Professional Documents
Culture Documents
SPH Mass Conservation
SPH Mass Conservation
SPH Mass Conservation
a r t i c l e i n f o a b s t r a c t
Article history: Anomalous uctuations in pressure associated with the Lagrangian smoothed particle hy-
Received 9 March 2015 drodynamics method (SPH) have recently been treated by introducing diffusive terms in
Revised 5 February 2016
the conservation of mass equation. Here, ve consistency conditions are proposed for such
Accepted 9 May 2016
diffusive terms; three that must be satised and two that add to the generality of the
Available online 27 May 2016
models. Each of the existing diffusive terms are reviewed and their consistency properties
Keywords: and relationships discussed and summarized to provide a guide for their usage.
SPH The equivalence of Riemann solver SPH formulations and conservation of mass equa-
Riemann solvers tion diffusive terms is demonstrated in this paper.
-SPH A practical application consisting of a simulation of a dam break ow is proposed to
Consistency show the consistency properties and relationship of some of the diffusive terms discussed
Conservation in the paper.
2016 Elsevier Inc. All rights reserved.
1. Introduction
Smoothed Particle Hydrodynamics (SPH) is a meshless Lagrangian method developed to solve uid dynamics equations
in which the uid domain is divided in a set of particles transporting the ow properties [1,2]. The most popular SPH
formulation is the weakly-compressible model, which uses an equation of state to relate pressure and the density elds,
allowing the usage of a purely explicit time integration scheme [3]. This scheme has the drawbacks of small time steps,
dictated by the (usually reduced) speed of sound in the uid, and some numerical instabilities, which have been traditionally
managed by the use of an articial viscosity term [3].
Even with the articial viscosity term, high frequency noise can be expected to be generated in the density and pressure
elds that result in poor simulations for pressure; although the uid kinematics are often well modeled. To deal with this
problem, some authors have adopted incompressible SPH formulations, based on the solution of the Poisson equation for
the pressure [4,5]. The incompressible SPH formulation has the main drawback that the Poisson equation is solved imposing
Dirichlet boundary conditions, and therefore, the pressure must be explicitly imposed on a uid particle in the free surface
of the domain, which usually requires a free surface detection algorithm and an algebraic system with a sparse matrix
solution each time step.
Corresponding author. Tel.: +34 629802681.
E-mail addresses: jl.cercos@upm.es (J.L. Cercos-Pita), rad@jhu.edu (R.A. Dalrymple), alexis.herault@cnam.fr (A. Herault).
http://dx.doi.org/10.1016/j.apm.2016.05.016
0307-904X/ 2016 Elsevier Inc. All rights reserved.
J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736 8723
Another approach to reduce the noise in the pressure eld, which we discuss in this paper, consists of modifying the
standard SPH numerical scheme by solving a Riemann problem each time step [68], resulting in an additional diffusive term
that results in a stable time integration. Inspired in this model, Ferrari et al. [9] proposed a simpler approach, introducing
a Rusanov (upwind ux-based) diffusive term directly in the conservation of mass equation. In a similar way Molteni and
Colagrossi [10] introduced another diffusive term in the conservation of mass equation. Both of these terms are unfortunately
incompatible with a hydrostatic equilibrium problem, such as still water in a tank, due to the instabilities arising at the free
surface. Antuono et al. [11] proposed an improvement to the approach proposed by Molteni and Colagrossi [10] by including
a correction term designed to x the problems along the free surface, calling it -SPH.
In a parallel way, Fatehi and Manzari [12] introduced a new term based on a numerical study of the pressure evolution
for a checker-board problem. This model is the rst one that is proposed as a Laplacian of the pressure eld rather than of
the density eld. Later Hashemi et al. [13] proposed a correction to adapt it to the free surface problems, in a similar way
as Antuono et al. [11] for the -SPH model.
Some of these terms have been already discussed by Antuono et al. [14], but in this work we will revisit them with the
following main objectives:
The conditions that these terms must satisfy will be reviewed, adding conditions that extend the applicability of the
terms are also introduced.
The relation between a Riemann-solver based SPH (hereinafter R-SPH) and the terms articially included in the conser-
vation of mass equation will be formalized in a similar way as used by Monaghan [15] for the momentum equation.
The term proposed by Hashemi et al. [13] is included in the analysis.
This work is organized as follows: First, the governing equations of the physical problem are presented. Second, the
weakly compressible SPH approach (hereinafter WCSPH), introducing the diffusive term, is described. Third, the numerical
diffusive term in the continuity equation is studied, imposing several conditions to assert its consistency. Fourth, the pre-
viously proposed terms of others will be discussed, focusing on their benets and drawbacks. Fifth, a practical application
will be carried out. Finally, a conclusions summary together with future work targets are provided.
2. Governing equations
Rather than examining the full NavierStokes equations, the importance of the conservation of mass diffusion terms can
be illustrated by the incompressible Euler equations (non-viscous Navier-Stokes equations) that are numerically solved for
the uid ow, using the weakly-compressible SPH (hereinafter WCSPH) technique. In this formulation the incompressibility
is sought by modeling the ow with a slightly compressible uid that in the ow regime expected is subject to very small
density uctuations [3].
The uid is assumed to be barotropic, which implies that the internal energy equation is decoupled from the continuity
and momentum equations, and the pressure is related to the density through an equation of state (hereinafter EOS). Under
these assumptions, the Euler equations in Lagrangian form read:
d
= div(u ), (1)
dt
du p
=g , (2)
dt
p = p( ), the EOS. (3)
Here, is the uid density, p is the pressure and g is a generic external volumetric force eld, usually gravity.
The ow velocity, u, is dened as the material derivative of a uid particle position r:
dr
= u. (4)
dt
In WCSPH, as aforementioned, an equation of state (EOS) relating the pressure and the density is imposed. Monaghan
[1] suggested the application of the following stiffened equation of state:
cs2 0
p = p0 + 1 , (5)
0
where 0 is the reference density, p0 is the ambient pressure, and cs is the speed of sound [16].
However, assumed the weakly-compressibility hypothesis, a Taylor series, around = 0 , demonstrates that the following
linear form is the main contribution in the previous equation (see [11]):
p = p0 + cs2 ( 0 ). (6)
3. Numerical model
In this section the SPH numerical method to approximate the solution of the governing equations (1)(3), is introduced,
with its two steps. First a smoothing process (associated to the SPH continuous model) is used and then the numerical
integration (associated with the SPH discrete model) is carried out.
3.1.1. Smoothing
In SPH [17,18] a kernel Wh (x) is dened as a positive even function such that
Wh (y x ) dy = 1, (7)
Rd
lim Wh (x ) = 0. (8)
|x|
For practical purposes, the kernel Wh (x) vanishes for |x| > s h, where s is an integer greater than 0, and h is the charac-
teristic length of the kernel. Hereinafter we will call the ball of radius s h centered at x (except at the boundaries).
The SPH convolution of a scalar or vector function f(x) with respect to the kernel Wh (x) is dened as
f (x ) := f (y )Wh (y x ) dy. (9)
In the same way, the value of a rst order differential operator can be also convoluted:
D f (x ) = D f (y ) Wh (y x ) dy, (10)
where D f (y ) is generally unknown. However, expanding the expression and applying the divergence theorem, the differen-
tial operators can be conveniently moved to the kernel function:
D f (x ) = f (y ) Wh (y x ) dy + f (y ) n(y )Wh (y x ) dy, (11)
d ( x )
= (x ) div(u(x ) ) div(F (x ) ). (13)
dt
In Appendix A the equivalence between the diffusive term of the mass variation rate equation, used in the SPH Riemann
solver approach [68], and the diffusive term added to the conservation of mass equation, used in -SPH, is investigated.
For consistency reasons associated with the free surface, in (13) the rst term of the right hand side is symmetrized (see
[19,24]). However the second term of the right hand side has not been dened yet, so a workaround can be performed
during the SPH convolution application to denote the possible symmetrized forms of the diffusive term:
d ( x )
( (y ) uxy + F xy ) Wh (y x ) dy, (14)
dt
where the terms uxy and Fxy denote [u(y ) u(x )] and F(x, y) respectively. Note that if the term F(x, y) is zero, the expression
reduces to the commonly used SPH form.
While Eq. (14) represents the SPH approximated version of the conservation of mass equation, here the focus will be on
the part affected by the numerical diffusive term.
J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736 8725
du ( x ) p(x )
+ g, (15)
dt (x )
where:
p(x ) p( x ) p( y )
= (y ) + Wh (y x ) dy. (16)
(x ) (x )2 (y )2
fb
f a = Wab mb , (18)
bFluid
b
fb
D f a = Wab mb . (19)
bFluid
b
In Eqs. (18) and (19) the index b of the sum moves along all the uid particles, including the self particle a, but
ltering out the ones further than the distance s h, where the kernel vanishes. In the case that uid extensions are used to
model the boundaries, such particles should be included in the summation as well.
In this work we focus on the added diffusive term, and therefore we are conveniently assigning a name to such contribution:
d d F ab
= + a div(u )a = Wab mb . (21)
dt
a
dt
a bFluid
b
with
p pa pb
= + Wab mb. (23)
a bFluid
a2 b2
The symmetric structure of the differential operator improves their conservation and consistency properties (see Mon-
aghan [2], Colagrossi et al. [19] for details).
8726 J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736
In order to assert the consistency of the model, the original governing equation (1) should be recovered in the continuum.
This condition can be written as
d
lim = 0, (24)
x/h 0 dt
h0 a
where x is the distance between particles. As it will be shown later, almost authors fullls the condition (24) dening
terms of the following form:
Even though the condition 4.1 is enough to assert the mass conservation the continuum level, achieving an intrinsic
conservation of the global mass, even at a discrete level, become a desirable feature of the diffusive model, i.e.:
d
= 0. (26)
dt
aFluid a
Antuono et al. [14] already demonstrated that symmetric forms of the term Fab are already fullling this condition.
Antuono et al. [11] introduced this condition. In such work the authors demonstrated that the term originally proposed
by Molteni and Colagrossi [10] becomes inconsistent close to the free-surface due to the singularity of the Morris formula
[19,25].
In a similar way, any other term based on the Laplacian of the density or pressure elds may be affected by such singu-
larity, as Antuono et al. [14] demonstrated for the term proposed by Ferrari et al. [9].
As Antuono et al. [31] already demonstrated, the diffusive term affects the energy balance of the system at the discrete
level, or even at the continuum if the condition 1 is not fullled. However, the methodology described in such work should
be generalized to can be applied to all the considered terms herein.
For a non-viscous uid, we can write the following expression [31]:
dEi dEc
= P PS , (27)
dt dt
where Ei and Ec are the internal energy and the energy due to the compressibility, P is the power due to the addition of
the diffusive term to the mass conservation equation, and PS is the power term due to the interactions with the boundaries.
As a consequence of the previous expression, if the term P becomes negative then it can be asserted that the diffusive
term is dissipating energy.
With the current notation, such power term can be expressed as follows:
NFluid
pa d pa
NFluid NFluid
ma mb
P = ma 2 = F ab Wab . (28)
a=1
a dt a a=1 b=1
a a b
However, it is convenient to rearrange the expression above, in order to point out its relation with the particle pairs inter-
actions. To this end, we can reorganize the summations and use the antisymmetry property of the kernel gradient, reading
as follows:
NFluid NFluid p pb
ma mb
F ba Wab
a
P = F ab . (29)
a=1 b=a+1
a b a b
Taking into account that the gradient of the kernel can be expressed as Wab = rabW
ab , where W
ab is a positive scalar,
we can establish the following strong condition:
p pa
b
F ba F ab rab 0 P 0. (30)
b a
J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736 8727
Antuono et al. [11] and Antuono et al. [14] discuss the process to set a convenient parameter to their model, and an
associated Courant condition for the time step. The stability analysis performed requires an expertise regarding the model
to can successfully apply it in the simulations. Hence it is a desirable, but not mandatory, property of a model if it could be
used without needing to tune parameters in order to have a stable time integration. Note that the commonly used articial
viscosity depends on a parameter to be tuned.
Vila [7] included the Riemann solvers in the SPH context, where the Godunov scheme [32] is approximately solved (see
[6,33]). However, this model can be studied in the context of the WCSPH model as we demonstrate in Appendix A, allowing
therefore the usage of the conditions introduced in Section 4 to test it. Such equivalence between the term of the mass
variation rate equation in R-SPH and the diffusive terms added in the mass conservation equation in WCSPH was already
intuited, but not formalized, by Ferrari et al. [9].
The WCSPH equivalent term can be written as follows:
ab = 2 a,ab ua,ab u (r
F Vi ab , t ) , (31)
where a, ab and ua, ab are the approximate solutions of a Riemann problem for the density and velocity respectively, and
ra + rb
r ab = . (32)
2
Such demonstration is out of the scope of this work. However, it can be asserted that the condition 2 is fullled since it is
a symmetric term.
Regarding the condition 3, since this term is proposed as a function of the divergence of the velocity resulting from the
Riemman problem solution, not using therefore the Morris formula [25], it is not affected by the singularity described by
Antuono et al. [11]. However, similar inconsistencies than the ones found by [19] for the original divergence of the velocity
operator should be considered.
Regarding the condition 4, for the symmetric terms like this one, where F ab = F ba , Eq. (30) can be slightly simplied as
p pa
b
F ab rab 0, (34)
b a
where introducing the term (31), it can be seen that the power term sign is not dened, not fullling the condition 3.
Vila [7] has already analyzed the stability of the time integration, which has been proven later in several practical appli-
cations (see for instance Molteni and Bilello [34] and Koukouvinis et al. [33])
It should be remarked that the solution of the Riemann problem adds a signicant computational cost to the model.
Ferrari et al. [9] proposed a diffusive term directly inside the conservation of mass equation as a simplication of the
model introduced by Vila [7]. In this case Ferrari et al. [9] chose an upwind Rusanov [35] ux, such that, taking a constant
sound speed for simplicity, the model can be written as
rab
ab = cs (b a )
F Fe . (35)
|rab |
5.2.1. Conditions test
Applying the nite differences approximation we can write:
ab cs b |r ab |.
F Fe (36)
However, in order to check the condition 1, we must recover the continuous formulation:
d
= cs div( (x )|rxy | ), (37)
dt
8728 J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736
The previous term is not vanishing for h 0, and therefore the condition 1 is not fullled. However, even though the term
is inconsistent with the mass conserving equation, it has a symmetric form (at least if a constant sound speed is considered),
and therefore it is intrinsically conserving the global mass.
Regarding the condition 3 , Antuono et al. [14] already demonstrated that this model is affected by the singularity of the
Morris formula close to the free surface.
Becoming a symmetric term the expression (34) can be used again, resulting as follows:
p pa
p pa
sign b
F Fe
ab r ab = sign
b
( b a ) . (39)
b a b a
Antuono et al. [31] already demonstrated that the expression above is negative if the EOS (6) is applied, such that the
condition 4 is fullled by this model.
In this term no tunning parameters are used, fullling in principle the condition 5. Unfortunately, as we will see in
Section 7.3 this term may turn the simulation unstable if a tuned parameter is not introduced in Eq. (35).
5.3. The diffusion term proposed by Molteni and Colagrossi [10] and Antuono et al. [11]
This term was initially introduced by Molteni and Colagrossi [10] as a function of the Laplacian of the density eld, and
corrected later by Antuono et al. [11] to x the problems arising on the free surface, transforming it into a fourth order
differential operator. The term can be written as follows
rab
ab = 2 h cs (b a )
F An + f ( , ) , (40)
|rab |2 L a b
where is a non-dimensional factor, and fL ( a , b ) is a renormalized correction term [11,13].
5.4. A diffusion term inspired by Fatehi and Manzari [12] and Hashemi et al. [13]
The last proposed term to include in the conservation of mass equation was proposed by Fatehi and Manzari [12], who
were the rst to use the Laplacian of the pressure instead of the Laplacian of the density. The term was later formalized by
Hashemi et al. [13], introducing a correction concerning the boundaries, in a similar way as Antuono et al. [11] did for the
term proposed by Molteni and Colagrossi [10]. During this work a modied version of the originally dened model is used,
where the averaged density value (a + b )/2 is replaced by density of reference 0 :
t a r
F Ha
ab = ( pb pa ) ab 2 + fL ( pa , pb ) , (41)
0 |rab |
which is really a more convenient formulation to analyze the consistency. t is the time step used.
Table 1
Properties of existing diffusive terms proposed for the Conservation of mass equa-
tion in SPH. FVi = Vila [7], FFe = Ferrari et al. [9], FAn = Antuono et al. [11], FHa =
modied Hashemi et al. [13].
In a similar way to the previous model, this model requires a correction to avoid the singularity close to the free surface.
To test the condition 4 the term can be expressed as follows:
Ha
ab = a F ab ,
F Ha (43)
with
Ha Ha
F ab = F ba . (44)
where the correction term, fL (pa , pb ), has been dropped from the analysis, in a similar way of Antuono et al. [31].
The term is clearly dissipating energy , hence condition 5 is fullled.
This term satises the conditions 5 as there are no tuning parameters.
6. Summary
In the previous section, the most popular diffusive terms used in WCSPH modeling have been discussed. In Table 1 the
properties of each term are shown.
7. Practical application
In this section we will simulate and analyze the largely documented experiments of Lobovsk et al. [36], and more
precisely the H = 0.3 water level case, in order to evaluate the effect of the diffusive terms.
One of the most popular test cases used to show the capabilities and the performance of a new SPH code is a dam break.
Lee et al. [37] were the rst whose carried an experimental study where detailed data, regarding both the ow kinematics
and dynamics, is provided, including the forces measured on a solid vertical wall downstream the dam, using for that
large circular force transducer plates (see previous works, [3843]). This experiment has been widely applied for validations
([44,45] are good examples).
Recently Lobovsk et al. [36] have revisited the same experimental study, providing more detailed data, and performing
a large number of test repetitions of the pressure measured on the solid vertical wall downstream the dam.
In Fig. 1 a schematic view of the simulation setup is shown. It should be remarked that the gate will not be simulated
for simplicity, therefore an initial column of uid in rest state, with the hydrostatic pressure freely evolves up to 1.925 s. 2D
simulations are performed. The wave height has been measured in the positions H1, H2, H3 and H4 along the experiments.
Also some sensors are placed on the solid vertical wall downstream the dam at 0.003, 0.015, 0.030, and 0.080 m from the
bottom, where the pressure will be captured. Lobovsk et al. [36] provides a complete pressure record with time obtained in
each experiment, and the associated statistically valid intervals of the pressure of each sensor, which we will use to compare
the results obtained with the simulations.
8730 J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736
762.5 W
514
265 300
H4 H3 H2 H1
H 4 D
H
D=600
z
80
3
2L 2
x
30
15
A=1010 B=600 1
3
37.5
L=1610 150
Fig. 1. A side view of the plexiglass tank used with the water level measuring positions (left) and a front view showing the locations of pressure sensors
on the impact wall (right). Dimensions are in millimeters. From Lobovsk et al. [36].
The dambreak experiment will be simulated using the following diffusive terms:
The implementation of the model proposed by Vila [7] is extremely different from the other ones due to the Riemann
problem approximate solution requirement, which is considered out of the scope of this paper.
One key benet of the model proposed by Fatehi and Manzari [12] is that it allows a stable time integration without
tuning any parameters. In order to perform a comparison as fair as possible, we will select parameters for the terms F Fe
ab
and F An
ab in order to achieve a similar effects with all of them.
If the linear EOS (6) is applied, then the term of Ferrari et al. [9] can be written as:
Fe rab
F Fe
ab = ( pb pa ) + f ( p , p )|r | , (46)
cs |rab | L a b ab
where Fe tuning parameter, and the correction term for the free surface fL (pa , pb ), have been appended to the model. The
term of Molteni and Colagrossi [10] and Antuono et al. [11] can be rewritten in the same way:
An h rab
F An
ab = 2 ( pb pa ) + f (p , p ) . (47)
cs |rab |2 L a b
Regarding the term of Fatehi and Manzari [12] and Hashemi et al. [13], we use the following time step:
x h
t = C f , (48)
h cs
such that the diffusive terms results:
C f h a rab
F Ha
ab = ( pb pa ) + fL ( pa , pb ) . (49)
c s 0 |rab |2
From Eqs. (46), (47) and (49) we can deduce the following parameters to achieve similar effects with all the analyzed
terms:
Fe = C f , (50)
Cf
An = . (51)
2
As it has been mentioned before, Antuono et al. [14] analyzed the An parameter doing a linear stability analysis propos-
ing an alternative way to tune the parameter (depending on the kernel used, and on the articial viscosity ), but here we
have established a condition based on the work of Fatehi and Manzari [12] to set the parameter based upon the stability of
J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736 8731
Fig. 2. Snapshots of the evolution of the dam break pressure eld at two different instants. First row: Simulation without viscosity or numerical diffusive
term. Second row: Corrected version of the term proposed by Ferrari et al. [9] ( Fe = 0.25). Third row: Term proposed by Molteni and Colagrossi [10] and
corrected by Antuono et al. [11] ( An = 0.125). Last row: Modied version of the term proposed by Fatehi and Manzari [12] and Hashemi et al. [13] ( Ha =
1.0).
Fig. 3. Energy dissipated along the simulations using each diffusive term. Two time lapses are highlighted with a gray background, indicating when the
wave impact and breaking are happening in the experiments.
a checkerboard problem, which consist in a lattice of particles in equilibrium where a punctual perturbation is introduced
to analyze if such perturbation is whether vanishing or diverging.
Fig. 4. Energy dissipated generated along the simulations using the models with the originally proposed parameter.
Fig. 5. Water surface elevation comparison between the models and the experimental data. Top-left: H1. Top-right: H2. Bottom-left: H3. Bottom-right: H4.
J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736 8733
Fig. 6. Wave height computation particles selection detail for probe H2 using the term F Ha
ab . The regular uid particles are drawn in black while the selected
ones to compute the uid height are highlighted in red color (The area of selection has been highlighted with a grey background). Top-left: t g/H = 7.43.
Top-right: t g/H = 7.62. Bottom-left: t g/H = 7.81. Bottom-right: t g/H = 8.6.
some energy is dissipated as discussed in Section 4.4. While the thermodynamic consistency close to the boundaries cannot
be granted, actually the global dissipated energy was positive for all the models. This result is similar to the obtained by
Antuono et al. [31] for the model F Anab .
From Fig. 3 it is clear that all the models behave similarly in regards the energy dissipation , while this is an unexpected
result considering that the model of Ferrari et al. [9] is slightly different from the other ones. This result reects the fact that
both F Fe An Ha
ab and F ab terms coecients has been tuned to match with the term F ab . If the originally recommended coecients
( = 1 and = 0.2) are used instead, the energy dissipation depicted in Fig. 4 is obtained, where it can be appreciated
Fe An
that both terms quickly diverge. In the case of the term F An ab , as Antuono et al. [14] already shown, it is caused by the
selected time step, which should be corrected with a Courant factor that depends on the An parameter. The model of
Hashemi et al. [13] explicitly takes into account the time step, becoming unnecessary to establish an additional Courant
factor for the time step.
In Fig. 5 we compare the wave heights obtained experimentally by Lobovsk et al. [36] at the positions H1, H2, H3 and
H4 to the various models. Good agreement is shown before the reected wave breaks and splashing starts by all the terms.
However, with the secondary wave the signal becomes noisy due to the ow fragmentation. In Fig. 6 the noise source can
be understood. The maximum height of the selected particles for H2 is found at the fragmented uid particles up to the
time t g/H = 8.6.
Fig. 7 depicts the pressure records computed at the sensors with each model, comparing them with the experimental
data. It can be seen that all the models have similar results. The agreement with the experimental data is reasonable con-
sidering the diculty of the problem. However the simulations falls out of the 95% condence interval for the pressure in
some time intervals.
It should be remarked again that there is good agreement between the three models when the parameters are tuned
using Eqs. (50) and (51).
8. Conclusions
In the present work some techniques to reduce the numerical noise generated in the pressure eld in WSPH have been
discussed, all based on the addition of a term (called the numerical diffusive term) to the conservation of mass equation. A
set of ve conditions that should be satised by these diffusive terms have been introduced, namely:
1. Mass conservation equation consistency, meaning that the term must vanish when the numerical solution converges to
the real one.
2. Intrinsic global mass conservation at a discrete level.
3. Consistency close to the free-surface, meaning that the terms are not affected by the Morris formula singularity.
4. Thermodynamic consistency, meaning that imposing the energy conservation (First Law of Thermodynamics) the diffusive
term is dissipating energy .
5. No parameter(s) tuning, meaning that the term should allow a stable time integration without needing to tune any
parameter, or set additional Courant conditions.
8734 J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736
Fig. 7. Sensors pressure record. Top-left: Sensor 1. Top-right: Sensor 2. Bottom-left: Sensor 3. Bottom-right: Sensor 4.
Several existing models have been discussed, tested against the ve conditions.
Each model discussed here has introduced new concepts and ideas:
It should be remarked that the equivalence between the mass rate of change equation used in R-SPH and the numerical
diffusive term has been proved (this idea was introduced by Ferrari et al. [9] without a formalization).
A practical application, consisting on a dambreak, has been simulated showing the improvements in the pressure eld
with the different models.
A relation between the models of Ferrari et al. [9] and Molteni and Colagrossi [10] with the model of Hashemi et al.
[13] has been established, demonstrating that the parameter used in the model of Molteni and Colagrossi [10] is related to
the Courant factor, and proposing therefore an alternative way to tune this parameter (Antuono et al. [14] already proposed
a way to tune the parameter, based on a linear stability analysis, with the main drawback that a new Courant factor
related with this term should be considered for the time step).
It is remarkable that, given the thermodynamic consistency close to the boundaries cannot be demonstrated (meaning
that we have not found a way to assert that the global energy dissipated is positive), in this practical application all the
models have generated a positive energy dissipation throughout the simulation.
J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736 8735
Future work
Although the Riemann solver based SPH model has been analyzed, it has been excluded from the practical application
due to the large differences and the added complexity with respect to all the other models. Therefore a more detailed
analysis must be performed.
It has been demonstrated that the added terms in the conservation of mass equations may result in energy variations
due to the changes in the compressibility energy component. It remains as future work to nd a diffusive term fullling
all the conditions introduced along the paper, including the null energy dissipation . However it seems that a new term
structure should be studied in order to achieve this goal.
During these work non-viscous uid has been modeled, however the coupling of the conservation of mass equation and
the momentum one for viscous uids must be considered in further publications. Also, a more detailed analysis regarding
the boundaries must be performed due to the Laplacian present in the formulation [28,50].
Acknowledgement
J.L. C.-P. has received funding from the Spanish Ministry for Science and Innovation under grant TRA2010-16988 Car-
acterizacin Numrica y Experimental de las Cargas Fluido-Dinmicas en el transporte de Gas Licuado and R.A.D. has been
supported by the Oce of Naval Research, Coastal Geosciences and Optics Program.
The authors are grateful to Antonio Souto-Iglesias from UPM-Madrid for the guidance and the long valuable discussions
during the paper preparation, and more specically the ones regarding the energy analysis of Section 4.4.
The authors would also express their gratitude to the anonymous reviewers, whose have signicantly contributed to
increase the paper quality.
Appendix A. Relation between R-SPH and WCSPH for the continuity equation
Along this appendix the relation between the mass rate of change equation proposed in the Riemann solver based SPH
(R-SPH) [6,7] and the numerical diffusive terms added to the continuity equation will be stablished. A similar analysis was
performed by Monaghan [15] for the momentum equation articial viscosity.
In R-SPH the domain is discretized in particles as well, but the following transport equation is used:
da
= a div(u(xa , t ) ), (A.1)
dt
where a is the volume of the particle a, such that
ma
a = . (A.2)
a
On the other hand a mass rate of change equation is proposed in the general form
d
( a a ) = a b Gb Wab . (A.3)
dt
bFluid
We are trying to relate this proposed model with the discussed along Section 3.1.2. First lets expand the derivative of
the volume in terms of the mass and the density:
da 1 dma m a d a
= 2 . (A.4)
dt a dt a dt
Solving for the density, the conservation of mass equation in the context of the R-SPH model reads
d a a d m a
= a div(u(xa , t ) ) + , (A.5)
dt ma dt
where the transport equation (A.1) has already been applied. Comparing this conservation of mass equation with Eq. (13) we
can relate the diffusive term inside the conservation of mass equation with the mass variation rate in the R-SPH approach:
a d m a
div(F )a = . (A.6)
ma dt
Regarding the term Gb , it can be related with the mass variation rate as well, because from Eq. (A.3) we can write
a d m a 1
= Gb Wab mb . (A.7)
ma dt
bFluid
b
Combining Eqs. (A.6) and (A.7) we can relate the R-SPH model and the numerical diffusive term added to the conserva-
tion of mass equation:
F b = Gb . (A.8)
Indeed the term proposed for the mass rate of change equation in R-SPH is the same than the introduced in the conservation
of mass equation. This idea had been hinted by Ferrari et al. [9], who did not formalized it.
8736 J.L. Cercos-Pita et al. / Applied Mathematical Modelling 40 (2016) 87228736
References
[1] J. Monaghan, Smoothed particle hydrodynamics and its diverse applications, Annu. Rev. Fluid Mech. 44 (1) (2012) 323346.
[2] J.J. Monaghan, Smoothed particle hydrodynamics, Rep. Progr. Phys. 68 (2005) 17031759.
[3] J. Monaghan, Simulating free surface ows with SPH, J. Comp. Phys. 110 (2) (1994) 39406.
[4] S. Shao, E.Y. Lo, Incompressible SPH method for simulating newtonian and non-Newtonian ows with a free surface, Adv. Water Resour. 26 (7) (2003)
787800.
[5] E.S. Lee, C. Moulinec, R. Xu, D. Violeau, D. Laurence, P. Stansby, Comparisons of weakly compressible and truly incompressible algorithms for the SPH
mesh free particle method, J. Comput. Phys. 227 (18) (2008) 84178436.
[6] E.F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics., Springer, Berlin, 1997.
[7] J. Vila, On particle weighted methods and Smooth Particle Hydrodynamics, Math. Models Methods Appl. Sci. 9 (2) (1999) 161209.
[8] A. Raee, K.P. Thiagarajan, J.J. Monaghan, SPH simulation of 2D sloshing ow in a rectangular tank, in: Proceedings of the International Offshore and
Polar Engineering Conference (ISOPE), The International Society of Offshore and Polar Engineers (ISOPE), 2009.
[9] A. Ferrari, M. Dumbser, E.F. Toro, A. Armanini, A new 3d parallel {SPH} scheme for free surface ows, Comput. Fluids 38 (6) (2009) 12031217.
[10] D. Molteni, A. Colagrossi, A simple procedure to improve the pressure evaluation in hydrodynamic context using the SPH, Comput. Phys. Commun.
180 (2009) 861872.
[11] M. Antuono, A. Colagrossi, S. Marrone, D. Molteni, Free-surface ows solved by means of SPH schemes with numerical diffusive terms, Comput. Phys.
Commun. 181 (3) (2010) 532549.
[12] R. Fatehi, M. Manzari, A remedy for numerical oscillations in weakly compressible smoothed particle hydrodynamics, Int. J. Numer. Methods Fluids 67
(9) (2011) 11001114.
[13] M. Hashemi, R. Fatehi, M. Manzari, Sph simulation of interacting solid bodies suspended in a shear ow of an oldroyd-b uid, J. Non-Newtonian Fluid
Mech. 166 (21) (2011) 12391252.
[14] M. Antuono, A. Colagrossi, S. Marrone, Numerical diffusive terms in weakly-compressible SPH schemes, Comput. Phys. Commun. 183 (12) (2012)
25702580.
[15] J. Monaghan, Sph and riemann solvers, J. Comp. Phys. 136 (1997) 298.
[16] J.R. MacDonald, Some simple isothermal equations of state, Rev. Mod. Phys. 38 (1966) 669679.
[17] L. Lucy, A numerical approach to the testing of the ssion hypothesis, Astron. J. 82 (1977) 10131024.
[18] R. Gingold, J. Monaghan, Smoothed particle hydrodynamics: theory and application to non-spherical stars, Mon. Not. Roy. Astron. Soc. (MNRAS) 181
(1977) 375389.
[19] A. Colagrossi, M. Antuono, D.L. Touz, Theoretical considerations on the free-surface role in the Smoothed-particle-hydrodynamics model, Phys. Rev. E
(Stat., Nonlinear, Soft Matter Phys.) 79 (5) (2009) 056701.
[20] P. Campbell, Some New Algorithms for Boundary Value Problems in Smooth Particle Hydrodynamics, Technical Report, Defende Nuclear Agency, 1989.
[21] M. De Leffe, D. Le Touz, B. Alessandrini, Normal ux method at the boundary for SPH, in: Proceedings of the 4th SPHERIC, 2009, pp. 149156.
[22] M. Ferrand, D.R. Laurence, B.D. Rogers, D. Violeau, C. Kassiotis, Unied semi-analytical wall boundary conditions for inviscid, laminar or turbulent
ows in the meshless SPH method, Int. J. Numer. Methods in Fluids 71 (4) (2013) 446472.
[23] F. Maci, L.M. Gonzlez, J.L. Cercos-Pita, A. Souto-Iglesias, A boundary integral SPH formulation. Consistency and applicationsto ISPH and WCSPH, Progr.
Theor. Physics 128 (3) (2012) 439462.
[24] J.J. Monaghan, Smoothed particle hydrodynamic simulations of shear ow, Mon. Notices R. Astron. Soc. 365 (2005) 199213.
[25] J.P. Morris, P.J. Fox, Y. Zhu, Modeling low Reynolds number incompressible ows using SPH, J. Comput. Phys. 136 (1997) 214226.
[26] A. Colagrossi, M. Antuono, A. Souto-Iglesias, D. Le Touz, Theoretical analysis and numerical verication of the consistency of viscous smoothed-parti-
cle-hydrodynamics formulations in simulating free-surface ows, Phys. Rev. E 84 (2011) 26705+.
[27] F. Maci, J.M. Snchez, A. Souto-Iglesias, L.M. Gonzlez, WCSPH viscosity diffusion processes in vortex ows, Int. J. Numer. Methods Fluids 69 (3)
(2012) 509533.
[28] F. Maci, M. Antuono, L.M. Gonzlez, A. Colagrossi, Theoretical analysis of the no-slip boundary condition enforcement in SPH methods, Progr. Theor.
Phys. 125 (6) (2011) 10911121.
[29] N.J. Quinlan, M. Lastiwka, M. Basa, Truncation error in mesh-free particle methods, Int. J. Numer. Methods Eng. 66 (13) (2006) 20642085.
[30] A. Amicarelli, J.-C. Marongiu, F. Leboeuf, J. Leduc, J. Caro, SPH truncation error in estimating a 3D function, Comput. Fluids 44 (1) (2011) 279296.
[31] M. Antuono, S. Marrone, A. Colagrossi, B. Bouscasse, Energy balance inthe -SPH scheme, Comput. Methods Appl. Mech. Eng. 289 (2015) 209226.
[32] S.K. Godunov, A difference method for numerical calculation of discontinuous solutions of the equations of hydrodynamics, Matematicheskii Sbornik
89 (3) (1959) 271306.
[33] P. Koukouvinis, J. Anagnostopoulos, D.E. Papantonis, An improved MUSCL treatment for the SPH-ALE method: comparison with the standard SPH
method for the jet impingement case, Int. J. Numer. Methods Fluids 71 (2013) 11521177.
[34] D. Molteni, C. Bilello, Riemann solver in SPH, Memorie della Soc. Astron. Italiana Suppl. 1 (2003) 36.
[35] V.V. Rusanov, Calculation of Interaction of Non-Steady Shock Waves with Obstacles, NRC, Division of Mechanical Engineering, 1962.
[36] L. Lobovsk, E. Botia-Vera, F. Castellana, J. Mas-Soler, A. Souto-Iglesias, Experimental investigationof dynamic pressure loads during dam break, J. Fluids
Struct. 48 (2014) 407434.
[37] T. Lee, Z. Zhou, Y. Cao, Numerical simulations of hydraulic jumps in water sloshing and water impacting, J. Fluids Eng. 124 (1) (2002) 215226.
[38] A. Ritter, Die Fortpanzung de Wasserwellen, Z. Ver. Deutscher Ingenieure 36 (33) (1892) 947954.
[39] A. Schoklitsch, ber Dambruchwellen, Sitz. Knigliche Akad. Wiss. 126 (IIa) (1917) 14891514.
[40] I.B. Eguiazaroff, Regulation of the water level in the reaches of canalized rivers, in: Proceedings XVI International Congress of Navigation, 1935.
[41] J.C. Martin, W.J. Moyce, Part IV. An experimental study of the collapse of liquid columns on a rigid horizontal plane, Philos. Trans. R. Soc. Lond. Ser. A,
Math. Phys. Sci. 244 (882) (1952) pp.312324.
[42] R.F. Dressler, Comparison of theories and experiments for the hydraulic dam-break wave, Int. Ass. Sci. Hydrol. Publi. 38 (3) (1954) 319328.
[43] P.K. Stansby, A. Chegini, T.C.D. Barnes, The initial stages of dam-break ow, J. Fluid Mech. 374 (1998) 407424.
[44] S. Marrone, M. Antuono, A. Colagrossi, G. Colicchio, D.L. Touz, G. Graziani, Delta-SPH model for simulating violent impact ows, Comput. Methods
Appl. Mech. Eng. 200 (13-16) (2011) 15261542.
[45] M. Asai, A.M. Aly, Y. Sonoda, Y. Sakai, A stabilized incompressible SPH method by relaxing the density invariance condition, J. Appl. Math. 2012 (2012)
Article ID 139583, doi:10.1155/2012/139583.
[46] H. Wendland, Piecewise polynomial, positive denite and compactly supported radial functions of minimal degree, Adv. Comput. Math. 4 (4) (1995)
389396.
[47] F. Maci, A. Colagrossi, M. Antuono, A. Souto-Iglesias, Benets of using a Wendland kernel for free-surface ows, in: Proceedings of the 6th ERCOFTAC
SPHERIC Workshop on SPH Applications, Hamburg University of Technology, 2011, pp. 3037.
[48] W. Dehnen, H. Aly, Improving convergence in smoothed particle hydrodynamics simulations without pairing instability, Mon. Notices R. Astron. Soc.
425 (2) (2012) 10681082.
[49] M. Antuono, A. Colagrossi, S. Marrone, C. Lugni, Propagation of gravity waves through an SPH scheme with numerical diffusive terms, Comput. Phys.
Commun. 182 (4) (2011) 866877.
[50] L. Gonzlez, J. Cercos-Pita, F. Maci, On the boundary condition enforcement in SPH methods, in: Proceedings of the 7th ERCOFTAC SPHERIC Workshop
on SPH Applications, Monash University, INSEAN, University of Pavia, 2012, pp. 303310.