since the stress tensor
ture approaches.116 In the first approach, the fluid is assumed in DPD is not expressed explicitly. It is noted that, however,
to be a continuum without microscopic features, and differ- many of the interesting phenomena in complex fluids have
ent restrictions are imposed on the constitutive equation for been found to often occur at the meso-scale,108 and even one
practical applications. Oldroyd-B is one of its typical repre- suggests that a defining feature of complex fluids is the pres-
sentatives, where a Jeffrey element is considered, consisting ence of a mesoscopic length scale which necessarily plays a
of a dashpot and a Voigt element in series, and the Voigt key role in determining the properties of the system.112 Thus,
element refers to the parallel combination of a spring and is there a direct and simple way to model complex fluids in
a dashpot.117 In the microstructure approach, constitutive mesoscopic methods? The answer is yes, and a common way
models are constructed with the assumption on the behav- is by adding a new force to describe complex materials, such
ior of molecules and their interaction with the flow so that as polymers, droplets, and cells, which is treated as an addi-
the relevant details to fluids can be included at the micro- tional structure contributing to the new force. This approach
structural level. Finitely extensible nonlinear elastic (FENE)118 has also been emerged in macroscopic methods actually, i.e.,
and Phan-Thien-Tanner (PTT)119 models are typical repre- modeling fluid-structure interaction, where a new force is
sentatives of this approach, and the Oldroyd-B model can added into the Navier-Stokes equations as a body force. At the
also be derived from this approach. In general, the stress τ is meso-scale, the new force could combine microscopic infor-
split into two parts, a solvent contribution τs and a polymer mation of complex materials so that the complexity is included
contribution τp 116 at the meso-scale and re-emerges as complex hydrodynamics
τ = τs + τp , (103) behavior at the continuum scale.111 This approach not only
and the solvent is Newtonian fluid with incorporates the microstructural information of complex flu-
ids but also bypasses the construction of constitutive models
τs = 2ηs D, (104) used in macroscopic methods. Hence, most of DPD and SDPD

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-18

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

studies on complex fluids adopted this approach, for exam-

ple, colloids,125 blood,126 droplets,127 and polymers.128 This
approach is also generalized in SPH easily, just like solving
a problem with fluid-structure interaction, e.g., cell simula-
tions.129,130 In summary, in SPH and SDPD, fluid complexity
can be introduced by both (i) constructing constitutive mod-
els and (ii) adding a new force to describe the contribution
of complex materials. In DPD, however, the latter approach
should be the first choice, and the former one may be infea-
sible because of the implicit expression of stress tensor in
DPD.

VI. APPLICATIONS
SPH has been extended to a vast range of problems in
both fluid and solid mechanics because of the strong ability to FIG. 6. RBC simulations with DPD. (a) Stretching behaviors of healthy and malaria-
incorporate complex physics into the SPH formulations. In this infected RBCs.154 Reproduced with permission from Ye et al. “Stretching and
review, rather than providing a complete overview on appli- relaxation of malaria-infected red blood cells,” Biophys. J. 105, 1103 (2013). Copy-
right 2013 Elsevier. (b) Tumbling motion of a healthy RBC in shear flow.156 Repro-
cations of SPH, DPD, and SDPD, we are here to show some duced with permission from Ye et al., “Numerical modelling of a healthy/malaria-
applications that we have done in the past years on complex infected erythrocyte in shear flow using dissipative particle dynamics method,” J.
fluids and complex flows. Appl. Phys. 115, 224701 (2014). Copyright 2014 AIP Publishing LLC. (c) Healthy
and malaria-infected RBCs in a tube flow.157 Reproduced with permission from
A. Biological flows Ye et al., “Dissipative particle dynamics simulations of deformation and aggrega-
tion of healthy and diseased red blood cells in a tube flow,” Phys. Fluids 26, 111902
Biological flows are of interest to a variety of natural and (2014). Copyright 2014 AIP Publishing LLC. (d) A file of RBCs in a tube flow.155
biological processes and microfluidic applications. Much of Reproduced with permission from Ye et al., “A file of red blood cells in tube flow: A
early interest has been motivated, in 1980s and 1990s, by their three-dimensional numerical study,” J. Appl. Phys. 116, 124703 (2014). Copyright
2014 AIP Publishing LLC.
relevance to the motion of blood through vessels of the human
circulation. These early studies were mainly concentrated on
experimental observations131–133 and simple theoretical anal-
ysis,134,135 where the blood is often assumed as a fully devel- RBCs, where three stages are considered with the parasite
oped Poiseuille flow with homogeneous or heterogeneous maturation for the malaria-infected RBCs, ring, trophozoite,
physical properties, and the behaviors of cells are not con- and schizont, and the parasites are modeled as rigid bodies
sidered separately. The subsequent studies gradually moved embedded in the infected RBCs.154 Two ends of the RBCs are
to numerical simulations, and the cells are often assumed stretched under a stepwisely increasing external force; once
rigid and suspended into the homogenous plasma.136–138 The the RBCs achieve to the final steady state, the external force
motion of cells can be examined separately in these studies, is released. The infected RBCs were found to gradually lose
but their deformation and aggregation are still excluded. Until their deformability from and the recoverability to the origi-
the last decade, more careful numerical simulations became nal shapes, with the maturation of the parasites. This study
popular, with not only modeling RBCs (red blood cells) as laid the reliable foundation for the viscoelastic model of a
deformable objects but also considering the cell-cell inter- cell. Based on this cell model, we studied the dynamic behav-
action.126,139,140 Secomb et al.141–144 were among the first to iors of healthy and malaria-infected RBCs in shear and tube
implement such simulations with FEM, where the RBC is rep- flow, respectively.146,147 Three typical modes are reproduced
resented as a set of interconnected viscoelastic springs sus- in shear flow: the unsteady tumbling, transitional trembling,
pended in a viscous fluid. Then, a large number of similar stud- and steady tank-treading. With the parasite maturation, the
ies have emerged by using different numerical methods, such infected RBCs are less likely to undergo tank-treading mode
as boundary integral methods (BIM),139 FEM,145,146 the lattice but, instead, prefer a tumbling mode in shear flow. In tube
Boltzmann method (LBM),147,148 SPH,129,130,149,150 DPD,151,152 flow, the infected RBCs, which are positioned in the front of
and so on. Despite these significant progresses achieved in a healthy RBC, have a stronger and stronger resistance on the
recent years, computational modeling of cellular-scale biolog- healthy RBC by virtue of intercellular interaction, with the par-
ical flows continues to pose new challenges,153 for example, asite maturation. We also studied a file of healthy RBCs in a
(i) the simulation of “whole” blood with RBCs, white blood cells tube flow, examined the dependence of apparent viscosity on
(WBCs), and platelets, (ii) the simulation of a large number of the tube hematocrit, and showed the typical plug-flow veloc-
nucleated and anuclear cells, and (iii) the simulation of cells in ity profile for the flow of cell suspension.155 We realized from
complex network-like structures. these studies that DPD may be not a good choice for RBC sim-
ulations because there are several issues encountered upon
We now show our applications by DPD and SDPD in the implementation. First, what is the value of conservative coef-
past years, as shown in Figs. 6 and 7. With DPD, we studied ficient αij set in RBC simulations? If it is set by the widely used
the stretching and relaxation of healthy and malaria-infected formula98

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-19

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

FIG. 7. RBC simulations with SDPD. (a) The motion and deformation of a single RBC in a rectangular microchannel.158 Reproduced with permission from Ye et al.,
“Numerical studies of a red blood cell in rectangular microchannels,” J. Appl. Phys. 122, 084701 (2017). Copyright 2017 AIP Publishing LLC. (b) The motion and deformation
of a single RBC in a cylindrical microchannel. Reproduced with permission from Ye et al., “Hybrid smoothed dissipative particle dynamics and immersed boundary method for
simulation of red blood cells in flows,” Phys. Rev. E 95, 063314 (2017). Copyright 2017 American Physical Society. (c) The motion and deformation of a single RBC in a curved
microchannel.160 Reproduced with permission from Ye et al., “Red blood cell motion and deformation in a curved microvessel,” J. Biomech. 65, 12 (2017). Copyright 2017
Elsevier. (d) The motion and deformation of a single RBC in a constricted microchannel.162 (e) The motion and deformation of a single RBC in a spiral microchannel.160
Reproduced with permission from Ye et al., “Red blood cell motion and deformation in a curved microvessel,” J. Biomech. 65, 12 (2017). Copyright 2017 Elsevier. (f) The
motion and deformation of a single RBC in a bifurcated microchannel.161 Reproduced with permission from Ye et al., “Three-dimensional motion and deformation of a red
blood cell in bifurcated microvessels,” J. Appl. Phys. 123, 064701 (2018). Copyright 2018 AIP Publishing LLC.

75kB T complex microvascular networks, as shown in Fig. 8, in the

αij = , (108)
d hope that it will be more realistic to simulate blood sus-
a blank zone without particles will be formed at the front pensions flowing through physiologically microvascular net-
of the RBCs, while a lot of particles are gathered behind works that are characterized by multiple bifurcating and
the RBCs along the flow direction. This is the compressibil- merging vessels, as well as geometrically complex lab-on-chip
ity issue of DPD, which becomes more serious in multiphase
flows or liquid-gas-solid co-existing systems, as mentioned in
Sec. V A 1. In order to make the particle distribution uniform,
i.e., enough incompressibility, we ignore the common expres-
sion in Eq. (108) but, instead, use a large αij value artificially,
for example, 200 sometimes in our work compared with 25 in
the most of the DPD studies with d = 3 and kB T = 1. We cannot
find the theoretical reason why αij is set to be 200, but just
know that this value can make the particles distributed uni-
formly. Second, what is the value of dissipative coefficient γ ij
in simulations? If it is set by the common expression in Eq. (89),
the time scale in simulations is often inconsistent with the real
physical scale because of much higher Reynolds number in
simulations than the real one. This is attributed to the scale-
invariant issue of DPD, as mentioned in Sec. V A 1. In order to
provide a small Reynolds number, we set γ ij = 150 in our work
much larger than 4.5 given by Eq. (89) for d = 3, rc = 1.0, and
kB T = 1. Similarly, we only know that γ ij = 150 can generate
the low Reynolds number about 0.3, but cannot clearly iden-
tify the theoretical reason why it is set to be 150. These issues
take our attention away from DPD to SDPD when performing FIG. 8. Simulation of 500 RBCs in a microvascular network using the FVM cou-
RBC simulations. pling immersed boundary method (IBM) by Bagchi’s group.164 Reproduced with
permission from Balogh and Bagchi, “Direct numerical simulation of cellular-scale
With SDPD, our final goal is to study the motion, defor- blood flow in 3D microvascular networks,” Biophys. J. 113, 2815 (2017). Copyright
mation, and aggregation of a large number of RBCs in very 2017 Elsevier.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-20

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

devices.153 Currently, we have finished the simulations of a coupled with interface tracking techniques, e.g., phase-field,
single RBC in various microvessels, such as rectangular, cylin- level-set, and volume of fluids methods, was developed to
drical, curved, constricted, spiral, and bifurcated vessels, as simulate droplet deformation and bubble oscillation. How-
shown in Fig. 7. We have studied (i) the effects of mechani- ever, the major challenge for Eulerian methods is to simulate
cal properties, initial positions, and orientations of the RBC in the problems with topology changes, such as droplet breakup
rectangular and cylindrical microvessels,158,159 (ii) the devel- and coalescence. The SPH method with its Lagrangian nature
opment of secondary flow and Dean vortices in curved and has potential advantages in simulating interface and topol-
spiral microvessels,160 (iii) the behaviors on motion and defor- ogy changes. Relevant techniques in SPH have been devel-
mation for a RBC through a bifurcation,161 and (iv) a numeri- oped in last decades for the simulation of droplets and bub-
cally designed microfluidic chip to measure the cell’s mechan- bles. The major tasks include the prescription of surface ten-
ical properties by recording the transit time of the cell through sion, modeling discontinuities of density and pressure, and
the stenosed part of a constricted microvessel.162 It is worth large density ration of the gas-liquid system (or large vis-
mentioning from these studies that SDPD performs much bet- cosity ratio of emulsion) near the interface. The numeri-
ter than DPD in RBC simulations. First, it is easy to adjust the cal issues have been discussed in a recent review paper by
fluid incompressibility via EOS or the sound speed in EOS. This Wang et al.20
implies that it is easy but reasonable to avoid the nonuniform
The SPH method has been successfully applied in the sim-
distribution of particles and to prevent the appearance of the
ulation of single droplet deformation with the viscosity ratio
blank zone without particles. Second, there are specific phys-
close to unity82,83 or far from unity,80 single bubble defor-
ical parameters that can be conveniently set in simulations,
mation and rising with the gas to liquid density ratio down
such as length, density, and viscosity so that we can answer
to 0.001,73,79 bubble formation, growth and departure from
such questions clearly: Why are these parameters set to such
a submerged orifice.171 However, the aforementioned stud-
specific values? More importantly, we adopted the immersed
ies mainly focus on simple and quasi-steady systems, and
boundary method (IBM) to treat with the fluid-cell interac-
complex systems with a large amount of droplets (or bub-
tion, where the action of the cell on the fluid is implemented
bles) have not been widely considered within the framework
by converting the forces resulting from the deformation and
of SPH. The challenge for SPH (or particle-based methods) in
aggregation of the cell into a singular force acting on the fluid
the simulation of complex systems is to precisely modeling
neighboring the cell, while the action of the fluid on the cell
droplet-droplet or bubble-bubble interactions. As two single
is reflected by using the fluid velocity to update the cell posi-
droplets come close, a thin film between the two droplets
tion. This enables us to blur the distinction between the fluid
exerts a large hydrodynamic force (lubrication force) on the
and cell and thus allows us to pay more attention to the fluid
droplets, preventing the two droplets from coming closer.
dynamics. Besides, the other important feature of IBM, that
With the particle-based method, there may not be enough
is, separation of solving the fluid dynamics and calculating
particles between the two droplets to prescribe the effects of
the RBC force (generally, deformation and aggregation forces),
hydrodynamic force of the thin film, and therefore the hydro-
makes its implementation easy and straightforward. On bal-
dynamic interactions are often underestimated. To overcome
ance, SDPD, as a particle-based method, is one of the easy
this, several approaches were developed recently in DPD,
and flexible methods to model complex structure fluids, and
SDPD, and also SPH. Bian et al.172 employed the analytical
meanwhile, it couples the advantages of both SPH and DPD,
lubrication force of Stokes flow and corrected the hydro-
having specific physical meaning like SPH and thermal fluctu-
dynamic interaction of two equal-sized rigid spheres in
ations like DPD. IBM is a preferred method for handling fluid-
their SDPD simulation and successfully predicted the shear-
structure interaction problems, especially in biological fluid
thickening behavior of particulate suspension. Pan et al.127
dynamics.163 Therefore, the combination of SDPD and IBM,
applied an additional pairwise repulsive force between each
abbreviated as SDPD-IBM,159 may be a worthwhile choice for
pair of DPD particles from different droplets calibrating the
consideration, if a numerical method is to be recommended
force amplitude with experimental data and predicted the rhe-
for modeling cellular-scale blood flow in arbitrary and highly
ological properties of the emulsion system with more than
complex geometry.126
hundreds of droplets. However, these two approaches are still
limited, the analytical lubrication approach is restricted in the
B. Microfluidics and droplet dynamics Stokes flow regime, and the pairwise repulsive force approach
Droplets (and bubbles)165–167 are the basic units in micro- is relied on experimental data. A more general approach was
scale or meso-scale systems (microfluidics). Droplets have recently proposed by Rahmat and Yildiz173 with SPH simu-
been used as the discrete chemical and biological reactors and lation. In their simulation, a film drainage model based on
have been applied in biomedical engineering, and bubbles have Reynolds lubrication theory174 was applied by solving sim-
been used to enhance contrast in ultrasound images, non- plified Navier-Stokes equations of the thin film between two
invasive therapy, and drug delivery with sonoporation.168,169 droplets. Yang and Kong166 simulated the process of a drop
Numerous microfluidic devices were developed to produce impacting on a dry solid surface by SPH with a new sur-
monodispersed droplets and bubbles.170 Besides, the dynamic face tension model. They also simulated the process of a fuel
behavior of droplets (and bubbles) would play important drop impacting on heated surfaces based on SPH with a phase
roles in microfluidics. A traditional Eulerian numerical method change model.165

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-21

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

FIG. 11. Particle distributions at different instants by SPH simulation of slosh-

ing.198 Reproduced with permission from Shao et al., “An improved SPH method
for modeling liquid sloshing dynamics,” Comput. Struct. 100, 18 (2012). Copyright
2012 Elsevier.

Lagrangian nature. Hu and Admas82 first simulated the equi-

librium triple-phase contact angle by using SPH coupled with
FIG. 9. Static contact angles of droplets on hydrophilic rectangular [(a)–(c)] and
dual-rectangular [(d)–(f)] surfaces. A pairwise force SPH simulation work by Shig- a momentum conservative approach. Recently, Breinlinger
orina et al.180 Reproduced with permission from Shigorina et al., “Smoothed et al.88 proposed a smoothed normal correction at the triple
particle hydrodynamics study of the roughness effect on contact angle and droplet line region to ensure smooth transition from prescribed to
flow,” Phys. Rev. E 96, 033115 (2017). Copyright 2017 American Physical Society. actually computed wall normal. Huber et al.178 developed the
contact line force model to analyze dynamical droplet shapes
with advancing and receding contact angles. Yeganehdoust
A recent progress of SPH application in microfluidics that et al.179 suggested an additional correction procedure of the
has not been discussed in previous review papers is the SPH color function (which is used for surface tension modeling)
simulation of contact angle dynamics and wetting phenomena. of the interface particles leading to an accurate computation
The contact angle (and contact line) is ubiquitous in nature
and industry, which has attracted a lot of attention in last few
decades.61,175,176 A typical example of the contact angle prob-
lem is a liquid droplet on a solid substrate, and there are three
different phases, i.e., solid-liquid, liquid-gas, and gas-solid,
which need to be considered. Numerical simulations with tra-
ditional Eulerian methods have been recently reviewed by Sui
et al.177 Meanwhile, SPH also has its advantages in simulat-
ing this kind of multiphase and multiscale problems, due to its

FIG. 10. Droplet impacts on an elasto-plastic substrate, a numerical simulation FIG. 12. Wave heights obtained by SPH simulation and experiment of slosh-
work by Li et al.181 Reproduced with permission from Li et al., “A smoothed ing.198 Reproduced with permission from Shao et al., “An improved SPH method
particle hydrodynamics framework for modelling multiphase interactions at meso- for modeling liquid sloshing dynamics,” Comput. Struct. 100, 18 (2012). Copyright
scale,” Comput. Mech. 62, 1071 (2018). Copyright 2018 Springer Nature. 2012 Elsevier.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-22

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

FIG. 13. Sketch of the 2D standing waves: tank model (left)

and initial condition (right).66 Reproduced with permission
from Zhang et al., “SPH method with applications of oscil-
lating wave surge converter,” Ocean Eng. 152, 273 (2018).
Copyright 2018 Elsevier.

FIG. 14. Time history of the total kinetic energy predicted by SPH with different kernel functions.66 The damping coefficient β = 4νk 2 depends on the wave number k and
the kinematic viscosity ν. (a) β = 0.3432. (b) β = 0.1648. Reproduced with permission from Zhang et al., “SPH method with applications of oscillating wave surge converter,”
Ocean Eng. 152, 273 (2018). Copyright 2018 Elsevier.

of the color function gradient. Tartakovsky and Panchenko90 collisions over an inclined surface,179 and three-phase flow
developed a pairwise SPH formulation and derived the rela- in porous media.90 More recently, a successful application of
tionship between the parameters in pairwise forces and the SPH on contact angle dynamics was conducted by Shigorina
surface tension and static contact angle.
Along with these relevant developments, SPH has also
been applied to the simulation of pinning of moving liquid
droplet at various geometries,88 droplet motion and droplet

FIG. 16. The time history of the wave elevation η at x/d w = 18 recorded by SPH.66
FIG. 15. The sketch of the numerical wave tank.66 Reproduced with permis- Reproduced with permission from Zhang et al., “SPH method with applications of
sion from Zhang et al., “SPH method with applications of oscillating wave surge oscillating wave surge converter,” Ocean Eng. 152, 273 (2018). Copyright 2018
converter,” Ocean Eng. 152, 273 (2018). Copyright 2018 Elsevier. Elsevier.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-23

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

FIG. 17. Snapshots of Rayleigh–Taylor instability at

t = 1s, 3s, and 5s (from the top to bottom).85 Left:
phase snapshot; right: pressure field (the black dot-
ted line indicates the interface between the two fluids
obtained from the left snapshot). Reproduced with per-
mission from Chen et al., “An SPH model for multiphase
flows with complex interfaces and large density differ-
ences,” J. Comput. Phys. 283, 169 (2015). Copyright 2015
Elsevier.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-24

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

FIG. 18. Simulation pictures at a typ-

ical instant: top: single-phase simula-
tion; bottom: multiphase simulation.85
Reproduced with permission from Chen
et al., “An SPH model for multi-
phase flows with complex interfaces and
large density differences,” J. Comput.
Phys. 283, 169 (2015). Copyright 2015
Elsevier.

et al.,180 and they used the pairwise force SPH model to study C. Non-Newtonian fluid flows
the effect of surface roughness and the microscopic contact A non-Newtonian fluid is a fluid that does not follow New-
angle on the macroscopic apparent contact angle. Typical sim- ton’s law of viscosity. The viscosity of non-Newtonian fluids
ulation results are shown in Fig. 9. Their simulations show is dependent on the shear rate or shear rate history. There-
that micro-scale hydrophobic rough surfaces exhibit effective fore, the non-Newtonian fluids also belong to the category of
hydrophobic behavior, and micro-scale hydrophilic surfaces complex fluid that has been introduced in Sec. V B. As men-
may be macroscopically hydrophilic or hydrophobic, depend- tioned in Sec. V B, there are two major approaches in mod-
ing on the type of roughness. Another application that is eling non-Newtonian fluid, i.e., the continuum approach and
worthy of special emphasis is the work by Li et al.;181 they the microstructure approach. The continuum approach, with
modeled the solid substrate as an elasto-plastic material, and an established constitutive model, is used to study complex
a multiphase-fluid-structure interaction system is developed flow patterns of non-Newtonian fluids. It is usually applied
in their SPH simulation as shown in Fig. 10. by top-down modeling methods, such as the traditional

FIG. 19. Oil spill containment without (left) and with (right)
wave effects.205 The green, red, and black particles rep-
resent water, oil, and boom, respectively. Reproduced with
permission from Yang and Liu, “Numerical modeling of oil
spill containment by boom using SPH,” Sci. China Phys.,
Mech. Astron. 56, 315 (2013). Copyright 2013 Springer
Nature.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-25

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

FIG. 20. Simulation results of the water

exit of a horizontal cylinder with com-
parisons among the results from other
sources at different time instants: (a) T =
0.0, (b) T = 0.2, (c) T = 0.4, (d) T = 0.6, (e)
T = 0.8, (f) T = 1.0, (g) T = 2.0, and (h) T
= 3.0. Left: numerical results by Lin (solid
line),217 numerical results by Greenhow
and Moyo (dashed line),216 and theoret-
ical results by Tyvand and Miloh (dotted
line).218 Right: SPH results.215 Repro-
duced with permission from Liu et al.,
“An SPH model for free surface flows
with moving rigid objects,” Int. J. Numer.
Methods Fluids 74, 684 (2014). Copy-
right 2014 John Wiley and Sons.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-26

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

FIG. 20. (Continued.)

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-27

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

Eulerian simulation methods and also the SPH method. stress model used in elastic solid simulation was applied and
While the microstructure approach mainly focuses182 on the it works well for viscoelastic fluid. In addition, the artificial
rheological properties of complex fluids by modeling the viscous term was also added in their simulation, which was
microstructures in a small scale. It is usually applied by also adopted by the following studies.190,191 For the model-
bottom-up modeling methods, such as DPD. In this subsec- ing of power-law fluids and viscoplastic fluids, a shear-rate
tion, we mainly focus on the continuum approach used in dependent viscosity is considered. Therefore, in SPH sim-
SPH that is applied to study the complex flow patterns of ulation, only the viscosity term is modified accordingly as
non-Newtonian fluids. reported in the literature. An improvement was reported by
Fan et al.187 for the simulation of high viscosity molding
SPH simulation of non-Newtonian fluids was first
flow. To retain the stability and efficiency of the SPH sim-
reported by Ellero et al.122 in 2002. Later on, SPH has been
ulation, an implicit time integration scheme was employed
widely applied to the simulations of viscoelastic fluids (e.g.,
in their simulation and the fluid viscosity was as high as
Oldroyd-B and Maxwell models),120,121,183–185 power law flu-
O(104 ) Pa s.
ids,70,186,187 and viscoplastic fluids (e.g., Herschel-Bulkley and
Bingham models).188,189 Ellero and Tanner184 applied SPH to With these developments, the SPH method has been
simulate viscoelastic fluid by solving an additional Maxwell applied in the studies on mudflow or debris flow,188,192
type equation of the elastic stress [similar as Eq. (106)] with solid particle suspended in shear flow of Oldroyd-B fluid,193
the same procedure for solving momentum equations in SPH. polymer melt and rod-climbing phenomenon caused by non-
Fang et al.120 applied a similar approach to simulate the zero normal stress difference,194 power-law fluid through
free surface problem (falling droplet towards a rigid wall) porous media,186 and 3D printing process.195 More recently,
of the Oldroyd-B fluid, and they found unrealistic fracture Pan et al.196 employed Glen’s constitutive law197 to model
and unphysical particle clustering in their simulation caused ice with SPH and investigated the ground line dynam-
by the tensile instability. In their simulation, the artificial ics of ice sheets. In view of this, SPH is applicable for

FIG. 21. SPH simulation snapshots219 and experimental

observations220 for a dam break with an elastic gate. (a)
T = 0.0s, (b) T = 0.04s, (c) T = 0.08s, (d) T = 0.12s, (e) T
= 0.16s, (f) T = 0.20s, (g) T = 0.24s, and (h) T = 0.32s.
Reproduced with permission from Liu et al., “Numerical
simulation of hydro-elastic problems with smoothed parti-
cle hydro-dynamics method,” J. Hydrodyn. 25, 673 (2013).
Copyright 2013 Springer Nature.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-28

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

the simulation of non-Newtonian fluids and can be further D. Free surface flows
extended to the simulation of complex material for practical In grid-based methods, the free surface can be precisely
applications. detected and a free surface condition (dynamic condition)

FIG. 22. Pressure distributions at different instants obtained using FPM for simulations of the interaction between solitary waves and the rigid plate.75 Reproduced with
permission from Huang et al., “Coupled finite particle method for simulations of wave and structure interaction,” Coastal Eng. 140, 147 (2018). Copyright 2018 Elsevier.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-29

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

should be imposed. In contrast, the classic SPH implicitly sat-

isfies two boundary conditions (kinematic and dynamic con-
ditions) of the free surface flows. Thus, SPH is comparatively
simpler in numerical implementation and more suitable for
flows with free surface than grid-based methods. Here, three
representative cases, including sloshing, standing waves, and
wave propagating, are introduced.
Sloshing is a process of violent fluid motions inside a
tank. The reciprocal movement of the water tank drives
the contained water to move front and back in the tank.
Figure 11198 shows the flow pattern of sloshing at nine typi-
cal time instants predicted by SPH when the liquid in a rect-
angular tank is excited under a periodic horizontal force.
Figure 12198 shows the wave heights at the probe point
obtained by the SPH simulation and experiment. The SPH
results agree well with experimental observations, in pattern,
period, and amplitude.
The standing wave is a typical free surface flow. For vis-
cous fluids, the total energy of fluid particles goes to zero and
the standing wave is completely damped for long time evo-
lution. The standing wave can be used to test the numerical
accuracy as its analytical solution of the total kinetic energy
exists. As shown in Fig. 13, the tank of the 2D standing waves
is shown. The water with the depth dw has an initial veloc-
ity which is defined by the potential theory.199 Figure 1466
shows the time history of the total kinetic energy predicted
by SPH with different kernel functions. The results by Gaus- FIG. 23. The wave forces acting on the rigid plate by FPM simulations, in com-
sian and quintic spline kernel functions are in good agreement parison with the experimental result221 (top: horizontal force, bottom: vertical
with analytical solutions, whereas cubic and quartic spline force).75 Reproduced with permission from Huang et al., “Coupled finite particle
functions cannot produce similar results. Therefore, the selec- method for simulations of wave and structure interaction,” Coastal Eng. 140, 147
(2018). Copyright 2018 Elsevier.
tion of the kernel function is critical for accuracy of the SPH
method.
Wave propagating and breaking are significantly impor- Rayleigh–Taylor instability (RTI), dam breaking, and oil spill
tant in ocean and costal engineering. Figure 15 shows the containment by boom, to show the great potential ability of
2D numerical wave tank. The time profiles of water sur- SPH in modeling the multiphase flows.
face elevation at x/dw = 18 are collected. Figure 1666 shows
the time profiles of water surface elevation η at x/dw = 18 Rayleigh–Taylor instability (RTI) is an instability that ini-
predicted by SPH with different spatial resolutions, in compar- tiates and develops at the interface between two horizon-
ison with the solutions of Mixed-Eulerian-Lagrangian Bound- tal parallel fluids with different densities and viscosities.
ary Element Method (BEM-MEL).200 With the increase of the
resolution, the wave elevation by SPH tends to the conver-
gent solution of the BEM-MEL signal. The results show the
good convergence properties in SPH simulations of water
waves.

E. Multiphase flows
As mentioned in Sec. IV B, SPH can naturally track
the interface in simulations of multiphase flows, which is
quite difficult for grid-based methods because of requiring
special algorithms to track the interface. Hence, SPH mod-
eling of multiphase flows has also attracted a lot of
researchers.13,79,201–203 For example, Yang and Kong202 used FIG. 24. The pressure counter at t = 6.87s by FPM for 3D simulations of waves
the SPH method to simulate the evaporating multiphase flows impacting on the OWSC.75 Reproduced with permission from Huang et al., “Cou-
and studied the heat and mass transfers across the liquid- pled finite particle method for simulations of wave and structure interaction,”
Coastal Eng. 140, 147 (2018). Copyright 2018 Elsevier.
gas interfaces. We here review three typical multiphase flows,

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-30

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

Figure 1785 shows the phase snapshots and the pressure fields multiphase simulation at a typical instant. It can be observed
at different instants obtained by SPH. It can be found that that the backward plunging breaker in multiphase simula-
SPH can successfully capture the complex interfaces between tion does not collapse quickly as that in single-phase simu-
the two different fluids, including the local separations and lation due to the existence of the air phase in the entrapped
discontinuities of the fluids over the interfaces. cavity.
Dam breaking is an important benchmark problem Oil spill containment, which involves water-oil two-phase
in hydrodynamic simulations, and it is usually associated flow, is significantly important in the protection of ocean
with a complex phenomenon, including high impact pres- environment from oil pollution.204 It is necessary to study
sure, surface break-up, and water jet. Figure 1885 shows the mechanism of boom failure in order to prevent oil from
the comparative pictures of single-phase simulation and spreading to wider areas. Figure 19205 shows the flow pattern

FIG. 25. Vertical velocity distribution dur-

ing the sedimentation process obtained
using the modified SPH (up) and
FE-FBM (down) at the same time
instants.37 Reproduced with permission
from Zhang et al., “Meshfree modeling
of a fluid-particle two-phase flow with an
improved SPH method,” Int. J. Numer.
Methods Eng. 116, 530 (2018). Copy-
right 2018 John Wiley and Sons.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-31

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

evolution of oil layer and the velocity field of water with and
without wave effects. There is no boom failure when water
wave is not considered. However, the spilled oil layer and
boom can move upwards and downwards with the advance-
ment of the water wave. Therefore, a more suitable design of
the boom is necessary in order to effectively contain oil spill in
an environment with water waves.

F. Fluid-structure interactions
Many studies have been conducted to investigate prob-
lems with fluid–structure interactions.206–213 The grid-based
methods encounter some difficulties when treating flows with
moving bodies or moving boundaries. In these models, the
moving structures in a Lagrangian formulation need to be
appropriately described in a grid, and the special treatment of
the moving boundary is inevitable. One of these approaches
is called the moving grid method,214 in which an adaptive
mesh is corrected or reconstructed during every time step.
The movement of the grid near the Lagrangian structure or
the communication between the moving Lagrangian struc-
ture and Eulerian meshes will increase the algorithm com-
plexity. As a fully Lagrangian particle model, the SPH method
is inherently well-suited for simulations of fluid–structure
interaction problems, and it has been widely applied into
various problems with moving bodies or a moving bound-
ary in past years. Here, we show several representative
examples.
Figure 20215 shows the snapshots of the SPH results
for simulations of the water exit of a horizontal cylinder, in
comparison with the existing numerical and theoretical
results. It is clear that the SPH results agree well with the
existing numerical results216,217 for both the free surface mor-
phology and position of the horizontal cylinder. All numerical
results are also close to the theoretical results218 at earlier
stages (before T = 0.4), but difference gradually becomes larger
as time marches to 0.6. After T = 0.6, the theoretical results
are no longer valid, and the free surface begins to breakup. It
is noted that the SPH method can not only well predict the
free surface morphology and position of the horizontal cylin-
der before its exit from water but also well describe the water
detachment, breakup, splash, and wave generation during the
exit of the cylinder from water.
Figure 21219 shows the SPH simulation snapshots and
the corresponding experimental observations for a dam break
with an elastic gate. The SPH simulation can well reproduce
this experimental phenomenon.220 It is shown that before
the dam break, both the water and the elastic gate are in a
still state. After the sudden removal of the clamp, under the
static water pressure, the elastic gate will deform and grad-
ually open. The contained water is discharged from the gate FIG. 26. Snapshots of the Al sphere impacting on the Al plate at dif-
and this leads to a large displacement of the elastic gate with ferent instants. From (a) to (f): (up) experimental results by Plassard et
a large water current out of the exit. al.233 and (down) results from SPH with KGC.232 Reproduced with per-
mission from Zhang and Liu, “Smoothed particle hydrodynamics with ker-
Figure 2275 shows the pressure distributions at different nel gradient correction for modeling high velocity impact in two- and three-
times by FPM for simulations of the interaction between soli- dimensional spaces,” Eng. Anal. Boundary Elem. 83, 141 (2017). Copyright 2017
Elsevier.
tary waves and the rigid plate, where FPM can successfully

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-32

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

FIG. 27. Welding interfaces obtained

from experimental observations234 and
present SPH simulations.208 (a)–(c)
are the experimental observations of
nickel-steel welding conducted at colli-
sion velocities of 1600 m/s, 1900 m/s,
and 2500 m/s, respectively. (d)–(f) are
the simulation results of steel-steel weld-
ing conducted at corresponding colli-
sion velocities. Reproduced with permis-
sion from Zhang et al., “Investigation of
explosive welding through whole process
modeling using a density adaptive SPH
method,” J. Manufact. Processes 35, 169
(2018). Copyright 2018 Elsevier.

capture the spray and splash in the vicinity of free surface. collision velocity increases [illustrated in Figs. 27(a)–27(c)].
Figure 2375 shows the time series of the total horizontal and The shape and position of the vortex, wave crest, and
vertical forces on the rigid plate, and the horizontal force wave trough [as shown in Figs. 27(d)–27(f)] by SPH simula-
predicted by FPM is in good agreement with experimental tion results are very close to those from the experimental
results.221 Figure 2475 shows a pressure counter at t = 6.87s observations.
by FPM for 3D simulations of waves impacting on an oscillat-
ing wave surge converter (OWSC) which is an effective con-
tender in the challenge to extract energy from waves.222–224 VII. CONCLUDING REMARKS
The simulation results confirm that FPM may well predict the
SPH is a truly meshfree, Lagrangian particle method,
development, change, and trend of the wave force acting on
and there is no need for a pre-defined or fixed connectivity
the structure.
between these particles. It uses particles with material prop-
Motions of solid particles in fluids have a wide range erties to represent the state of a system, and this set of parti-
of science and industrial fields including fluidized suspen- cles is employed as the computational frame for approximat-
sions,225 transport properties of colloidal particles,226 sedi- ing partial differential equations governing the fluid flow. As a
mentation on seabed,227,228 particle motion in a viscoplastic meshfree method, SPH does not have a mesh entangle prob-
fluid,229 and particle rotation driven by external flow fields.230 lem, compared with the Lagrangian finite element method. As
SPH shows good results for simulations of particulate two- a Lagrangian particle method, SPH can naturally track mov-
phase flows. Figure 2537 shows the vertical velocity distribu- ing features such as evolving free surfaces, moving material
tion during the sedimentation process obtained by a modified interfaces, and deformable solid boundaries, while these are
SPH and finite element fictitious boundary method (FE-FBM). challenging issues for Eulerian grid-based methods like FDM
The FE-FBM approach is based on a software package FEAT- and FVM. Furthermore, SPH shares some similar features to
FLOW231 which is well validated for particulate flow simula- MD and DPD and it is feasible to couple them together for
tions. The modified SPH can accurately simulate the sedimen- multi-scale applications or introduce some concepts of MD
tation process with a smoothed velocity field and a similar and DPD into SPH to model meso- or even micro-scale fluid
particle position as the FE-FBM simulation produces velocity flows. In summary, these special features make SPH attractive
and position. in modeling complex fluid flows, especially those with large
fluid deformation, fluid-structure interaction, and multi-scale
SPH is also attractive in modeling hydrodynamics with
physics.
complex material strength. For example, Fig. 26232 shows
the evolution snapshots of the Al sphere impacting on In this paper, we review the latest development of SPH
the Al plate at different time instants. It is seen that the for modeling complex fluid flows. Some important numerical
SPH method with KGC can well reproduce the experimen- aspects of the SPH are described including improved approx-
tal observations233 with main physics including the debris imation or discretization schemes, different particle regu-
cloud and rear plumes. Figure 27 shows the experimental234 larization techniques, and various boundary treatment algo-
and numerical208 snapshots of the welding interfaces in the rithms. All these are closely related to the computational accu-
processes of explosive welding, respectively. In the exper- racy of SPH when modeling complex fluid flows. Moreover,
iments, the interfacial morphology changes from straight we discuss the link between SPH and DPD and further intro-
to wavy and then to wavy with vortex shedding as the duce the SDPD for modeling complex fluids in meso-scales.

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-33

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

Then, we present some latest applications of SPH to mod- its derivatives. Unlike FPM, DFPM avoids the trouble-some
eling complex fluids, including biological flows, microfluidics matrix inversion and is more stable and efficient while pos-
and droplet dynamics, non-Newtonian fluid flows, free sur- sessing similar accuracy. Unlike the conventional SPH, DFPM
face and multiphase flows, flows with strong fluid-structure has much better accuracy while keeping the advantages of
interactions, and even material strength. flexibility, cost-effectiveness, and easiness in coding. There-
fore, DFPM leverages the high accuracy of FPM and the effi-
ciency and flexibility of the conventional SPH, and it should be
A. Some remarks on SPH approximation schemes
appealing in modeling complex fluid flows.
Despite its successful application in modeling complex
fluid flows, the conventional SPH has been hindered with low B. Some remarks on the particle
accuracy as it cannot exactly reproduce quadratic and lin- regularization techniques
ear functions and even cannot exactly reproduce a constant.
Particle regularization techniques including XSPH, ten-
The particle inconsistency is an important problem closely
sile correction, dynamics stabilization, and particle shifting
related to the accuracy of the SPH method. This problem orig-
can be helpful in modeling complex fluid flows especially for
inates from the SPH particle approximation process and is a
irregular particle distributions. It is important to note that
manifestation of the discrepancy between the spatially dis-
these particle regularization techniques may not be valid for
cretized equations and the corresponding kernel approxima-
every circumstance and also not all scenarios need particle
tions in continuous form. During the last decades, different
regularization. The first thing before applying a particle reg-
approaches have been proposed to restore the particle con-
ularization technique is to check when and where to conduct
sistency and hence to improve the SPH approximation accu-
the particle regularization, i.e., to establish a rule to eval-
racy. Some of them involve reconstruction of a new smooth-
uate the disorderedness of particle distribution. The fact is
ing function so as to satisfy the discretized consistency con-
that most particle regularization techniques (e.g., XSPH, TC,
ditions. One other more flexible approach is to generate a
and DS) do not check disorderedness of particle distribu-
correction matrix for the conventional SPH method without
tion while simply applying a particle regularization technique
changing the conventional smoothing function.
everywhere is not necessary and may even produce additional
In this paper, we discussed some widely used SPH errors.
approximation schemes including the conventional SPH,
A simple but effective way is to check the constant and
CSPH, KGC, FPM, KGF-SPH, and DFPM. These SPH approx-
linear consistency conditions. For SPH approximations, if par-
imation schemes in general originate from the Taylor series
ticles are evenly distributed, the following two equations can
analysis of the SPH kernel approximation of a field function
be satisfied:
and its derivatives, while different treatments lead to differ-
N
ent modified versions. FPM can be regarded as the improved X
version of both SPH and CSPM. If only the gradients of a W(x − xj , h)∆vj = 1 (109)
j=1
field function are concerned, FPM can be converted to KGC,
which is also straightforward and popular in modeling com- and
plex fluid flows. DFPM is a decoupled FPM without matrix N
X
inversion. KGF-SPH involves replacing ∇W with (rj − ri )W in (x − xj )W(x − xj , h)∆vj = 0. (110)
the SPH approximation so as to avoid directly using the kernel j=1
gradient, which sometimes can be troublesome. For irregular particle distributions, these two dis-
It is important to note in FPM, KGF-SPH, KGC, and CSPM, cretized constant and linear consistency conditions are
 N 
some kind of corrective matrix is used to restore particle not necessarily satisfied. As such,  W(x − xj , h)∆vj − 1 and
P
consistency so as to improve approximation accuracy. These  j=1 
approaches put more emphasis on the consistency (hence  N 
 P (x − xj )W(x − xj , h)∆vj  can be used separately or simulta-
hopefully accuracy), but the stability sometimes can be in  j=1 
question. For example, for complex fluid flows with singu-
neously to evaluate the disorderedness of particle distribu-
lar particle distributions, these methods may suffer from ill-
tion. If these two values exceed some pre-defined threshold
conditioned corrective matrices, which may further lead to
value (i.e., 0.1), particle regularization techniques should be
numerical instability or unexpected termination of the sim-
applied.
ulation. The question is that can we have both the consistency
and the stability at the same time? The answer is yes, pro- One other thing is that we must be cautious when apply-
vided we are willing to balance the accuracy and stability. ing a particle regularization technique. For complex fluid flows
The DFPM is such a solution, which is obtained through Tay- with material interfaces or discontinuities (e.g., cracks and
lor series analyses on integration approximation and assum- shocks), applying a particle regularization technique over the
ing diagonal dominance of the resultant corrective matrix.36 entire domain may lead to “smoothed” rather than “physi-
As DFPM is a decoupled finite particle method, it does not cal” results along the interface area. That is why Lind et al.54
need to solve redundant and challenging pointwise correc- and Khayyer et al.56 developed their particle shifting tech-
tive matrix equations when approximating a field variable and niques for free surface, which is a special type of material

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-34

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

interface. When a complex fluid flow involves the fluid inter- SPH. This treatment involves Eulerian-Lagrangian dual com-
face, it is feasible to apply particle regularization techniques putational frames, mesh frames for solving PDEs, and par-
locally in the separate domain with a single fluid phase rather ticle frames for tracking moving features. Typical examples
than globally in the whole domain with different fluid phases. include the particle-in-cell (PIC) method243 and the mate-
This is somewhat similar to the application of improved SPH rial point method244 (MPM). One other way is to intro-
approximation schemes for problems with material interface duce some good features from the other particle method
of discontinuity-like shock waves. Just as pointed out by Liu et into SPH, thus leading to the SPH method for various
al., when applying modified SPH approximations like CSPM to physics. For example, by introducing thermal fluctuations as
model discontinuities like shock wave over the entire domain, in MD and DPD into SPH, Español and Revenga25 developed
the discontinuities can be unreasonably averaged, thus ignor- SDPD for modeling complex fluids in meso-scale. Bui et al.
ing real physics.3,235 developed an SPH model for large deformation and failure
flows of geo-material through introducing the elastic-plastic
soil constitutive model, which is frequently used in DEM
C. Combined approaches with SPH
simulations.183
In principle, as a particle-based discretization technique,
SPH can be used to approximate any PDE to obtain the cor- D. Multi-scale modeling of complex fluids with SPH
responding SPH formulation and to model the evolution of
In Sec. V A, we introduced the dissipative particle dynam-
inherent physics governed by the PDE. SPH, as a meshfree,
ics and smoothed dissipative particle dynamics within the SPH
Lagrangian particle method, is attractive in modeling com-
framework. We note that
plex fluid flows with large deformations and moving inter-
faces. However, the accuracy of the SPH method is not easy • SPH was proposed initially for simulations of astro-
to be enhanced to more than 2nd order and it is usually dif- physical problems at cosmoscale.10,11,14
ficult to have both high accuracy and high stability at the • SPH was modified to solve problems of viscous flow
same time. Treatment of boundaries in particle-based meth- and thermal conduction at macro-scale.93–96
ods in general is not as accurate, easy, and straightforward • SPH was extended to solve complex flows with thermal
as grid-based methods. Moreover, particle-based methods fluctuations at meso-scale, named SDPD.25
are in general computationally more expensive than grid- • SDPD can be either a mesoscopic or a macroscopic
based methods as the interaction between particle pairs usu- flow solver. For macroscopic flows, SDPD is just a
ally increases drastically when increasing the total num- particular, thermodynamically consistent version of
ber of particles. As each numerical method has special fea- SPH.111
tures with specific advantages and disadvantages, it would • SDPD can be considered as the final outcome of the
be beneficial to combine SPH with other numerical methods endeavors to improve DPD, grouping all the partial
(referred to herein as coupling approaches) or to introduce solutions to the issues mentioned in Sec. V A 2.108
some good features from other methods into SPH (referred to
Hence, SPH seems to be attractive in multi-scale model-
as hybrid approaches), so as to leverage the advantages of each
ing of complex fluids. Along the path that SPH was modified to
method.
the macro-scale and extended to the meso-scale, a question
As to the coupling approaches, since it is convenient for is straightforward whether it can be further extended to the
grid-based methods to treat boundaries, it is natural to cou- micro-scale, in an elegant way like SPH to SDPD, for example,
ple SPH with grid-based methods like FEM, FVM, or FDM. For introducing one more force in the momentum equation (95) or
example, SPH was coupled with FEM for fluid structure inter- revising the existing forces.
action (FSI) problems in which SPH is used to model fluid flows
If the answer is yes, a multi-scale SPH scheme will be gen-
and FEM is used to model solid dynamics.236,237 SPH-FEM
erated naturally, abbreviated here as MSPH for convenience.
coupling was also applied to problems with extremely local
Thus, a scene in science fiction might come true by MSPH eas-
large deformations in which SPH is used to model regions with
ily, that one in the galaxy might predict behaviors of a molecule
large deformation and FEM is used for other regions.238,239
in a human on earth, where only particles are used and even
One other trend is to couple SPH with other particle-based
their governing equations do not need changing. From this
methods for different purposes. For example, SPH was cou-
perspective, we believe that the potential MSPH may be very
pled with the element bending group (EBG) to model flexible
competitive in multiscale simulations, compared with other
fibers interacting with viscous fluids, while SPH is used to
particle-based methods, not to mention grid-based methods
model the fluid flow and EBG is used to model the move-
that may be complicated or even unreachable in micro and
ment and deformation of flexible fibers.211,212 SPH was also
cosmoscales. Note that we here emphasize the elegant way
coupled with the discrete element method (DEM)240 for sim-
by which SDPD is extended to the micro-scale. From SPH to
ulating environmental flows in which SPH is used for mod-
SDPD, only a random force FR is added into the momentum
eling the fluid flows and DEM is used to model granular
equation, and SDPD can also replace SPH at the macro-scale.
flows.241,242
From SDPD to the microscopic approach, one also expects
As to the hybrid approaches, one most popular way is that only one more force is added or the existing forces are
to introduce a background mesh into particle methods like revised, and the microscopic SDPD can replace the original

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-35

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

SDPD at both the meso- and macro-scales. Unfortunately, no 15

J. J. Monaghan, “Smoothed particle hydrodynamics,” Rep. Prog. Phys. 68,
such extension has been done so far, although several multi- 1703 (2005).
16
scale SDPD methods have been developed by coupling SDPD D. Violeau and B. D. Rogers, “Smoothed particle hydrodynamics (SPH)
for free-surface flows: Past, present and future,” J. Hydraul. Res. 54, 1
with MD.245–247 These coupling multiscale SDPD methods are
(2016).
still complicated and not straightforward, requiring MD in the 17
H. Gotoh and A. Khayyer, “On the state-of-the-art of particle methods for
microscopic region, SDPD in the mesoscopic region, and the coastal and ocean engineering,” Coastal Eng. J. 60, 79 (2018).
combination of MD and SDPD in the transition region. Hence, 18
A. M. Zhang, P. Sun, F. Ming, and A. Colagrossi, “Smoothed particle hydro-
an extension from SDPD to the micro-scale is still an open dynamics and its applications in fluid-structure interactions,” J. Hydrodyn.
problem. 29, 187 (2017).
19
M. S. Shadloo, G. Oger, and D. Le Touze, “Smoothed particle hydrody-
namics method for fluid flows, towards industrial applications: Motivations,
current state, and challenges,” Comput. Fluids 136, 11 (2016).
ACKNOWLEDGMENTS 20
Z. B. Wang, R. Chen, H. Wang, Q. Liao, X. Zhu, and S.-Z. Li, “An overview
The work was financially supported by the of smoothed particle hydrodynamics for simulating multiphase flow,” Appl.
Math. Modell. 40, 9625 (2016).
National Natural Science Foundation of China (Grant Nos. 21
M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids (Oxford
51779003, U1530110, 11172306, and 10942004). Some of the
University Press, 1987).
simulations were conducted at Tianhe II Super Computer in 22
P. J. Hoogerbrugge and J. Koelman, “Simulating microscopic hydrody-
the National Supercomputing Center in Guangzhou with the namic phenomena with dissipative particle dynamics,” Europhys. Lett. 19,
effective support from Beijing Paratera Technology Co., Ltd. 155 (1992).
Special thanks are given to Dr. Muhammad Usman and Dr. 23
M. B. Liu, P. Meakin, and H. Huang, “Dissipative particle dynamics with
Zhilang Zhang for reviewing the paper and giving valuable attractive and repulsive particle-particle interactions,” Phys. Fluids 18,
comments. 017101 (2006).
24
P. Espanol, “Fluid particle dynamics: A synthesis of dissipative parti-
cle dynamics and smoothed particle dynamics,” Europhys. Lett. 39, 605
REFERENCES (1997).
25
P. Español and M. Revenga, “Smoothed dissipative particle dynamics,”
1
M. B. Liu, G. R. Liu, L. W. Zhou, and J. Z. Chang, “Dissipative particle dynam- Phys. Rev. E 67, 026705 (2003).
ics (DPD): An overview and recent developments,” Arch. Comput. Methods 26
P. W. Randles and L. D. Libersky, “Smoothed particle hydrodynamics:
Eng. 22, 529 (2015). Some recent improvements and applications,” Comput. Methods Appl.
2
O. C. Zienkiewicz and R. L. Taylor, The Finite Element Method (McGraw-Hill, Mech. Eng. 139, 375 (1996).
2000). 27
J. K. Chen and J. E. Beraun, “A generalized smoothed particle hydrody-
3
G. R. Liu and M. B. Liu, Smoothed Particle Hydrodynamics: A Mesh- namics method for nonlinear dynamic problems,” Comput. Methods Appl.
free Particle Method (World Scientific Publishing Co. Pte. Ltd., Singapore, Mech. Eng. 190, 225 (2000).
2003). 28
M. B. Liu, W. P. Xie, and G. R. Liu, “Modeling incompressible flows using a
4
C. W. Hirt and B. D. Nichols, “Volume of fluid (VOF) method for the finite particle method,” Appl. Math. Modell. 29, 1252 (2005).
dynamics of free boundaries,” J. Comput. Phys. 39, 201 (1981). 29
M. B. Liu and G. R. Liu, “Restoring particle consistency in smoothed
5
M. Sussman, P. Smereka, and S. Osher, “A level set approach for comput- particle hydrodynamics,” Appl. Numer. Math. 56, 19 (2006).
ing solutions to incompressible two-phase flow,” J. Comput. Phys. 114, 146 30
J. Bonet and T.-S. L. Lok, “Variational and momentum preservation
(1994). aspects of smooth particle hydrodynamic formulations,” Comput. Methods
6
S. Koshizuka and Y. Oka, “Moving-particle semi-implicit method Appl. Mech. Eng. 180, 97 (1999).
for fragmentation of incompressible fluid,” Nucl. Sci. Eng. 123, 421 31
M. B. Liu, G. R. Liu, and K. Y. Lam, “Constructing smoothing functions in
(1996). smoothed particle hydrodynamics with applications,” J. Comput. Appl. Math.
7
G. T. Duan and B. Chen, “Large eddy simulation by particle method coupled 155, 263 (2003).
with sub-particle-scale model and application to mixing layer flow,” Appl. 32
C. Huang, J. M. Lei, M. B. Liu, and X. Y. Peng, “A kernel gradient free (KGF)
Math. Modell. 39, 3135 (2015). SPH method,” Int. J. Numer. Methods Fluids 78, 691 (2015).
8
M. B. Liu and G. R. Liu, “Smoothed particle hydrodynamics (SPH): An 33
C. Huang, J. M. Lei, M. B. Liu, and X. Y. Peng, “An improved KGF-SPH with
overview and recent developments,” Arch. Comput. Methods Eng. 17, 25 a novel discrete scheme of Laplacian operator for viscous incompressible
(2010). fluid flows,” Int. J. Numer. Methods Fluids 81, 377 (2016).
9
A. Di Mascio, M. Antuono, A. Colagrossi, and S. Marrone, “Smoothed parti- 34
G. M. Zhang and R. C. Batra, “Symmetric smoothed particle hydrodynam-
cle hydrodynamics method from a large eddy simulation perspective,” Phys. ics (SSPH) method and its application to elastic problems,” Comput. Mech.
Fluids 29, 035102 (2017). 43, 321 (2009).
10
L. B. Lucy, “Numerical approach to testing of fission hypothesis,” Astron. J. 35
T. Jiang, Z. C. Chen, W. G. Lu, J. Y. Yuan, and D. S. Wang, “An effi-
82, 1013 (1977). cient split-step and implicit pure mesh-free method for the 2D/3D
11
R. A. Gingold and J. J. Monaghan, “Smoothed particle hydrodynamics: The- nonlinear Gross–Pitaevskii equations,” Comput. Phys. Commun. 231, 19
ory and application to non-spherical stars,” Mon. Not. R. Astron. Soc. 181, 375 (2018).
(1977). 36
Z. L. Zhang and M. B. Liu, “A decoupled finite particle method for mod-
12
J. J. Monaghan, “Smoothed particle hydrodynamics and its diverse appli- eling incompressible flows with free surfaces,” Appl. Math. Modell. 60, 606
cations,” Annu. Rev. Fluid Mech. 44, 323 (2012). (2018).
13
F. R. Ming, P. N. Sun, and A. M. Zhang, “Numerical investigation of ris- 37
Z. L. Zhang, K. Walayat, J. Z. Chang, and M. B. Liu, “Meshfree modeling of a
ing bubbles bursting at a free surface through a multiphase SPH model,” fluid-particle two-phase flow with an improved SPH method,” Int. J. Numer.
Meccanica 52, 2665 (2017). Methods Eng. 116, 530 (2018).
14
J. J. Monaghan, “Smoothed particle hydrodynamics,” Annu. Rev. Astron. 38
J. P. Morris, P. J. Fox, and Y. Zhu, “Modeling low Reynolds number
Astrophys. 30, 543 (1992). incompressible flows using SPH,” J. Comput. Phys. 136, 214 (1997).

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-36

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

39 62
J. J. Monaghan, “Simulating free surface flows with SPH,” J. Comput. Phys. M. B. Liu, J. R. Shao, and J. Z. Chang, “On the treatment of solid boundary
110, 399 (1994). in smoothed particle hydrodynamics,” Sci. China Technol. Sci. 55, 244 (2012).
40 63
S. J. Cummins and M. Rudman, “An SPH projection method,” J. Comput. B. D. Rogers and R. A. Dalrymple, “SPH modeling of tsunami waves,” in 3rd
Phys. 152, 584 (1999). International Workshop on Long-wave Runup Models (Catalina, 2008).
41 64
A. Rafiee and K. P. Thiagarajan, “An SPH projection method for simulating J. J. Monaghan and J. B. Kajtar, “SPH particle boundary forces for arbitrary
fluid-hypoelastic structure interaction,” Comput. Methods Appl. Mech. Eng. boundaries,” Comput. Phys. Commun. 180, 1811 (2009).
198, 2785 (2009). 65
Y. W. Han, H. F. Qiang, J. L. Zhao, and W. R. Gao, “A new repulsive model for
42
M. Antuono, A. Colagrossi, S. Marrone, and D. Molteni, “Free-surface flows solid boundary condition in smoothed particle hydrodynamics,” Acta Phys.
solved by means of SPH schemes with numerical diffusive terms,” Comput. Sin. 62, 044702 (2013).
Phys. Commun. 181, 532 (2010). 66
D. H. Zhang, Y. X. Shi, C. Huang, Y. L. Si, B. Huang, and W. Li, “SPH method
43
S. Marrone, M. Antuono, A. Colagrossi, G. Colicchio, D. Le Touze, and with applications of oscillating wave surge converter,” Ocean Eng. 152, 273
G. Graziani, “δ-SPH model for simulating violent impact flows,” Comput. (2018).
Methods Appl. Mech. Eng. 200, 1526 (2011). 67
P. N. Sun, A. Colagrossi, S. Marrone, M. Antuono, and A. M. Zhang, “Multi-
44
D. Le Touze, A. Colagrossi, G. Colicchio, and M. Greco, “A critical inves- resolution Delta-plus-SPH with tensile instability control: Towards high
tigation of smoothed particle hydrodynamics applied to problems with Reynolds number flows,” Comput. Phys. Commun. 224, 63 (2018).
free-surfaces,” Int. J. Numer. Methods Fluids 73, 660 (2013). 68
M. S. Shadloo, A. Zainali, M. Yildiz, and A. Suleman, “A robust weakly com-
45
M. Antuono, B. Bouscasse, A. Colagrossi, and S. Marrone, “A measure of pressible SPH method and its comparison with an incompressible SPH,” Int.
spatial disorder in particle methods,” Comput. Phys. Commun. 185, 2609 J. Numer. Methods Eng. 89, 939 (2012).
(2014). 69
T. Long, D. Hu, D. Wan, C. Zhuang, and G. Yang, “An arbitrary bound-
46
J. J. Monaghan, “Smoothed particle hydrodynamics,” J. Comput. Phys. 82, ary with ghost particles incorporated in coupled FEM-SPH model for FSI
1 (1989). problems,” J. Comput. Phys. 350, 166 (2017).
47 70
S. Shahriari, I. G. Hassan, and L. Kadem, “Modeling unsteady flow char- S. D. Shao and E. Y. M. Lo, “Incompressible SPH method for simulat-
acteristics using smoothed particle hydrodynamics,” Appl. Math. Modell. 37, ing Newtonian and non-Newtonian flows with a free surface,” Adv. Water
1431 (2013). Resour. 26, 787 (2003).
48 71
J. J. Monaghan, “SPH without a tensile instability,” J. Comput. Phys. 159, E. S. Lee, C. Moulinec, R. Xu, D. Violeau, D. Laurence, and P. Stansby,
290 (2000). “Comparisons of weakly compressible and truly incompressible algorithms
49
T. Jiang, L. G. Lu, and W. G. Lu, “The numerical investigation of spreading for the SPH mesh free particle method,” J. Comput. Phys. 227, 8417 (2008).
72
process of two viscoelastic droplets impact problem by using an improved S. Marrone, A. Colagrossi, D. Le Touze, and G. Graziani, “Fast free-surface
SPH scheme,” Comput. Mech. 53, 977 (2014). detection and level-set function definition in SPH solvers,” J. Comput. Phys.
50
Y. Melean, L. D. Sigalotti, and A. Hasmy, “On the SPH tensile instability in 229, 3652 (2010).
73
forming viscous liquid drops,” Comput. Phys. Commun. 157, 191 (2004). A. Colagrossi and M. Landrini, “Numerical simulation of interfacial flows
51
X. F. Yang, M. B. Liu, and S. L. Peng, “Smoothed particle hydrodynamics by smoothed particle hydrodynamics,” J. Comput. Phys. 191, 448 (2003).
74
modeling of viscous liquid drop without tensile instability,” Comput. Fluids Q. W. Ma and J. T. Zhou, “MLPG R method for numerical simulation of 2D
92, 199 (2014). breaking waves,” Cmes-Comput. Model. Eng. Sci. 43, 277 (2009).
52 75
N. Tsuruta, A. Khayyer, and H. Gotoh, “A short note on dynamic stabi- C. Huang, D. H. Zhang, Y. L. Si, Y. X. Shi, and Y. G. Lin, “Coupled finite
lization of moving particle semi-implicit method,” Comput. Fluids 82, 158 particle method for simulations of wave and structure interaction,” Coastal
(2013). Eng. 140, 147 (2018).
53 76
R. Xu, P. Stansby, and D. Laurence, “Accuracy and stability in incompress- A. Zhang, P. Sun, and F. Ming, “An SPH modeling of bubble rising and coa-
ible SPH (ISPH) based on the projection method and a new approach,” J. lescing in three dimensions,” Comput. Methods Appl. Mech. Eng. 294, 189
Comput. Phys. 228, 6703 (2009). (2015).
54 77
S. J. Lind, R. Xu, P. Stansby, and B. Rogers, “Incompressible smoothed par- K. Szewc, J. Pozorski, and J. P. Minier, “Spurious interface fragmentation
ticle hydrodynamics for free-surface flows: A generalised diffusion-based in multiphase SPH,” Int. J. Numer. Methods Eng. 103, 625 (2015).
algorithm for stability and validations for impulsive flows and propagating 78
A. Krimi, M. Rezoug, S. Khelladi, X. Nogueira, M. Deligant, and L. Ramı́rez,
waves,” J. Comput. Phys. 231, 1499 (2012). “Smoothed particle hydrodynamics: A consistent model for interfacial mul-
55
A. Colagrossi, B. Bouscasse, M. Antuono, and S. Marrone, “Particle packing tiphase fluid flow simulations,” J. Comput. Phys. 358, 53 (2018).
algorithm for SPH schemes,” Comput. Phys. Commun. 183, 1641 (2012). 79
N. Grenier, M. Antuono, A. Colagrossi, T. Le, and B. Alessandrini, “An
56
A. Khayyer, H. Gotoh, and Y. Shimizu, “Comparative study on accuracy and Hamiltonian interface SPH formulation for multi-fluid and free surface
conservation properties of two particle regularization schemes and pro- flows,” J. Comput. Phys. 228, 8380 (2009).
posal of an optimized particle shifting scheme in ISPH context,” J. Comput. 80
S. Adami, X. Y. Hu, and N. A. Adams, “A new surface-tension formula-
Phys. 332, 236 (2017). tion for multi-phase SPH using a reproducing divergence approximation,”
57
C. Huang, D. H. Zhang, Y. X. Shi, Y. L. Si, and B. Huang, “Coupled finite J. Comput. Phys. 229, 5011 (2010).
particle method with a modified particle shifting technology,” Int. J. Numer. 81
X. Y. Hu and N. A. Adams, “A constant-density approach for incompress-
Methods Eng. 113, 179 (2018). ible multi-phase SPH,” J. Comput. Phys. 228, 2082 (2009).
58 82
G. Oger, S. Marrone, D. L. Touze, and M. D. Leffe, “SPH accuracy improve- X. Y. Hu and N. A. Adams, “A multi-phase SPH method for macroscopic
ment through the combination of a quasi-Lagrangian shifting transport and mesoscopic flows,” J. Comput. Phys. 213, 844 (2006).
velocity and consistent ALE formalisms,” J. Comput. Phys. 313, 76 (2016). 83
X. Y. Hu and N. A. Adams, “An incompressible multi-phase SPH method,”
59
D. H. Zhang, Y. X. Shi, C. Huang, Y. L. Si, and W. Li, “A mixed character- J. Comput. Phys. 227, 264 (2007).
istic boundary condition for simulating viscous incompressible fluid flows 84
G. X. Zhu, L. Zou, Z. Chen, A. M. Wang, and M. B. Liu, “An improved SPH
around a hydrofoil,” J. Marine Sci. Technol. (published online). model for multiphase flows with large density ratios,” Int. J. Numer. Methods
60
C. Huang, J. M. Lei, and M. B. Liu, “An iterative method for homogeniz- Fluids 86, 167 (2017).
ing particles,” in IVth International Conference on Particle-based Methods, 85
Z. Chen, Z. Zong, M. B. Liu, L. Zou, H. T. Li, and C. Shu, “An SPH model for
Barcelona, Spain, 2015. multiphase flows with complex interfaces and large density differences,” J.
61
R. Vacondio and B. D. Rogers, “Consistent iterative shifting for SPH meth- Comput. Phys. 283, 169 (2015).
ods,” in Proceedings of 12th International SPHERIC Workshop, Ourense, 86
M. Zhang, “Simulation of surface tension in 2D and 3D with smoothed
Spain, 2017. particle hydrodynamics method,” J. Comput. Phys. 229, 7238 (2010).

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-37

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

87 112
M. Zhang and X.-L. Deng, “A sharp interface method for SPH,” J. Comput. W. M. Gelbart and A. Benshaul, “The ‘new’ science of ‘complex fluids,’” J.
Phys. 302, 469 (2015). Phys. Chem. 100, 13169 (1996).
88 113
T. Breinlinger, P. Polfer, A. Hashibon, and T. Kraft, “Surface tension and S. E. Spagnolie, Complex Fluids in Biological Systems (Springer, New York,
wetting effects with smoothed particle hydrodynamics,” J. Comput. Phys. 2015).
243, 14 (2013). 114
J. Li and M. Kwauk, “Exploring complex systems in chemical
89
G. T. Duan, S. Koshizuka, and B. Chen, “A contoured continuum surface engineering—The multi-scale methodology,” Chem. Eng. Sci. 58, 521 (2003).
force model for particle methods,” J. Comput. Phys. 298, 280 (2015). 115
W. Ge, J. Ma, J. Zhang, D. Tang, F. Chen, X. Wang, L. Guo, and J. Li, “Par-
90
A. M. Tartakovsky and A. Panchenko, “Pairwise force smoothed particle ticle methods for multi- scale simulation of complex flows,” Chin. Sci. Bull.
hydrodynamics model for multiphase flow: Surface tension and contact line 50, 1057 (2005).
dynamics,” J. Comput. Phys. 305, 1119 (2016). 116
X. Chen, “Numerical modeling of fluid-structure interaction with rhe-
91
J. U. Brackbill, D. B. Kothe, and C. Zemach, A Continuum Method for ologically complex fluids,” Ph.D. thesis, Technische Universität Darmstadt,
Modeling Surface Tension (Academic Press Professional, Inc., 1992). 2014.
92 117
B. Lafaurie, C. Nardone, R. Scardovelli, and G. Zanetti, “Modelling merging J. G. Oldroyd, “On the formulation of rheological equations of state,”
and fragmentation in multiphase flows with SURFER,” J. Comput. Phys. 113, Proc. R. Soc. London 200, 523 (1950).
134 (1994). 118
A. Peterlin, “Hydrodynamics of macromolecules in a velocity field with
93
H. Takeda, S. M. Miyama, and M. Sekiya, “Numerical simulation of vis- longitudinal gradient,” J. Polym. Sci. Part B: Polym. Lett. 4, 287 (1996).
cous flow by smoothed particle hydrodynamics,” Prog. Theor. Phys. 92, 939 119
N. P. Thien and R. I. Tanner, “A new constitutive equation derived from
(1994). network theory,” J. Non-Newtonian Fluid Mech. 2, 353 (1977).
94
O. Kum, W. G. Hoover, and H. A. Posch, “Viscous conducting flows with 120
J. N. Fang, R. G. Owens, L. Tacher, and A. Parriaux, “A numerical study
smooth-particle applied mechanics,” Phys. Rev. E 52, 4899 (1995). of the SPH method for simulating transient viscoelastic free surface flows,”
95
S. J. Watkins, A. S. Bhattal, N. Francis, J. A. Turner, and A. P. Whitworth, “A J. Non-Newtonian Fluid Mech. 139, 68 (2006).
new prescription for viscosity in smoothed particle hydrodynamics,” Astron. 121
A. Rafiee, M. T. Manzari, and M. Hosseini, “An incompressible SPH
Astrophys. Suppl. 119, 177 (1996). method for simulation of unsteady viscoelastic free-surface flows,” Int. J.
96
P. W. Cleary and J. J. Monaghan, “Conduction modelling using smoothed Non-Linear Mech. 42, 1210 (2007).
particle hydrodynamics,” J. Comput. Phys. 148, 227 (1999). 122
M. Ellero, M. Kröger, and S. Hess, “Viscoelastic flows studied by
97
P. Espanol and P. Warren, “Statistical mechanics of dissipative particle smoothed particle dynamics,” J. Non-Newtonian Fluid Mech. 105, 35 (2002).
dynamics,” Europhys. Lett. 30, 191 (1995). 123
A. Vázquez-Quesada, M. Ellero, and P. Español, “Smoothed particle
98
R. D. Groot and P. B. Warren, “Dissipative particle dynamics: Bridging the hydrodynamic model for viscoelastic fluids with thermal fluctuations,” Phys.
gap between atomistic and mesoscopic simulation,” J. Chem. Phys. 107, 4423 Rev. E 79, 056707 (2009).
(1997). 124
A. Vázquez-Quesada, M. Ellero, and P. Español, “A SPH-based particle
99
M. B. Liu, P. Meakin, and H. Huang, “Dissipative particle dynamics simu- model for computational microrheology,” Microfluidics Nanofluidics 13, 249
lation of pore-scale multiphase fluid flow,” Water Resour. Res. 43, W04411, (2012).
https://doi.org/10.1029/2006wr004856 (2007). 125
S. B. Dan, G. S. Grest, J. B. Lechman, F. Pierce, S. J. Plimpton, and P. R.
100
M. B. Liu, P. Meakin, and H. Huang, “Dissipative particle dynamics simula- Schunk, “Particle dynamics modeling methods for colloid suspensions,”
tion of multiphase fluid flow in microchannels and microchannel networks,” Comput. Particle Mech. 1, 321 (2014).
Phys. Fluids 19, 033302 (2007). 126
T. Ye, N. Phan-Thien, and C. T. Lim, “Particle-based simulations of red
101
I. Pagonabarraga and D. Frenkel, “Dissipative particle dynamics for blood cells—A review,” J. Biomech. 49, 2255 (2016).
interacting systems,” J. Chem. Phys. 115, 5015 (2001). 127
D. Y. Pan, N. Phan-Thien, and B. C. Khoo, “Dissipative particle dynamics
102
P. B. Warren, “Hydrodynamic bubble coarsening in off-critical vapor- simulation of droplet suspension in shear flow at low Capillary number,”
liquid phase separation,” Phys. Rev. Lett. 87, 225702 (2001). J. Non-Newtonian Fluid Mech. 212, 63 (2014).
103 128
S. Y. Trofimov, E. L. F. Nies, and M. A. J. Michels, “Thermodynamic con- J. T. Padding and W. J. Briels, “Systematic coarse-graining of the dynam-
sistency in dissipative particle dynamics simulations of strongly nonideal ics of entangled polymer melts: The road from chemistry to rheology,” J.
liquids and liquid mixtures,” J. Chem. Phys. 117, 9383 (2002). Phys.: Condens. Matter 23, 233101 (2011).
104 129
P. B. Warren, “Vapor-liquid coexistence in many-body dissipative parti- S. M. Hosseini and J. J. Feng, “A particle-based model for the transport of
cle dynamics,” Phys. Rev. E 68, 066702 (2003). erythrocyte in capillaries,” Chem. Eng. Sci. 64, 4488 (2009).
105 130
M. Arienti, W. Pan, X. Li, and G. Karniadakis, “Many-body dissipative par- H. N. Polwaththe-Gallage, S. C. Saha, E. Sauret, R. Flower, and Y. Gu, “A
ticle dynamics simulation of liquid/vapor and liquid/solid interactions,” J. coupled SPH-DEM approach to model the interactions between multiple
Chem. Phys. 134, 204114 (2011). red blood cells in motion in capillaries,” Int. J. Mech. Mater. Design 12, 477
106
A. Ghoufi, J. Emile, and P. Malfreyt, “Recent advances in many body dissi- (2016).
131
pative particles dynamics simulations of liquid-vapor interfaces,” Eur. Phys. G. W. Schmid-Schönbein, R. Skalak, S. Usami, and S. Chien, “Cell distri-
J. E 36, 10 (2013). bution in capillary networks,” Microvasc. Res. 19, 18 (1980).
107 132
R. M. Füchslin, H. Fellermann, A. Eriksson, and H. J. Ziock, “Coarse grain- I. H. Sarelius and B. R. Duling, “Direct measurement of microvessel hema-
ing and scaling in dissipative particle dynamics,” J. Chem. Phys. 130, 214102 tocrit, red cell flux, velocity, and transit time,” Am. J. Physiol. 243, H1018
(2009). (1982).
108 133
P. Español and P. B. Warren, “Perspective: Dissipative particle dynamics,” A. R. Pries, T. W. Secomb, P. Gaehtgens, and J. F. Gross, “Blood flow in
J. Chem. Phys. 146, 150901 (2017). microvascular networks. Experiments and simulation,” Circ. Res. 67, 826
109
I. V. Pivkin and G. E. Karniadakis, “Coarse-graining limits in open and (1990).
134
wall-bounded dissipative particle dynamics systems,” J. Chem. Phys. 124, B. M. Fenton, R. T. Carr, and G. R. Cokelet, “Nonuniform red cell dis-
184101 (2006). tribution in 20 to 100 micrometers bifurcations,” Microvasc. Res. 29, 103
110
Z. Li, Y. H. Tang, H. Lei, B. Caswell, and G. E. Karniadakis, “Energy- (1985).
135
conserving dissipative particle dynamics with temperature-dependent S. P. Sutera and R. Skalak, “The history of Poiseuille’s law,” Annu. Rev.
properties,” J. Comput. Phys. 265, 113 (2014). Fluid Mech. 25, 1 (1993).
111 136
M. Ellero and P. Español, “Everything you always wanted to know S. Chien, C. D. Tvetenstrand, M. A. Epstein, and G. W. Schmid-Schönbein,
about SDPD (but were afraid to ask),” Appl. Math. Mech. Engl. Ed. 39, 103 “Model studies on distributions of blood cells at microvascular bifurcations,”
(2018). Am. J. Physiol. 248, H568 (1985).

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-38

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

137 162
A. W. El-Kareh and T. W. Secomb, “A model for red blood cell motion in T. Ye, H. Shi, N. Phan-Thien, C. T. Lim, and Y. Li, “Relationship between
bifurcating microvessels,” Int. J. Multiphase Flow 26, 1545 (2000). transit time and mechanical properties of a cell through a stenosed
138
T. Hyakutake, T. Matsumoto, and S. Yanase, “Lattice Boltzmann simula- microchannel,” Soft Matter 14, 533 (2018).
163
tion of blood cell behavior at microvascular bifurcations,” Math. Comput. C. S. Peskin, “The immersed boundary method,” Acta Numer. 11, 479
Simul. 72, 134 (2006). (2002).
139 164
J. B. Freund, “Numerical simulation of flowing blood cells,” Annu. Rev. P. Balogh and P. Bagchi, “Direct numerical simulation of cellular-
Fluid Mech. 46, 67 (2014). scale blood flow in 3D microvascular networks,” Biophys. J. 113, 2815
140
Y. Imai, T. Omori, Y. Shimogonya, T. Yamaguchi, and T. Ishikawa, (2017).
165
“Numerical methods for simulating blood flow at macro, micro, and multi X. F. Yang, M. Ray, S.-C. Kong, and C.-B. M. Kweon, “SPH simulation of
scales,” J. Biomech. 49, 2221 (2016). fuel drop impact on heated surfaces,” Proc. Combust. Inst. (in press).
141 166
T. W. Secomb, B. Styp-Rekowska, and A. R. Pries, “Two-dimensional sim- X. F. Yang and S.-C. Kong, “3D simulation of drop impact on dry surface
ulation of red blood cell deformation and lateral migration in microvessels,” using SPH method,” Int. J. Comput. Methods 15, 1850011 (2018).
Ann. Biomed. Eng. 35, 755 (2007). 167
J. Kordilla, A. M. Tartakovsky, and T. Geyer, “A smoothed particle hydro-
142
J. O. Barber, J. P. Alberding, J. M. Restrepo, and T. W. Secomb, “Simu- dynamics model for droplet and film flow on smooth and rough fracture
lated two-dimensional red blood cell motion, deformation, and partitioning surfaces,” Adv. Water Resour. 59, 1 (2013).
in microvessel bifurcations,” Ann. Biomed. Eng. 36, 1690 (2008). 168
K. Ferrara, R. Pollard, and M. Borden, “Ultrasound microbubble contrast
143
T. W. Secomb, J. O. Barber, J. P. Alberding, and J. M. Restrepo, “Motion agents: Fundamentals and application to gene and drug delivery,” Annu. Rev.
of red blood cells in microvessels and bifurcations: Computational simula- Biomed. Eng. 9, 415 (2007).
tions,” Biorheology 45, 35 (2008). 169
C. C. Coussios and R. A. Roy, “Applications of acoustics and cavitation
144
T. W. Secomb, “Mechanics and computational simulation of blood flow to noninvasive therapy and drug delivery,” Annu. Rev. Fluid Mech. 40, 395
in microvessels,” Med. Eng. Phys. 33, 800 (2011). (2008).
145 170
W. K. Liu, Y. Liu, D. Farrell, L. Zhang, X. S. Wang, Y. Fukui, N. Patankar, S. L. Anna, “Droplets and bubbles in microfluidic devices,” Annu. Rev.
Y. Zhang, C. Bajaj, and J. Lee, “Immersed finite element method and its appli- Fluid Mech. 48, 285 (2016).
cations to biological systems,” Comput. Methods Appl. Mech. Eng. 195, 1722 171
A. K. Das and P. K. Das, “Incorporation of diffuse interface in smoothed
(2006). particle hydrodynamics: Implementation of the scheme and case studies,”
146
Y. Liu and W. K. Liu, “Rheology of red blood cell aggregation by computer Int. J. Numer. Methods Fluids 67, 671 (2011).
simulation,” J. Comput. Phys. 220, 139 (2006). 172
X. Bian, S. Litvinov, M. Ellero, and N. J. Wagner, “Hydrodynamic
147
Y. Sui, Y. T. Chew, P. Roy, Y. P. Cheng, and H. T. Low, “Dynamic motion of shear thickening of particulate suspension under confinement,” J. Non-
red blood cells in simple shear flow,” Phys. Fluids 20, 112106 (2008). Newtonian Fluid Mech. 213, 39 (2014).
148 173
J. Zhang, “Lattice Boltzmann method for microfluidics: Models and A. Rahmat and M. Yildiz, “A multiphase ISPH method for simulation of
applications,” Microfluidics Nanofluidics 10, 1 (2011). droplet coalescence and electro-coalescence,” Int J Multiphas Flow 105, 32
149
S. M. Hosseini and J. J. Feng, “How malaria parasites reduce the deforma- (2018).
174
bility of infected red blood cells,” Biophys. J. 103, 1 (2012). L. G. Leal, Advanced Transport Phenomena: Fluid Mechanics and Convec-
150
H. N. Polwaththe-Gallage, S. C. Saha, E. Sauret, R. Flower, W. Senadeera, tive Transport Processes (Cambridge University Press, 2007).
175
and Y. T. Gu, “SPH-DEM approach to numerically simulate the deformation D. Bonn, J. Eggers, J. Indekeu, J. Meunier, and E. Rolley, “Wetting and
of three-dimensional RBCs in non-uniform capillaries,” Biomed. Eng. Online spreading,” Rev. Mod. Phys. 81, 739 (2009).
15, 161 (2016). 176
J. H. Snoeijer and B. Andreotti, “Moving contact lines: Scales, regimes,
151
D. A. Fedosov, B. Caswell, and G. E. Karniadakis, “Systematic coarse- and dynamical transitions,” Annu. Rev. Fluid Mech. 45, 269 (2013).
graining of spectrin-level red blood cell models,” Comput. Methods Appl. 177
Y. Sui, H. Ding, and P. D. M. Spelt, “Numerical simulations of flows with
Mech. Eng. 199, 1937 (2010). moving contact lines,” Annu. Rev. Fluid Mech. 46, 97 (2014).
152
L. Xiao, Y. Liu, S. Chen, and B. Fu, “Simulation of deformation and aggre- 178
M. Huber, F. Keller, W. Sackel, M. Hirschler, P. Kunz, S. M. Hassanizadeh,
gation of two red blood cells in a stenosed microvessel by dissipative and U. Nieken, “On the physically based modeling of surface tension and
particle dynamics,” Cell Biochem. Biophys. 74, 513 (2016). moving contact lines with dynamic contact angles on the continuum scale,”
153
P. Balogh and P. Bagchi, “A computational approach to modeling cellular- J. Comput. Phys. 310, 459 (2016).
scale blood flow in complex geometry,” J. Comput. Phys. 334, 280 (2017). 179
F. Yeganehdoust, M. Yaghoubi, H. Emdad, and M. Ordoubadi, “Numeri-
154
T. Ye, N. Phan-Thien, B. C. Khoo, and C. T. Lim, “Stretching and relaxation cal study of multiphase droplet dynamics and contact angles by smoothed
of malaria-infected red blood cells,” Biophys. J. 105, 1103 (2013). particle hydrodynamics,” Appl. Math. Modell. 40, 8493 (2016).
155 180
T. Ye, N. Phan-Thien, B. C. Khoo, and C. T. Lim, “A file of red blood cells in E. Shigorina, J. Kordilla, and A. M. Tartakovsky, “Smoothed particle
tube flow: A three-dimensional numerical study,” J. Appl. Phys. 116, 124703 hydrodynamics study of the roughness effect on contact angle and droplet
(2014). flow,” Phys. Rev. E 96, 033115 (2017).
156 181
T. Ye, N. Phan-Thien, B. C. Khoo, and C. T. Lim, “Numerical modelling L. Li, L. Shen, G. D. Nguyen, A. El-Zein, and F. Maggi, “A smoothed particle
of a healthy/malaria-infected erythrocyte in shear flow using dissipative hydrodynamics framework for modelling multiphase interactions at meso-
particle dynamics method,” J. Appl. Phys. 115, 224701 (2014). scale,” Comput. Mech. 62, 1071 (2018).
157 182
T. Ye, N. Phan-Thien, B. C. Khoo, and C. T. Lim, “Dissipative parti- D. Y. Pan, J. X. Hu, and X. M. Shao, “Lees-Edwards boundary condition
cle dynamics simulations of deformation and aggregation of healthy and for simulation of polymer suspension with dissipative particle dynamics
diseased red blood cells in a tube flow,” Phys. Fluids 26, 111902 (2014). method,” Mol. Simul. 42, 328 (2016).
158 183
T. Ye, H. Shi, L. Peng, and Y. Li, “Numerical studies of a red blood cell in H. H. Bui, R. Fukagawa, K. Sako, and S. Ohno, “Lagrangian meshfree par-
rectangular microchannels,” J. Appl. Phys. 122, 084701 (2017). ticles method (SPH) for large deformation and failure flows of geomaterial
159 using elastic–plastic soil constitutive model,” Int. J. Numer. Anal. Methods
T. Ye, N. Phan-Thien, C. T. Lim, L. Peng, and H. Shi, “Hybrid smoothed
dissipative particle dynamics and immersed boundary method for simula- Geomech. 32, 1537 (2008).
tion of red blood cells in flows,” Phys. Rev. E 95, 063314 (2017). 184
M. Ellero and R. I. Tanner, “SPH simulations of transient viscoelastic
160 flows at low Reynolds number,” J. Non-Newtonian Fluid Mech. 132, 61 (2005).
T. Ye, N. Phan-Thien, C. T. Lim, and Y. Li, “Red blood cell motion and
deformation in a curved microvessel,” J. Biomech. 65, 12 (2017). 185
J. Ren, J. Ouyang, and T. Jiang, “An improved particle method for simu-
161 lation of the non-isothermal viscoelastic fluid mold filling process,” Int. J.
T. Ye, L. Peng, and Y. Li, “Three-dimensional motion and deformation of
a red blood cell in bifurcated microvessels,” J. Appl. Phys. 123, 064701 (2018). Heat Mass Transfer 85, 543 (2015).

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-39

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

186 209
M. Vakilha and M. T. Manzari, “Modelling of power-law fluid flow through C. Antoci, M. Gallati, and S. Sibilla, “Numerical simulation of fluid-
porous media using smoothed particle hydrodynamics,” Transp. Porous structure interaction by SPH,” Comput. Struct. 85, 879 (2007).
Media 74, 331 (2008). 210
J. B. Kajtar and J. J. Monaghan, “On the swimming of fish like bodies near
187
X. J. Fan, R. I. Tanner, and R. Zheng, “Smoothed particle hydrodynam- free and fixed boundaries,” Eur. J. Mech. B: 33, 1 (2012).
ics simulation of non-Newtonian moulding flow,” J. Non-Newtonian Fluid 211
X. F. Yang, M. B. Liu, S. L. Peng, and C. G. Huang, “Numerical modeling
Mech. 165, 219 (2010). of dam-break flow impacting on flexible structures using an improved SPH-
188
D. Laigle, P. Lachamp, and M. Naaim, “SPH-based numerical investiga- EBG method,” Coastal Eng. 108, 56 (2016).
tion of mudflow and other complex fluid flow interactions with structures,” 212
X. F. Yang, M. B. Liu, and S. L. Peng, “Smoothed particle hydrodynam-
Comput. Geosci. 11, 297 (2007). ics and element bending group modeling of flexible fibers interacting with
189
H. N. Zhu, N. S. Martys, C. Ferraris, and D. De Kee, “A numerical study viscous fluids,” Phys. Rev. E 90, 063011 (2014).
of the flow of Bingham-like fluids in two-dimensional vane and cylinder 213
S. S. P. Kumar, B. S. V. Patnaik, and K. Ramamurthi, “Prediction of air
rheometers using a smoothed particle hydrodynamics (SPH) based method,” blast mitigation in an array of rigid obstacles using smoothed particle
J. Non-Newtonian Fluid Mech. 165, 362 (2010). hydrodynamics,” Phys. Fluids 30, 046105 (2018).
190
X. Y. Xu, J. Ouyang, T. Jiang, and Q. Li, “Numerical simulation of 3D- 214
P. D. Thomas and C. K. Lombard, “Geometric conservation law and
unsteady viscoelastic free surface flows by improved smoothed particle its application to flow computations on moving grids,” AIAA J. 17, 1030
hydrodynamics method,” J. Non-Newtonian Fluid Mech. 177, 109 (2012). (1979).
191
J. Ren, T. Jiang, W. Lu, and G. Li, “An improved parallel SPH approach 215
M. B. Liu, J. R. Shao, and H. Q. Li, “An SPH model for free surface
to solve 3D transient generalized Newtonian free surface flows,” Comput. flows with moving rigid objects,” Int. J. Numer. Methods Fluids 74, 684
Phys. Commun. 205, 87 (2016). (2014).
192
W. Wang, G. Q. Chen, Z. Han, S. H. Zhou, H. Zhang, and P. D. Jing, “3D 216
M. Greenhow and S. Moyo, “Water entry and exit of horizontal circular
numerical simulation of debris-flow motion using SPH method incorporat- cylinders,” Philosoph. Trans. R. Soc. A 355, 551 (1997).
ing non-Newtonian fluid behavior,” Nat Hazards 81, 1981 (2016). 217
P. Lin, “A fixed-grid model for simulation of a moving body in free surface
193
M. R. Hashemi, R. Fatehi, and M. T. Manzari, “SPH simulation of inter- flows,” Comput. Fluids 36, 549 (2007).
acting solid bodies suspended in a shear flow of an Oldroyd-B fluid,” J. 218
P. A. Tyvand and T. Miloh, “Free-surface flow due to impulsive motion of
Non-Newtonian Fluid Mech. 166, 1239 (2011). a submerged circular cylinder,” J. Fluid Mech. 286, 67 (1995).
194
X. Y. Xu and J. Ouyang, “A SPH-based particle method for simulating 3D 219
M. B. Liu, J. R. Shao, and H. Q. Li, “Numerical simulation of hydro-elastic
transient free surface flows of branched polymer melts,” J. Non-Newtonian problems with smoothed particle hydro-dynamics method,” J. Hydrodyn.
Fluid Mech. 202, 54 (2013). 25, 673 (2013).
195
E. Bertevas, J. Ferec, B. C. Khoo, G. Ausias, and P.-T. Nhan, “Smoothed 220
G. Zhou, Z. Chen, W. Ge, and J. Li, “SPH simulation of oil displacement in
particle hydrodynamics (SPH) modeling of fiber orientation in a 3D printing cavity-fracture structures,” Chem. Eng. Sci. 65, 3363 (2010).
process,” Phys. Fluids 30, 103103 (2018). 221
B. Seiffert, M. Hayatdavoodi, and R. C. Ertekin, “Experiments and compu-
196
W. Pan, A. M. Tartakovsky, and J. J. Monaghan, “Smoothed particle hydro- tations of solitary-wave forces on a coastal-bridge deck. Part I: Flat Plate,”
dynamics non-Newtonian model for ice-sheet and ice-shelf dynamics,” Coastal Eng. 88, 194 (2014).
J. Comput. Phys. 242, 828 (2013). 222
Y. Wei, T. Abadie, A. Henry, and F. Dias, “Wave interaction with an
197
J. W. Glen, “The creep of polycrystalline ice,” Proc. R. Soc. London Seri. A oscillating wave surge converter. Part II: Slamming,” Ocean Eng. 113, 319
228, 519 (1955). (2016).
198
J. R. Shao, H. Q. Li, G. R. Liu, and M. B. Liu, “An improved SPH method for 223
F. Dias, E. Renzi, S. Gallagher, D. Sarkar, Y. Wei, T. Abadie, C. Cummins,
modeling liquid sloshing dynamics,” Comput. Struct. 100, 18 (2012). and A. Rafiee, “Analytical and computational modelling for wave energy sys-
199
M. Antuono, A. Colagrossi, S. Marrone, and C. Lugni, “Propagation of tems: The example of oscillating wave surge converters,” Acta Mech. Sin. 33,
gravity waves through an SPH scheme with numerical diffusive terms,” 647 (2017).
Comput. Phys. Commun. 182, 866 (2011). 224
F. Dias and J.-M. Ghidaglia, “Slamming: Recent progress in the evaluation
200
C. Lugni, “ A study about the free-surface waves and the freely floating of impact pressures,” Annu. Rev. Fluid Mech. 50, 243 (2018).
structure interactions,” Ph.D. thesis, University of Rome La Sapienza, 1999. 225
A. Shakeri, K.-W. Lee, and T. Poeschel, “Limitation of stochastic rota-
201
X. F. Yang and M. B. Liu, “Numerical study of Rayleigh-Taylor instabil- tion dynamics to represent hydrodynamic interaction between colloidal
ity by using smoothed particle hydrodynamics,” Acta Phys. Sin. 66, 164701 particles,” Phys. Fluids 30, 013603 (2018).
(2017). 226
J. W. Swan and G. Wang, “Rapid calculation of hydrodynamic and trans-
202
X. F. Yang and S.-C. Kong, “Smoothed particle hydrodynamics method port properties in concentrated solutions of colloidal particles and macro-
for evaporating multiphase flows,” Phys. Rev. E 96, 033309 (2017). molecules,” Phys. Fluids 28, 011902 (2016).
203
A. M. Tartakovsky and P. Meakin, “A smoothed particle hydrodynam- 227
T.-D. Thien, P.-T. Nhan, and B. C. Khoo, “A smoothed particle hydrody-
ics model for miscible flow in three-dimensional fractures and the two- namics (SPH) study on polydisperse sediment from technical activities on
dimensional Rayleigh Taylor instability,” J. Comput. Phys. 207, 610 (2005). seabed,” Phys. Fluids 30, 023302 (2018).
204
G. T. Duan, B. Chen, X. M. Zhang, and Y. C. Wang, “A multiphase 228
T.-D. Thien, P.-T. Nhan, and B. C. Khoo, “A smoothed particle hydrody-
MPS solver for modeling multi-fluid interaction with free surface and its namics (SPH) study of sediment dispersion on the seafloor,” Phys. Fluids 29,
application in oil spill,” Comput. Methods Appl. Mech. Eng. 320, 133 (2017). 083302 (2017).
205 229
X. F. Yang and M. B. Liu, “Numerical modeling of oil spill containment by E. Chaparian, A. Wachs, and I. A. Frigaard, “Inline motion and hydrody-
boom using SPH,” Sci. China Phys., Mech. Astron. 56, 315 (2013). namic interaction of 2D particles in a viscoplastic fluid,” Phys. Fluids 30,
206
F. R. Ming, A. M. Zhang, H. Cheng, and P. N. Sun, “Numerical simulation of 033101 (2018).
a damaged ship cabin flooding in transversal waves with smoothed particle 230
Z. Ouyang, J.-Z. Lin, and X. Ku, “Hydrodynamic interactions between
hydrodynamics method,” Ocean Eng. 165, 336 (2018). a self-rotation rotator and passive particles,” Phys. Fluids 29, 103301
207
F. R. Ming, A. M. Zhang, Y. Z. Xue, and S. P. Wang, “Damage character- (2017).
istics of ship structures subjected to shockwaves of underwater contact 231
S. Turek and C. Berker, FEATFLOW: Finite Element Software for the
explosions,” Ocean Eng. 117, 359 (2016). Incompressible Navier Stokes Equations (IWR, 1996).
208 232
Z. L. Zhang, D. L. Feng, and M. B. Liu, “Investigation of explosive weld- Z. L. Zhang and M. B. Liu, “Smoothed particle hydrodynamics with kernel
ing through whole process modeling using a density adaptive SPH method,” gradient correction for modeling high velocity impact in two- and three-
J. Manufact. Processes 35, 169 (2018). dimensional spaces,” Eng. Anal. Boundary Elem. 83, 141 (2017).

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-40

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

233 240
F. Plassard, J. Mespoulet, and P. Hereil, “Hypervelocity impact of alu- P. A. Cundall and O. D. L. Strack, “A discrete numerical model for granular
minium sphere against aluminium plate : Experiment and LS-DYNA cor- assemblies,” Geotechnique 29, 47 (1979).
relation,” in 8th European LS-DYNA Users Conference, Strasbourg, France, 241
U. E. Shamy and M. Zeghal, “Coupled continuum-discrete model for
2011. saturated granular soils,” J. Eng. Mech. 131, 413 (2005).
234
G. R. Cowan, O. R. Bergmann, and A. H. Holtzman, “Mechanism of bond 242
P. W. Cleary and M. Prakash, “Discrete–element modelling and
zone wave formation in explosion-clad metals,” Metall. Mater. Trans. B 2, smoothed particle hydrodynamics: Potential in the environmental sci-
3145 (1971). ences,” Philosoph. Trans.: Math., Phys. Eng. Sci. 362, 2003 (2004).
235
M. B. Liu, G. R. Liu, and K. Y. Lam, “A one-dimensional meshfree 243
J. U. Brackbill and H. M. Ruppel, “Flip: A method for adaptively zoned,
particle formulation for simulating shock waves,” Shock Waves 13, 201 particle-in-cell calculations of fluid flows in two dimensions,” J. Comput.
(2003). Phys. 65, 314 (1986).
236
D. A. Hu, T. Long, Y. Xiao, X. Han, and Y. Gu, “Fluid–structure interaction 244
D. Sulsky, Z. Chen, and H. L. Schreyer, “A particle method for history-
analysis by coupled FE–SPH model based on a novel searching algorithm,” dependent materials,” Comput. Methods Appl. Mech. Eng. 118, 179 (1994).
Comput. Methods Appl. Mech. Eng. 276, 266 (2014). 245
P. M. Kulkarni, C. C. Fu, M. S. Shell, and L. G. Leal, “Multiscale modeling
237
A. M. Zhang, F. R. Ming, and S. P. Wang, “Coupled SPHS–BEM method with smoothed dissipative particle dynamics,” J. Chem. Phys. 138, 234105
for transient fluid–structure interaction and applications in underwater (2013).
impacts,” Appl. Ocean Res. 43, 223 (2013). 246
N. D. Petsev, L. G. Leal, and M. S. Shell, “Hybrid molecular-continuum
238
G. R. Johnson, R. A. Stryk, and S. R. Beissel, “SPH for high veloc- simulations using smoothed dissipative particle dynamics,” J. Chem. Phys.
ity impact computations,” Comput. Methods Appl. Mech. Eng. 139, 347 142, 044101 (2015).
(1996). 247
N. D. Petsev, L. G. Leal, and M. S. Shell, “Multiscale simulation of ideal
239
M. B. Liu and G. R. Liu, Particle Methods for Multi-Scale and Multi-Physics mixtures using smoothed dissipative particle dynamics,” J. Chem. Phys. 144,
(World Scientific, 2015). 084115 (2016).

Phys. Fluids 31, 011301 (2019); doi: 10.1063/1.5068697 31, 011301-41

Published under license by AIP Publishing