Computers & Fluids 50 (2011) 24–34

Contents lists available at ScienceDirect

Computers & Fluids

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

A multiphase compressible model for the simulation of multiphase flows

Jean-Paul Caltagirone a, Stéphane Vincent a,⇑, Céline Caruyer b
a
Tranferts Ecoulements Fluides Energétique (TREFLE), UMR CNRS 8508, ENSCPB, Université Bordeaux 1, 16 Avenue Pey-Berland, 33607 Pessac Cedex, France
b
CEA, DAM, Le Ripault, F-37260 Monts, France

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

Article history: A compressible model able to manage incompressible two-phase flows as well as compressible motions
Received 22 July 2010 is proposed. After a presentation of the multiphase compressible concept, the new model and related
Received in revised form 13 May 2011 numerical methods are detailed on fixed structured grids. The presented model is a 1-fluid model with
Accepted 17 June 2011
a reformulated mass conservation equation which takes into account the effects of compressibility.
Available online 12 July 2011
The coupling between pressure and flow velocity is ensured by introducing mass conservation terms
in the momentum and energy equations. The numerical model is then validated with four test cases
Keywords:
involving the compression of an air bubble by water, the liquid injection in a closed cavity filled with
1-Fluid compressible model
Two-phase flows
air, a bubble subjected to an ultrasound field and finally the oscillations of a deformed air bubble in
VOF melted steel. The numerical results are compared with analytical results and convergence orders in space
Surface tension are provided.
Interface Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction finite difference method with a second order accuracy in space. Cai-
den et al. [3] detailed a numerical method for modeling two phase
The goal of this work is to build a general model in order to sim- flows consisting of separate compressible and incompressible re-
ulate, in the near future, the interaction between a plasma flow and gions. They use the level-set method to track the interface between
a liquid jet within the framework of plasma spraying. the compressible and the incompressible regions, as well as the Ghost
As the plasma flow is compressible, a compressible model fluid method to design accurate discretizations across the interface.
[12,29] has to be built to describe at the same time the plasma Nourgaliev et al. [13] presented a pseudo-compressibility method
and the two phase character of the problem. A mixed model capa- for the numerical simulation of incompressible multifluid flows. This
ble of simultaneously managing compressible plasma motions and approach is based on artificial compressibility, the concept being to
the incompressible liquid jet features of the two-phase flow is re- add a time derivative of pressure into the continuity equation, thus
quired. As a first step, the present work aims at proposing a model providing a direct coupling between pressure and flow velocity. The
able to take into account two-phase motions on a fixed mesh as the authors developed an adaptive characteristics-based matching
liquid interface strongly varies over time, as well as the low Mach method for compressible multiphase simulation [14].
compressible character of the flow in the gas phases and the Concerning weakly compressible flows, which are not formulated
incompressible character of the liquid phase. A 1-fluid model [8] on a 1-fluid model, ‘‘Low Mach number’’ models are established from
is used for its ability to handle two phase flows involving separated compressible Navier–Stokes equations and asymptotic expansions of
phases on fixed grids. unknowns [11,32]. Stable resolution methods are difficult to set up
Some works on 1-fluid models which are compressible, multifluid for low Mach numbers. The model presented in this article is not a part
and made of separated phases exist in the literature. A unified numer- of the category of ‘‘low Mach number’’ models previously presented
ical procedure is proposed as a solver both for compressible and but a general compressible model for multiphase flows.
incompressible fluids by Yabe and Yuan [36]. The predictor–corrector The coupling between pressure and flow velocity can either be
scheme is adopted to correctly calculate pressure and to account for of type predictor–corrector [3,36] or artificial compressibility
compressible effects. The governing equations are separated into [13,14]. In all the methods quoted above, the meshes are fixed
advection and non-advection terms by the time splitting method. and structured. These numerical approaches avoid remeshing but
The advection is simulated by the CIP method with third order spatial require specific functions to track interfaces such as Level Set
accuracy, while the non-advection terms are discretized by a centered [15], Volume Of Fluid [37] or Front Tracking [24]. A pseudo com-
pressibility method, fixed structured grids and Volume Of Fluid
⇑ Corresponding author. (VOF) interface tracking methods will be also utilized in the
E-mail address: [email protected] (S. Vincent). present work.

0045-7930/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compfluid.2011.06.011
 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34 25

In this paper, the motion equations dedicated to a deterministic and analyzed in Section 4 for Mach numbers less than 1. Conclu-
description of a compressible two-phase flow are presented based sions and perspectives are finally drawn in Section 5.
on the 1-fluid model. A reformulated mass conservation equation
is derived by developing the partial derivative of the pressure 2. Multiphase compressible concept
according to time in function of temperature and density. Specific
terms are added in the momentum and energy equations to ac- 2.1. Motivation
count for mass conservation and velocity pressure coupling in
compressible flows. The present model is devoted to the prediction of macroscopic
The existing compressible models are based on conservative fields such as pressure, temperature and velocity, when different
formulation and need hyperbolic schemes to be approximated. equations of state govern the fluids in interaction. Each fluid is gov-
These models generate numerical difficulties when the Mach num- erned by its own equation of state and is characterized by temper-
ber tends to zero because q is of the order of M2. In our model, the ature and pressure dependent on physical properties, such as
objective is to develop a two-phase compressible model in a non- density, viscosity, thermal conductivity, specific heat, dilatation
conservative expression so that P and V, which is continuous and compression coefficients. The surface tension coefficient of
through the interface R (contrary to q and qV), are the variables. the fluids is taken as known.
This will allow us to obtain a stable and strong formulation for The main difficulty to overcome in a multiphase 1-fluid ap-
mixed incompressible and compressible flows in particular for a proach is that the pressure and temperature are dependant on
Mach number less than 1. In this model, based on a non-conserva- the physical properties, but also on the discontinuous phase func-
tive approach, all fluids can be considered as compressible, more- tion C. The nature of fluid induces huge density variations
over all numerical parameters are determined by the physics (104 kg m3) while temperature variations for instance result in
except the choice of resolution schemes. This approach is original very weak changes in density (105 kg m3). These two scales of
because the coupling between pressure and flow velocity is not density fluctuation constrain the global model in a noticeably dif-
based on projection or artificial compressibility types but on a ferent way. Thus, new models and numerical methods unifying
method inspired of an augmented Lagrangian type. these effects are at stake.
To establish a two-phase compressible model, we start from the The only unknowns of the incompressible model are the velocity
incompressible single fluid model, also called 1-fluid, as described and the temperature, since the pressure is a numerical parameter
by Kataoka [8]: allowing the system to meet local volume compression or the
incompressibility constrain rV = 0. The other quantities are the
8
physical characteristics among which the viscosity l, the thermal
< r V ¼ 0
>
   
q @V þ V  rV ¼ rp þ qg þ r  l rV þ rt V þ F ST ð1Þ conductivity k, the specific heat cp and the density q are very com-
@t
>
: @C mon. They are independent of the Navier–Stokes equations resolu-
þ V  rC ¼ 0
@t tion, and can be calculated according to the nature of fluid,
equations of state, or color function C in multiphase flows. Extending
where V is the velocity, q the density, t the time, p the pressure, g a 1-fluid model to compressible flows requires building a reformu-
the gravity vector, l the dynamic viscosity coefficient, FST the sur- lated mass conservation equation to account for incompressibility
face tension force and C the phase function. The 1-fluid Navier– in the liquid and dilatation in the gas or the plasma.
Stokes equations are solved in the whole field D. An equivalent fluid
containing all the phases, i.e. D1 and D2, is built thanks to a phase 2.2. Mass conservation
function C which depends on time and space. The value of the phase
function C is 1 in the liquid and 0 elsewhere (Fig. 1). The interface R Generally speaking, the flow is compressible (rV – 0) and an
between the liquid phase and the rest of the fluids is defined as equation of state must be applied to the fluids. For an interfacial
C = 0.5. flow, the mass conservation equation follows:
The concept of fictitious domains consists in considering a do- Z Z Z Z Z Z  
d @q
main of interest including several sub-domains associated with q dv ¼ þ r  ðqVÞ dv
different fluids as a single equivalent fluid with physical properties
dt D D @t
Z Z
which are variable in time and space. The interest of this approach þ ½½qðV  WÞ  N ds ð2Þ
is to consider a single equation system for each physical variable R
such as velocity, temperature or concentration.
and
The article is organized as follows: Section 2 presents the mul-      
tiphase compressible model in general and in the particular case ½½qðV  WÞ  N ¼ qD1 VD1  W  N  qD2 VD2  W  N ð3Þ
where the Mach number is less than 1. The numerical methods
are exposed in Section 3. Finally, four test cases are performed where W is the interface velocity, N the normal to the interface R
which separates the sub-domains D1 and D2. This can be written
in local formulation:
(
@q
@t
þ r  ðqVÞ ¼ 0 in D ¼ D1 [ D2 =R
ð4Þ
½½qðV  WÞ  N ¼ 0 on R

If the relative interface velocity is Vr = (V  W), the jump condition

associated to the mass conservation gives:
½½qV r  ¼ 0 ð5Þ
The mass flow rate qVr = m through a discontinuity surface R is con-
tinuous. If the interface is supposed to be a contact surface, with no
Fig. 1. Phase function C and fictitious domains D1 and D2 in the computational mass flux through the discontinuity surface R, i.e. no phase change
domain D. occurs, the jump equation can be simplified:
26 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34

½½V  N ¼ 0 on R ð6Þ p V tc T
p ¼ ; V ¼ ; t ¼ ; T ¼ ð12Þ
which establishes the continuity of the normal component of the qV 20 V0 H dT
velocity. In our case, the tangent component is also supposed to and the non-dimensional equation for pressure becomes:
be continuous, which is verified experimentally using a macro- 
scopic approach. @p 1 Ga dT
¼  r  V  MV  rp þ 2  ð13Þ
The mass conservation equation in D can be reformulated as an @t  M M dt
equation for the pressure by developing the derivative of the pres- Noticing that dT/T  dp/p  M2 for weakly compressible flows, the
sure according to time in function of temperature and density: last term of Eq. (13) is of the first order O(1). This can be shown
 
dp @p dT @p  dq by considering the following equations:
¼  þ 
dt @T q dt @ q T dt dT
 ¼ ar2 T ð14Þ
dp b dT @p  dq dt
¼ þ  @p 1 1 1
dt v dt @ q dt
T T ¼  r  V  MV  rp þ Gar2 T  ð15Þ
dq
@t M M Pe
Using the compressible mass conservation dt
þ qr  V ¼ 0, we
obtain: where a is the thermal diffusivity and Pe ¼ HVa 0 is the Peclet number.
 For low Mach number and highly inertial flows, the relations (16)
dp b dT @p  are generally satisfied:
¼ þ  ðqr  VÞ
dt vT dt @ q T
M ! 0; Ga  1; Pe  1 ð16Þ
and with
Thus, the two last terms in Eq. (13) can be neglected in Mach num-
dp @p bers limited to zero, including highly compressible flows, involving
¼ þ V  rp
dt  @t fluids with a high compressibility coefficient. The resulting equation
@p  1 for pressure is:
q  ¼
@ q T vT @p 1
¼  r  V ð17Þ
we finally obtain
@t  M
which is written in real variables, after a semi-discretization in
@p b dT 1
¼  r  V  V  rp ð7Þ time:
@t vT dt vT
Dt
From a numerical perspective, after a semi-discretization in time, it pnþ1 ¼ pn  rV ð18Þ
vT
reads:
where n is a time index such as t = nDt and Dt is the time step.
!
1 b dT
nþ1 The density is updated as a function of the temperature and
pnþ1 ¼ pn  Dt r  Vnþ1 þ Vnþ1  rpn  ð8Þ pressure:
vT vT dt
q ¼ f ðp; TÞ ð19Þ
where b is the coefficient of thermal expansion and vT is the iso- The multiphase compressible model is based on the formulation of
thermal compressibility coefficient. the mass conservation equation which will be partially or com-
The relative importance of the different terms in Eq. (7) is in pletely integrated in the 1-fluid momentum equation. This equation
function to physical properties and constraints. It is achieved replaces rV = 0 in the 1-fluid model even for incompressible flows,
through its non-dimensional expression, involving the following which will be related to high values of vT.
non-dimensional variables:
2.3. Numerical treatment of the momentum conservation
p V tV 0 T
p ¼ ; V ¼ ; t ¼ ; T ¼ ð9Þ
qV 20 V0 H dT
The goal is to interpret Eq. (7) in order to establish the numer-
ical relation between mass and momentum conservation equation
where V0 is a reference velocity, dT a reference temperature, H a
in the Navier–Stokes equations. The momentum equation of (1) in
characteristic length. pffiffiffiffiffiffiffiffiffiffiffiffiffiffi semi-discrete form is written as:
By noticing that the sound velocity is c ¼ Cte 1=qvT , the Mach
!
number is M ¼ Vc0 and by introducing the dimensionless variables ðVnþ1  Vn Þ  
in (7), we obtain: q þ Vn  rVnþ1 ¼ rpnþ1 þ qg þ r  l rVnþ1
Dt
 
@p 1 bdT dT þrt Vnþ1 þ F TS
¼  2 r  V  V  rp þ 2  ð10Þ ð20Þ
@t  M M dt
The Gay-Lussac number Ga = bdT is introduced, this parameter With Eq. (8), we obtain:
quantifies the density differences of the working fluid under the !
temperature field.
The non-dimensional expression becomes:
q
ðVnþ1  Vn Þ
Dt
þ Vn  rVnþ1 ¼ r ð pn 
s
vT
r  Vnþ1  sVnþ1  rpn
!

sb dT nþ1  
@p 1 Ga dT
¼  2 r  V  V  rp þ 2  ð11Þ þ þ qg þ r  l rVnþ1
@t  M M dt vT dt

In the dimension analysis (9), the time scale of the pressure þrt Vnþ1 þ F TS ð21Þ
fluctuations is taken to be H/V0. In order to capture the pressure
waves, a time scale s defined as Hc seems more appropriate (where This formulation is self-consistent since the local constraint
c is the sound velocity). The dimensionless set of parameters is accounting for the degree of fluid compressibility is directly in-
written again as: cluded in the equation. The parameter k = s/vT behaves as a
 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34 27

compression resistance which depends on the time scale of the Dt

pnþ1 ¼ pn  r  Vnþ1 ð27Þ
phenomenon. Indeed, a medium behaves differently to low and vT
high frequency stresses which makes the time scale a crucial no-
tion in this formulation. The total mass conservation issue differs where pressure is the thermodynamic pressure or the dynamic
drastically from the incompressible approaches and is a matter of pressure if the total pressure only depends on time. In fact r = w
volume mass transfers in D or surface mass transfers through the and no minimization procedure is required. The physical parame-
R boundary.For a low Mach number assumption, using Eq. (18), ters involved in the equation system are the isothermal compress-
Eq. (20) becomes: ibility vT, expansion coefficients b, imposed or calculated by a
state equation, and the density q(p, T) calculated by a state
! equation.
ðVnþ1  Vn Þ s
q þ Vn  rVnþ1 ¼ r pn  r  Vnþ1 þ qg
Dt vT
   2.4. Energy conservation
þr l rVnþ1 þ rt Vnþ1
For the sake of generality, the formulation of non-conservative
þ F TS ð22Þ forms of conservation equations in the framework of compressible
Eqs. (22), coupled with (18), provide a modeling framework for multiphase flows is also proposed for the energy equation, even if
two-phase compressible isothermal flows. This version of the com- no example will be investigated in this article. The energy equation
pressible one-fluid model is simpler to implement than the com- is written in its non-conservative form with compressibility terms:
plete formulation and is useful for many problems. Several @T dp
examples are provided in the validation section. qcp þ V  rT ¼ r  ðkrTÞ þ bT þqþ/ ð28Þ
@t dt
The Stokes Eq. (23), is deduced from the continuum approach
assuming that the volumic viscosity coefficient is equal to zero: where q is the volumic production and / the dissipation function.
Another form of the energy equation is obtained by developing
3k þ 2l ¼ 0 ð23Þ
the derivative of the pressure according to time in function of tem-
The compression viscosity coefficient k is unknown and the validity perature T and density q:
of this relation is limited to monoatomic gases. For other fluids,  
especially for incompressible liquids, it is not demonstrated that
dp @p dT @p  dq
¼  þ  ð29Þ
this coefficient has a physical sense [4]. For a compressible and an dt @T q dt @ q T dt
incompressible flow as well, the pressure gradient in the Navier–
with ddtq ¼ qr  V according to the compressible mass conservation.
Stokes equations can be associated to the term r(k rV Id) = r(k @p
For the term @T , the thermodynamic expression including the
rV) as:
specific heat is used:
rðp  kr  VÞ ð24Þ 
@p  T @q @p
bT ¼ ¼ qðcp  cv Þ ð30Þ
@T q q @T P @T q
Incompressibility, usually formalized by the divergence free condi-
tion rV = 0, can here be associated to an infinite compression vis- With expressions (29) and (30), the term bT dp
dt
of (28) is written
cosity coefficient. The association of this coefficient to a viscosity as:
effect is rather artificial. Indeed, even if its dimension is similar to
that of a viscosity, it is closely related to the fluid compressibility dp dT bT
bT ¼ qðcp  cv Þ  rV
coefficient. In our numerical formulation, the parameter which dt dt vT
plays the same role as k is called w and is written as: @T bT
¼ qðcp  cv Þ þ V  rT  rV ð31Þ
s @t vT
w¼ ð25Þ
vT Finally, the new formulation of the conservation energy equa-
tion is obtained by introducing (31) in (28):
where s is the time scale of the physical phenomenon, which ex-
tends from a wave propagation with a small time scale to large fluid @T bT
qcv þ V  rT ¼ r  ðkrTÞ  rVþqþ/ ð32Þ
motions. For instance, at a fixed time scale and in normal pressure @t vT
and temperature conditions, the compression resistance of water is
A phase change source term Fp can be added in (32) to take into ac-
105 higher than for air. In a classical approach, air would be consid-
count the related interface jump conditions involving latent heat.
ered as a compressible fluid, for which k = 2/3l, whereas the
For more details, the reader can consult the work of Voller and
water would be strictly incompressible with a non-physical k coef-
Swaminathan [34]. This specific physical aspect will not be consid-
ficient and would require a second step to ensure the divergence-
ered in the present article.
free constraint. Moreover, the assumption k = 2/3l for air would
The form (32) of the conservation energy equation is more
result in a highly compressible non-physical behavior forcing the
interesting than (28) as the derivative of the pressure according
fluid to collapse under its own weight.
to time has been replaced by a divergence term of velocity in the
The simplified one-fluid model (22-18) can be interpreted as a
second member. Furthermore, now the equation depends on the
generalization of the formulation obtained with the augmented
constant volume specific heat and not the constant pressure spe-
Lagrangian method [5,31] for incompressible flows which calcu-
cific heat.
lates the pressure as follows, after a semi-discretization in time:

pnþ1 ¼ pn  r r  Vnþ1 ð26Þ 2.5. Comparison with classic ‘‘low Mach number’’ model

where r is a numerical parameter which allows to locally ensuring The compressible Navier–Stokes equations in their conservative
rV = e with the chosen accuracy e ? 0. In this case, the pressure and non-dimensional form are given in [33]. The term M12 present in
p can be seen as a compressibility storage variable. The extension the momentum equation complicates the resolution for low Mach
of this method to compressible flows results in the expression: numbers as this term tends to 0 for weak flow velocities. This
28 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34

8  nþ1 n   
difficulty is avoided by performing asymptotic expansions of the >
> q V V
þ V n
 rV nþ1
¼ r pn
 s r  Vnþ1 þ qg
unknowns of the Navier–Stokes equations [11].
>
> D t v T
>
>   
>
> þr  l rVnþ1 þ rt Vnþ1 þ F TS
>
>
>
>
qðx; t; eÞ ¼ q0 ðx; tÞ þ eq1 ðx; tÞ þ e2 q2 ðx; tÞ þ Oðe3 Þ > pnþ1 pn
>
> 1 nþ1
< Dt ¼  vT r  V
>
Tðx; t; eÞ ¼ T 0 ðx; tÞ þ eT 1 ðx; tÞ þ e2 T 2 ðx; tÞ þ Oðe3 Þ  nþ1 n  ð35Þ
>
> qC v T DT þ ðVnþ1  rÞT nþ1 ¼ r  ðkrT nþ1 Þ
Vðx; t; eÞ ¼ V0 ðx; tÞ þ eV1 ðx; tÞ þ e2 V2 ðx; tÞ þ Oðe3 Þ >
>
t
>
>
>
>
nþ1
 bTv r  Vnþ1 þ q þ /
pðx; t; eÞ ¼ p0 ðx; tÞ þ ep1 ðx; tÞ þ e2 p2 ðx; tÞ þ Oðe3 Þ >
>
>
>
T
>
> dC
¼ @C
þ V  rC ¼ 0
>
>
with e = cM2. : dt @t
These asymptotic expansions are substituted into the com-
q ¼ f ðp; T; CÞ; k ¼ f ðT; CÞ; l ¼ f ðT; CÞ; C p ¼ f ðT; CÞ
pressible Navier–Stokes equations in conservative formulation This equation system will be considered in the rest of the article.
and the powers of M are equated. The terms of order M2 in the The model is characterized by time and space dependent
momentum equation imply that rp0 = 0. The quantity p0 is thus physical properties q, l, b, vT which are functions of the phase
constant in space. function C. More than two fluids can be compressible and dilat-
At the zero order, ‘‘low Mach number’’ Navier–Stokes equations able. They have their own state equation. The local constraint of
are obtained: compression resistance is implicitly taken into account as a func-
8 @q tion of the phenomenon time scale. The initial conservation
>
>
0
þ r  ðq 0 V 0 Þ ¼ 0 equations are not subjected to any assumption. Any time scale
> @t
>
>
> @ q0 V0 can be resolved, extending from wave propagation in a contin-
< @t þ r  ðq0 V0 V0 Þ ¼ rp1
>
  uum to common flows on large time scales. It aims at unifying
~ ð33Þ
>
> þ 1
r   2
lð rV Þ  ~
I þ l ð rV þ r t
V Þ the resolution of the equations in compressible configurations
>
> Re 3 0 0 0
>
> with Mach number Ma < 1 or incompressible flow configurations,
>
: @p0 c
þ cr  ðp0 V0 Þ ¼ RePr r  ðlrT 0 Þ in the presence of several phases or not. It enables resolving
@t
the propagation and more generally the constraints and the mo-
Preconditionners and stable resolution methods are necessary to tions in an elastic medium through the definition of the viscosity
solve (33). Their development and implementation is rather compli- coefficients l and k from Young’s modulus and Poisson’s
cated. It is based on a low Mach preconditioning of iterative solvers coefficient. The extension to fluid–structure interaction is
[33]. These models are used for low compression rate and low straightforward.
speed.
The model presented in this article for Ma < 1 is not a low Mach 3. Numerical methods
compressible model, no asymptotic expansions are led. Our model
is more general because it can support a high compression rate and 3.1. General methodologies
a strong density variation even with very low speeds. A simplified
model can be obtained by assuming that Ma ? 0 but without Numerical methodology is of great importance in the use of the
developing pressure according to the Mach number M as in the SCOF model presented here. Indeed, the explicit resolution of each
low Mach number approach. Moreover, the 1-fluid compressible velocity component, or the use of any prediction–correction meth-
model can be used in its general form to treat various compressible od, does not involve any physical constraint in the prediction step.
configurations with Ma > 1. Specific advection schemes devoted to This results in a divergence which not only contains purely physi-
the hyperbolic character of the non-linear term in the momentum cal data but also numerical errors due to the resolution. The
equations must be used in this case [7]. achievement of velocity and pressure fields in agreement with
the physics of the phenomenon is only possible through the cou-
2.6. Synthesis of numerical formulations pled resolution of all the velocity components with the simulta-
neous imposition of the compression resistance constraint. The
In the general case, the discretized in time equation system of subsequent determination of the pressure and density is just an
the Compressible One-Fluid (COF) model is written as: update associated with the implicit constraint already applied in
the equation. The determination of the couple of unknowns (V, p)
8  nþ1 n 
>
> q V V
þ Vn  rVnþ1 ¼ at the time step n + 1 only requires the knowledge of the density,
>
> Dt
viscosity, expansion and compressibility coefficients at the time
>
>  
>
> r pn  s r  Vnþ1  sVnþ1  rpn þ sb dT n
>
> vT vT dt step n. Their values are calculated through the state equation of
>
>
>
>    each fluid, which are located by the phase function C(x, t) devoted
>
> nþ1
> þqg þ r  l rV
> þ rt Vnþ1 þ F TS to the interface description.
<
nþ1 n n ð34Þ After a semi-discretization in time, Eqs. (35) become for isother-
>
>
p p
Dt
¼  v1 r  Vnþ1  Vnþ1  rpn þ vb dT mal flows:
> T T dt
>
>  
>
> nþ1 n nþ1 nþ1 8  nþ1 n   
>
> qC v T DT t
þ ðV  rÞT ¼ r  ðkrT nþ1 Þ > qn V DV þ Vn  rVnþ1 ¼ r pn  wr  Vnþ1 þ qgþ
>
> >
> t
>
> >
>   
>
>
nþ1
 bTv r  Vnþ1 þ q þ / dC ¼ @C þ V  rC ¼ 0 >
> nþ1 t nþ1
>
> dt @t >
> þ r  l n
r V þ r V þ F TS
>
: T >
>
>
>
q ¼ f ðp; T; CÞ; k ¼ f ðT; CÞ; l ¼ f ðT; CÞ; C p ¼ f ðT; CÞ >
<p nþ1 n
¼ p  wr  V nþ1

@C ð36Þ
where s = Dt in a discrete formulation. In Eq. (34), the compressible >
> @t
þ V  rC ¼ 0
>
>
effects are implicitly act on the interface evolutions through the >
> q ¼ C nþ1 q1 þ ð1  C nþ1 Þq0
nþ1
>
>
velocity field which does not verify rV. >
>
> lnþ1 ¼ C nþ1 l þ ð1  C nþ1 Þl
>
When the Mach number tends towards zero, the previous >
> 1 0
: nþ1
equation system, called Simplified Compressible One-Fluid (SCOF) vT ¼ C nþ1 v1T þ ð1  C nþ1 Þv0T
model, becomes:
 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34 29

The numerical parameter w is known, in fact w ¼ vDt with s = Dt 4.1. Test case 1: Compression of an air bubble
T
which is chosen equal to the numerical discretization time step
used in the Navier–Stokes equations and vT is the isothermal com- 4.1.1. Presentation of the case
pressibility coefficient of the fluid. This approach is unsteady and An air bubble, with a radius of R = 30 mm, is compressed by
directly adapts the numerics to the physics according to the time water in a closed cavity of L = 0.1 m sides (Fig. 2). The water flow
scale s. For wave compressible propagation, s is chosen to 108 s, is very slow, with V0 = 0.0025 m s1. The gravity and surface ten-
whereas for weakly compressible problems, s can be equal to sev- sion forces are not taken into account.The Reynolds number is
eral orders of magnitude larger in accordance to the physical prob- Re = 250.
lem, i.e. the time scale of the interface advection for example.
This 1-fluid approach is fully independent on the way the inter- 4.1.2. Analytical solution
face is located, which enables the use of any Lagrangian (Front- The initial bubble volume is known, V ol ¼ m
q0 , as is the injected
0

Tracking Unverdi and Tryggvason [27], Tryggvason et al. [26], water flow. The theoretical equation of the density during time is:
Markers And Cells (Raad and Bidoae [19], Vincent et al. [28]) or
Eulerian (VOF Lopez et al. [10], Level-Set Osher and Fedkiw [15], q0
qðtÞ ¼ ð37Þ
etc.) interface tracking method to account for interface motions. 1  pqtR2
Here, the Piecewise Linear Interface Method (VOF–PLIC) of Youngs
[37] is utilized as it preserves sharp interfaces on one grid cell with q the volumetric flow rate of injected water. Here,
length while ensuring mass conservation. We choose to implement q = 0.001 m3 s1.
this method for its robustness and easy programming. A Level-Set
or a Front-Tracking method could have been used in an equivalent 4.1.3. Numerical simulations
way by remplacing the VOF function C by an heavyside function The bubble volume is very quickly reduced until the collapse is
obtained as the projection of the markers on the Eulerian grid obtained and the mesh effect becomes more significant. The col-
[24] or as an analytical function depending on the Level-Set [25]. lapse takes place at t = 2.82 s, i.e. q(t) ? +1, in accordance with
The Navier–stokes equations are discretized by an implicit finite the theory. In order to test the grid convergence, the simulation
volume method on a staggered mesh. In 2D, a direct resolution is repeated with different meshes: 32 32, 64 64, 128 128,
solver MUMPS [1] is chosen to solve the linear systems. In 3D, an 256 256, 512 512. The time step is Dt = 0.00125 s and the iso-
iterative bi-conjugate gradient stabilized BICG-Stab II [35], precon- thermal compressibility coefficient is vair T ¼ 9:8692327 106 P1
9 1
ditioned under a Modified and Incomplete LU method [6] was for air and vwater
T ¼ 0:444 10 P for water. Numerical density
implemented. The spatial discretizations employed centered for each mesh is given in Table 1. The comparison between simu-
schemes. The surface tension force is modeled according to the lations on different meshes and the analytical solution is given in
Continuum Surface Force (CSF) of Brackbill et al. [2]; Popinet and Fig. 3 for the density and in Fig. 4 for the pressure. The numerical
Zaleski [17] as: FST = rjN di with r the surface tension coefficient, results converge towards the analytical solution when the number
j the local curvature of the interface and di the Dirac function indi- of grid cells is increased.
cating the interface. The numerical methods have been detailed The order p of the method, calculated by the Richardson method
 
and validated in previous articles [29–31]. q q
ln q4 q2
3 1
[20], is p ¼ lnðkÞ
¼ 2:07, where k ¼ 12. The asymptotic numerical
3.2. Advantages and limits density is q ¼ q4   q4 q3 p  ¼ 2:50055 kg m3 , i.e. a relative var-
N3
1 N4

The model and the related numerical methodology are able to iation of 0.24% with the theoretical density. In spite of a very low
account for flows characterized by density ratios of 107, viscosity speed, we obtain a good precision. The index 1, 2, 3 and 4 are re-
ratios of 1012, compressibility ratios of 10n (n = 5–10) and surface lated to grids 64 64, 128 128, 256 256 and 512 512.
tension between the two fluids higher than 10 N m1. These limi-
tations are due to computational errors. Furthermore, no particular 4.2. Test case 2: Liquid injection in a closed cell
state laws are required. The model manages high compression rate
even with very low speeds. However, the model is fully coupled on 4.2.1. Presentation of the case
velocities, which is numerically difficult to solve and requires effi- A gas volume initially placed in a closed cavity, is compressed
cient solvers. by a liquid water piston which is quasi-incompressible. The mass
of air is constant and its volume regularly varies over time as water
is injected at a constant speed. The cavity is a square of 0.1 m sides
4. Physical test cases
( Fig. 5). The Reynolds number is Re = 10000. The simulations are
carried out for several time steps and meshes.
The numerical simulation of compressible two-phase flows
involving separated phases is not often studied in the literature.
The majority of existing works is related to flows in which Mach
numbers are up to 1. The transition regime between 0 and 1 is very
useful for real problems such as plasma interaction with liquid jets,
liquid injection in closed molds. The main interest of this work is to
propose a specific model in non-conservative form that does not
require hyperbolic scheme equations as in a non-conservative for-
mulation, velocity and pressure remain continuous while Ma < 1.
No comparison with existing models of the literature is possible
as these low Mach number models are only formulated for single
phase problems. The following test cases that are presented in this
section have been designed to compare and validate the numerical
results, and therefore the new compressible one-fluid model, to
analytical references. Fig. 2. Compression of an air bubble by water.
30 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34

Table 1
Density at t = 1.5 s. The theoretical density is
2.50665 kg m3.

Mesh q(1.5 s)
32 32 2.29658
64 64 2.45194
128 128 2.46224
256 256 2.49142
512 512 2.49481
Fig. 5. Water injection in a closed cavity of air.

14 Table 2
Density at t = 0.3 s. The theoretical density is 1.96138 kg m3.
12 Mesh Dt q(0.3 s) Er
32*32
64*64 32 32 0.008 1.92831 1.6861
10 512*512
Density [kg/m3]

64 64 0.004 1.94444 0.8637

Analytical
128 128 0.002 1.95269 0.4431
256 256 0.001 1.95696 0.2254
8
512 512 0.0005 1.95916 0.1132

106 P1 for air and vwater

T ¼ 0:444 109 P1 for water. Time steps
4
are indicated in Table 2.

2
4.2.3. Numerical simulations
0
The pressure increases over time (Fig. 6) but it is uniform
0 0.5 1 1.5 2 2.5 throughout the cavity, in both water and air. Without gravity, the
Time [s] pressure is strictly uniform.
A correct behavior of the numerical solution is noticed. The liquid
Fig. 3. Evolution of the air density for different meshes. remains incompressible with a constant density of 1000 kg m3
(Fig. 7) while the density in air is linear.
Different meshes are tested. In Table 2, the density is given for
800000
these meshes as is the relative error with the analytical solution (in
32*32 %). A regular convergence towards the theoretical value is ob-
64*64 served. On the finest grid, the difference between the analytical
512*512
Analytical
and simulated density is around 0.1%. In this case, the free surface
600000
is moving and can be subjected to deformations. However, since a
coarse 64x64 grid, the error on density in air is less than 1% in this
Pressure [Pa]

case.
400000
4.3. Test case 3: Bubble subjected to an ultrasound field

4.3.1. Presentation of the case

200000 A transducer emits a signal of frequency f = 100 kHz with an
amplitude of e = 106 m on the lower wall of the cavity of 0.1 m
length, i.e. a sinusoidal celerity with an amplitude e and a fre-
quency f is imposed on the lower wall of the cavity. The case cor-
0 responds to the action of power ultrasounds on an air bubble of
0 0.5 1 1.5 2 2.5
radius R = 0.03 m in a water cavity. The mesh is 128 128 and
Time [s]
the time step is 107 s. The isothermal compressibility coefficient
Fig. 4. Evolution of the pressure in the air bubble. is vair
T ¼ 9:8692327 106 P1 for air and vwater
T ¼ 0:444 109
P1 for water.

4.2.2. Analytical solution

4.3.2. Numerical simulations
The constant air mass,which is equal to m0, is compressed by a
The propagation of an ultrasonic wave in water is first studied.
liquid piston, which moves at a constant speed V0 = 0.1 m s1. The
The ultrasonic wave propagation in a liquid creates phases of
density varies then as:
compression and depression (Figs. 8 and 9). The pressure varies
from  1 106 to 1 106 Pa and the velocity between 0.6 and
m0 q0 H 0 S q0 H0 q0 0.6 m s1. Numerically, the wave length is 0.015298 m on the
qðtÞ ¼ ¼ ¼ ¼ ð38Þ
X0  V 0 tS H0 S  V 0 tS H0  V 0 t 1  VH0 t 128 128 grid.
0

The speed of sound in water is equal to 1500 m s1. The wave

The density after 0.5 s must be equal to q = 3.5304 kg m3. The length is proportional to the sound celerity: k ¼ cf ¼ 0:015 m. Thus
pressure must be equal to p = qrT0 = 3.5304  287  300 = 303967 Pa. there is a variation of 1.99% between the numerical and analytical
The isothermal compressibility coefficient is vair T ¼ 9:8692327 wave length. The velocity varies between ±0.6 m s1 (Fig. 10).
 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34 31

32*32
128*128
512*512
80000 Analytical
Pressure [Pa]

70000

60000

50000

0.25 0.3 0.35

Time [s]

Fig. 6. Evolution of the pressure in air for different meshes.

Fig. 8. Pressure in the cavity.

10 3

500000

2
Density [kg/m3]

10
Pressure [Pa]

10 1

-500000

0 0.1 0.2 0.3 0.4 0.5

Time [s] 0 0.02 0.04 0.06 0.08 0.1
Y [m]
Fig. 7. Evolution of the density at the center of the cavity.
Fig. 9. Axial cut (x = 0.05 m) of the pressure field.

Now an air bubble is placed in the cavity. The bubble is repre- elliptic shape oscillates until a circular profile (equilibrium) is
sented by a black line. The amplitude and frequency characteristics reached.
remain equal to 106 m and 100 kHz. The new pressure variations
reach p ± 15 bar (Fig. 11). For such pressures, we are aware that 4.4.2. Analytical solution
cavitation effects take place. However, these effects [21] are not According to the analytical solution obtained by Lamb [9,23],
considered in this test case, but they were already studied with the interfacial position of a drop of radius R is given in polar coor-
the same numerical tool [22]. The present test case demonstrates dinates by:
that the compressible two phase model also makes it possible to
take into account acoustic wave interaction with fluid interfaces. rðh; tÞ t
¼1þ
~cosð2hÞcosðxd tÞexp
R sd
where  ~ is the initial perturbation of the bubble, h the angular coor-
4.4. Test case 4: Oscillations of an air bubble in steel
dinate and xd and sd are respectively the frequency and character-
istic time of decay due to viscous damping:
4.4.1. Presentation of the case
A deformed air bubble of elliptic shape is now considered. The 6r R2 ðqL þ qG Þ
bubble is inserted in a closed cavity which contains melted steel x2d ¼ ; sd ¼
3
R ðqL þ qG Þ 4 ðlL þ 3lG Þ
of density qL and viscosity lL. Under the effect of the surface ten-
sion forces, the bubble starts to oscillate and finally becomes cylin- The following two-phase characteristics are chosen:
drical [18]. r = 1.5 N m1 is the tension surface force, qG = 1.17683 kg m3 is
The computational domain is a square, with 0.05 m long sides. the air density, qL = 7000 kg m3 is the melted steel density, lL = 4
Fig. 12 shows the air-steel interface over time. It can be observed 103 Pa s is the dynamic viscosity of melted steel and lG = 1.85
that the interface is described with a regular shape. The initial 105 Pa s is the dynamic viscosity of air.
32 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34

0.012

Length of short semi-axe [m]

0.0115

0.011

0.0105

0.01

0.0095

0.009
0 0.2 0.4 0.6 0.8 1
Time [s]

Fig. 13. 
~ ¼ 0:2, mesh 128 128.

4.4.3. Numerical simulations

Fig. 10. Velocity in the cavity.
Numerical frequencies are compared to analytical values. In or-
der to compare simulation results with linear theory, small initial
perturbations must be applied. However, the semi-axe oscillations
are disturbed when ~  is too small (inferior to 0.1), i.e. when ~ is
smaller than the space step of the calculation grid. The displace-
ment of the interface at h ¼ p2 as a function of time and for an initial
perturbation of 0.2 is shown in Fig. 13. The mesh must be suffi-
ciently fine to describe accurately the initial interface perturbation
~.
The numerical frequency tends towards the analytical fre-
quency when ~  decreases. The convergence study was led using a
512 512 mesh and the results are presented in Table 3.
A mesh study is carried out for  ~ ¼ 0:2 to avoid generating a too
fine grid. The frequencies obtained are given in Table 4.
We notice that the numerical frequency tends towards the ana-
lytical frequency when the mesh is refined even if the difference
between the linear theory and simulations remains significant.
As previously stated, the initial deformation is too great, even for
 ¼ 0:05, to fulfill the linear theory assumption. Very small ~, less
~
than 0.01, are required to satisfy the linear theory. But they were
not simulated in order to avoid using too great a number of cells
and too long computing times. However, Tables 3 and 4 demon-
strate that the simulations converge to the linear theory as ~  and
Dx tends to zero.
Fig. 11. Bubble subjected to an ultrasound field.
The characteristic time of decay due to viscous damping sd is
calculated for different initial perturbations  ~: 0.4, 0.2, 0.1 and
0.05, with a 128 128 mesh. For ~  ¼ 0:4, we obtain sd = 9.5877 s
t=0.1125s
and for  ~ ¼ 0:05; sd ¼ 39:1055 s. The analytical characteristic time
is sd = 43.1585 s. As for the frequency, we notice a convergence of
simulated characteristic time of decay sd to the analytical solution
when  ~ tends to 0. This parameter is much more sensitive to the
initial interface perturbation than the bubble frequency.

t=0s
5. Conclusions and perspectives

A new model for the numerical simulation of compressible mul-

tiphase flows has been developed from a non-conservative formu-
t=0.0625s lation of the 1-fluid Navier–Stokes equations. It allows the
simultaneous management of the compressible and incompress-
ible features of the multiphase flow. The mass conservation equa-
t=0.075s tion is used to build an evolution equation on the pressure. The
pressure gradient in the momentum equation is formulated with
Fig. 12. First oscillations of an air bubble in steel with a 256  ¼ 0:4
256 mesh, ~ the pressure equation to account for compressible effects. Three
and Dt = 2.5 104 s. new terms are now in the momentum equation. The modeling of
 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34 33

Table 3 model will be used to simulate for example the injection of gas
Frequency of bubble oscillations for different
in melted metals [16] or the interaction between a plasma and a
initial perturbations for a 512 512 mesh. The
theoretical frequency is 5.70632 s1. liquid jet [29].


~ Frequency (s1)

0.4 4.66479 Acknowledgements

0.2 5.03814
0.1 5.21205 The authors thank the Aquitaine Regional Council for the finan-
0.05 5.50479
cial support dedicated to a 256-processor cluster investment, lo-
cated in the TREFLE laboratory. We are grateful for access to the
computational facilities of the French CINES (National computing
Table 4 center for higher education), IDRIS (National computing center of
Frequency of bubble oscillations with 
~ ¼ 0:2. The theoretical frequency is CNRS) and CCRT (National computing center of CEA) under Project
5.70632 s1.
Number x2009026115.
Mesh Dt Frequency (s1)
32 32 0.002 4.36758
References
64 64 0.001 4.79187
128 128 0.0005 4.95222
[1] Amestoy P, Duff I, L’Excellent JY, Li X. Analysis, tuning and comparison of two
256 256 0.00025 4.98973
general sparse solvers for distributed memory computers. ACM Trans Math
512 512 0.000125 5.03814
Softw 2001;27(4):388–421.
[2] Brackbill J, Kothe D, Zemach C. A continuum method for modeling surface
tension. J Comput Phys 1992;100:335–54.
the compressibility in the energy equation is also presented. Spe- [3] Caiden R, Fedkiw RP, Anderson C. A numerical method for two phase flow
consisting of separate compressible and incompressible regions. J Comput
cific divergence terms appear when the pressure equation account- Phys 2001;166:1–27.
ing for mass conservation is satisfied. The model is first presented [4] Caltagirone J, Vincent S. Sur une méthode de pénalisation tensorielle pour la
in its general form without any simplification. Then, thanks to a résolution des équations de Navier–Stokes. CR Acad Sci Paris 2001;IIb:607–13.
[5] Fortin M, Glowinski R. Méthodes de Lagrangien augmenté. Application à la
non-dimensional analysis of the new mass conservation equation, résolution numérique de problème aux limites. Dunod; 1982.
a simplification of the model for a Mach number which tends to- [6] Gustafsson I. On first and second order symmetric factorization methods for
ward zero and for isothermal flows, is given. As a first step, this the solution of elliptic difference equations. Chalmers University of
Technology; 1978.
simplified version of the model is approximated and validated.
[7] Hirsch C. Numerical computation of internal and external flows. Computational
Contrary to usual explicit discretization schemes used for solving methods for inviscid and viscous flows, vol. 2. Chichester: John Wiley and Sons;
compressible flows, we proposed implementing a fully implicit 1990.
discretization to solve the velocity component in a coupled way, [8] Kataoka I. Local instant formulation of two-phase flow. Int J Multiphase Flow
1986;12:745–58.
without any time splitting assumption [32] and without any [9] Lamb H. Hydrodynamics. 6th ed. Cambridge University Press; 1932.
hyperbolic scheme requirement. [10] Lopez J, Hernandez J, Gomez P, Faura F. An improved PLIC–VOF method for
The numerical methods were based on finite volumes on stag- tracking thin fluid structures in incompressible two-phase flows. J Comput
Phys 2005;208:51–74.
gered Cartesian grids, centered schemes and direct numerical solv- [11] Majda A. Compressible fluid flow and systems of conservation laws in several
ers. The proposed compressible 1-fluid model is independent on space variable. Springer-Verlag; 1984.
the interface tracking approach and may be established by any [12] Meillot E, Vincent S, Caruyer C, Caltagirone J, Damiani D. From dc time-
dependent thermal plasma generation to suspension plasma-spraying
Lagrangian (Front-Tracking, MAC) or Eulerian method (VOF, interactions. J Therm Spray Technol 2009;18:875–86.
Level-Set, etc.). A VOF–PLIC method is used in this work for its abil- [13] Nourgaliev R, Dinh T, Theofanous T. A pseudocompressibility method for the
ity to handle mass conservation without involving numerical diffu- numerical simulation of incompressible multifluid flows. Int J Multiphase Flow
2004;30:901–37.
sion during interface tracking. [14] Nourgaliev R, Dinh T, Theofanous T. Adaptive characteristics-based matching
Several test cases have been performed on different isothermal for compressible multifluid dynamics. J Comput Phys 2006;213:500–29.
physical problems. The capability of the SCOF model to handle [15] Osher S, Fedkiw R. Level set methods: an overview and some recent results. J
Comput Phys 2001;169:463–502.
compressible and incompressible conditions as well as two-phase
[16] Pianet G, Vincent S, Leboi J, Caltagirone J, Anderhuber M. Simulating
flows has been demonstrated. First the compression of an air bub- compressible gas bubbles with a smooth volume tracking 1-fluid method.
ble by water is studied. The calculated convergence order is 2 in Int J Multiphase Flow 2010;36:273–83.
space with the Richardson method. Then liquid injection in a [17] Popinet S, Zaleski S. A front tracking algorithm for accurate representation of
surface tension. Int J Numer Meth Fluids 1999;30:775–93.
closed cavity initially full of air is considered. Close agreement [18] Prosperetti A. Free oscillations of drops and bubbles: the initial problem. J
was obtained with the theory by comparing analytical and simu- Fluid Mech 1980;100:333–47.
lated pressures. The case of the bubble subjected to an ultrasound [19] Raad P, Bidoae R. The three-dimensional Eulerian–Lagrangian marker and
micro cell method for the simulation of free surface flows. J Comput Phys
field demonstrates that the compressible two phase model also 2005;203:668–99.
makes it possible to take into account acoustic waves if adapted [20] Roache P, Verification and validation in computational science and
time steps are used. Finally, the case of a bubble oscillating in engineering. Hermosa; 1998.
[21] Servant G. Contribution à l’étude et à la modélisation de la propagation des
melted steel is approached. A correct behavior of the solution is ob- ultrasons de puissance en milieu cavitant. PhD thesis. Université de Bordeaux
served by comparison to the linear theory. The numerical frequen- 1; 2001.
cies and characteristic times converge towards analytical values. [22] Servant G, Caltagirone J, Gerard A, Laborde J, Hita A. Numerical simulation of
cavitation bubble dynamics induced by ultrasound waves in a high frequency
Future works will be devoted to the validation of the presented reactor. Ultrason Sonochem 2000;7:217–27.
model in the case of heat transfer and larger Mach number com- [23] Shepel S, Smith B. On surface tension modelling using the level set method. Int
pressible flows. Thermal effects must be added to the mass conser- J Numer Meth Fluids 2008;59:147–71.
[24] Shin S, Juric D. Modeling three-dimensional multiphase flow using a level
vation equation written in terms of a pressure equation. A
contour reconstruction method for Front Tracking without connectivity. J
numerical study will be also provided to evaluate the effect of Comput Phys 2002;180:427–70.
the interface tracking method on the physical meaning of the sim- [25] Sussman M, Smereka P, Osher S. A Level Set approach for computing solutions
ulation. Front-Tracking, Level-Set and VOF methods will be com- to incompressible two-phase flow. J Comput Phys 1994;114:146–59.
[26] Tryggvason G, Brunnen B, Esmaeli A, Juric D, Al-rawahi N, Han W, et al. A
pared. No major effects of the interface tracking method is Front-Tracking next term method for the computations of multiphase flow. J
forseen. The isothermal and non-isothermal compressible 1-fluid Comput Phys 2001;169:708–59.
34 J.-P. Caltagirone et al. / Computers & Fluids 50 (2011) 24–34

[27] Unverdi S, Tryggvason G. A front-tracking method for viscous, incompressible, [32] Viozat C. Implicit upwind schemes for number low Mach compressible flows.
multi-fluid flows. J Comput Phys 1992;100:25–37. Rapport de recherche n3084 INRIA; 1997.
[28] Vincent S, Balmigere G, Caltagirone J, Meillot E. Eulerian–Lagrangian [33] Viozat C. Calcul d’écoulements stationnaires et instationnaires à petit nombre
multiscale methods for solving scalar equations application to de Mach et en maillages étirés. PhD thesis. Université de Nice-Sophia
incompressible two-phase flows. J Comput Phys 2010;229:73–106. Antipolis; 1998.
[29] Vincent S, Balmigere G, Caruyer C, Meillot E, Caltagirone J. Contribution to the [34] Voller V, Swaminathan C. Fixed grids techniques for phase change problems: a
modeling of the interaction between a plasma flow and a liquid jet. Surf Coat review. Int J Numer Meth Fluids 1990;30:875–98.
Technol 2009;203:2162–71. [35] Vorst HVD. A fast and smoothly converging variant of Bi-CG for the solution of
[30] Vincent S, Caltagirone J. Efficient solving method for unsteady incompressible non-symmetric linear systems. J Sci Stat Comput 1992;44:631–44.
flow problems. Int J Numer Meth Fluids 1999;30:795–811. [36] Yabe T, Yuan P. Unified numerical procedure for compressible and
[31] Vincent S, Caltagirone J, Lubin P, Randrianarivelo T. An adaptative augmented incompressible flow. J Phys Soc Jpn 1991;60:2105–8.
Lagrangian method for three-dimensional multimaterial flows. Comput Fluids [37] Youngs D. Time-dependent multi-material flow with large fluid distorsion.
2004;33:1273–89. K.W. Morton and M.J. Baines; 1982.