ESAIM: M2AN 46 (2012) 1029–1054 ESAIM: Mathematical Modelling and Numerical Analysis

DOI: 10.1051/m2an/2011069 [Link]

MODELLING AND SIMULATION OF LIQUID-VAPOR PHASE TRANSITION

IN COMPRESSIBLE FLOWS BASED ON THERMODYNAMICAL
EQUILIBRIUM ∗

Gloria Faccanoni1 , Samuel Kokh2 and Grégoire Allaire3,4

Abstract. In the present work we investigate the numerical simulation of liquid-vapor phase change
in compressible flows. Each phase is modeled as a compressible fluid equipped with its own equation
of state (EOS). We suppose that inter-phase equilibrium processes in the medium operate at a short
time-scale compared to the other physical phenomena such as convection or thermal diffusion. This
assumption provides an implicit definition of an equilibrium EOS for the two-phase medium. Within
this framework, mass transfer is the result of local and instantaneous equilibria between both phases.
The overall model is strictly hyperbolic. We examine properties of the equilibrium EOS and we propose
a discretization strategy based on a finite-volume relaxation method. This method allows to cope with
the implicit definition of the equilibrium EOS, even when the model involves complex EOS’s for the
pure phases. We present two-dimensional numerical simulations that shows that the model is able to
reproduce mechanism such as phase disappearance and nucleation.

Mathematics Subject Classification. 76T10, 76N10, 65M08.

Received January 24, 2011. Revised July 13, 2011.
Published online February 13, 2012.

1. Introduction
The simulation of liquid-vapor phase change phenomena in fluid flows raises challenging problems pertaining
to physical modeling, mathematical analysis and numerical analysis. It is all the more important that it concerns
many industrial applications. For example, the prediction of the boiling crisis is a crucial problem for the safety
studies in the nuclear industry [14–16, 18, 46, 48, 54]. In the present article, we focus on a model of phase change
for compressible two-phase flows (away from the critical point) and propose a numerical algorithm for solving it.
Keywords and phrases. Compressible flows, two-phase flows, hyperbolic systems, phase change, relaxation method.
∗ This work has been achieved within the framework of the NEPTUNE project, financially supported by CEA (Commissariat à
l’Énergie Atomique), EDF (Électricité de France), IRSN (Institut de Radioprotection et de Sûreté Nucléaire) and AREVA-NP.
The authors thank the CEA for its financial support. Grégoire Allaire is a member of the DEFI project at INRIA Saclay
Ile-de-France and is partially supported by the Chair “Mathematical modeling and numerical simulation”, F-EADS – École
Polytechnique – INRIA.
1 IMATH – Université du Sud Toulon-Var, Avenue de l’Université, 83957 La Garde, France. faccanon@[Link]
2 DEN/DANS/DM2S/SFME/LETR, Commissariat à l’Énergie Atomique Saclay, 91191 Gif-sur-Yvette, France.
[Link]@[Link]
3 Conseiller Scientifique du DM2S – Commissariat à l’Énergie Atomique Saclay, 91191 Gif-sur-Yvette, France.

4 CMAP, École Polytechnique, CNRS, 91128 Palaiseau, France. allaire@[Link]

Article published by EDP Sciences 

c EDP Sciences, SMAI 2012
1030 G. FACCANONI ET AL.

A classic approach consists in modeling the two-phase system by a single fluid with a Van der Waals type
equation of state (EOS) [22]. Far from the critical point, each phase corresponds to a stable branch of the EOS,
which are connected by the spinodal zone. The intrinsic instability of the fluid in the spinodal zone requires
to account for very small scale effects by the means of Korteweg-type tensors (see [6, 24, 27, 33, 37, 52] and the
references therein) including dissipative effects as well as dispersive effects. In other words, the interface between
the two phases is implicitly modeled as a thick transition zone. A drawback of such models is that they call for
very fine grids when discretizing the system.
It is possible to model the interface as a discontinuity locus connecting two pure phase states. Such dis-
continuity can be obtained by taking the limit in the Korteweg model when the interface thickness tends to
zero [10,17,38,51,53]. However this requires the knowledge of the whole EOS for the two-phase material including
the spinodal zone. Unfortunately, experimental data for the spinodal zone are often not available.
We follow here an alternate approach that has been examined by several authors for the past
years [2, 4, 10, 11, 18, 19, 21, 23, 25, 29, 31, 34, 45, 48]. The key idea lies in avoiding the spinodal zone by assum-
ing that the driving physical phenomena are kinetical effects that act on both phases at a short time scale
compared to the hydrodynamics and thermal diffusion in agreement with classic thermodynamics. This comes
down to assuming that each phase is modeled by a compressible fluid and that the EOS of the two-phase system
is obtained by a convexification procedure (see Prop. 4.2). We note that this construction is more general than
the classic convexification procedure based on the Maxwell rule of a Van der Waals EOS [9, 28, 34, 45]: in the
present model each pure phase is equipped independently with its own EOS. If we consider the particular case
when each pure phase EOS matches a stable branch of the Van der Waals EOS, the resulting equilibrium EOS
of our model coincides with the classic Maxwell construction for the Van der Waals EOS [30]. For all other cases
we propose a simple way to compute the equilibrium EOS, based on solving a single nonlinear equation that we
call “phase-change equation”. We demonstrate its usefulness in the case of pure phases having a stiffened gas
EOS.
Within this context, the mass transfer between both phases is driven by instantaneous equilibria in the
composition of the two-phase medium with respect to phasic pressures, temperatures and Gibbs potentials.
An important issue is that, as it is expected, the resulting EOS does no longer fulfill classic strict convexity
thermodynamics hypotheses for the entropy. Moreover, the process of reaching equilibria (and equivalently
the convexification process) involves solving a nonlinear system that combines the EOS of the pure phases.
Nevertheless, it is possible to show that, under simple hypotheses on the pure phase EOSs, the system is strictly
hyperbolic (Thm. 5.2). This property, announced in [2], is important for the theoretical and numerical resolution
of the system.
Once this well-posedness property is obtained, we can use the relaxation framework of [7, 12, 13, 35, 43] to
interpret the equilibria between both phases (and equivalently the convexification process) as the limit of a
relaxation procedure. This modeling interpretation provides means of discretizing the system thanks to a two-
step relaxation strategy.
This paper is structured as follows: in Section 2 we present the overall structure of our model. In Section 3 we
first recall basic thermodynamics principles for pure phases, then we provide our system with an off-equilibrium
entropy. Then we introduce the equilibrium entropy of the system. In Section 4 we detail properties of this
entropy law related to the geometrical construction of its graph, convexification principles of the pure phases
entropy and convexity properties. The definition of this equilibrium law involves the resolution of a system of
nonlinear equations. We show that it is possible to reduce this system to a nonlinear scalar equation by using
classic thermodynamical state laws for saturated states. We recall in Section 5 key well-posedness property of
this system. In Section 6 we show that the system can be interpreted by means of a relaxation approach to
other two-phase system. We propose in Section 7 a multi-step relaxation discretization strategy that allows to
decouple computation of the equilibrium thermodynamical parameters from the approximation of the convection
by confining it in the projection step. Finally in Section 8 we present two-dimensional numerical simulations of
boiling-type tests.
 LIQUID-VAPOR PHASE TRANSITION 1031

2. Model structure
In the sequel the subscript α = l (resp. α = v) will refer to the liquid (resp. vapor) phase. Let τα and εα
denote respectively the specific volume and the specific internal energy of the phase α = l, v. We define the cone
of admissible states
C ={(τ, ε) | τ > 0, ε > 0}
def
(2.1)
and wα =(τα , εα ). In the following we shall always assume that wα ∈ C.
def

We suppose that the fluid will never reach the critical point and therefore each phase can be modelled as
a compressible fluid equipped with its complete EOS wα → sα , where sα denotes the specific entropy of the
phase α = l, v. The assumption on the pure phase entropy laws will be detailed in Section 3.1.
In order to define the composition of the two-phase media, we introduce three frac-
tions [2, 4, 10, 11, 18, 19, 21, 29, 31]:
• yl (resp. yv ) the mass fraction of the phase α = l (resp. α = v); we note y = yl and we suppose y ∈ [0, 1]
def

(that implies yv = 1 − y);

• zl (resp. zv ) the volume fraction of the phase α = l (resp. α = v); we note z = zl and we suppose z ∈ [0, 1]
def

(that implies zv = 1 − z);

• ψl (resp. ψv ) the energy fraction of the phase α = l (resp. α = v); we note ψ = ψl and we suppose ψ ∈ [0, 1]
def

(that implies ψv = 1 − ψ).

Let us now define the global specific volume τ and internal energy ε of the two-phase medium: for w =(τ, ε)
def

∈ C, the additivity of the volume and the internal energy implies that
     
τ τ
yα α = y α wα .
def
w= =
ε εα
α=l,v α=l,v

We define the set Q(w) as follows

  
 1 = yl + yv

Q(w) = (wl , wv , yl , yv ) ∈ C × [0, 1] .
def 2 2
 (2.2)
 w = y l wl + y v wv

Therefore
τl εl
z=y and ψ=y ·
τ ε
We suppose that both phases have the same velocity u and we set  = 1/τ . If we neglect the viscous dissipative
def

effects, then the two-phase medium is governed by the system

∂t U + div[Feq (U)] = Stension (U, z) + Sgravity (U) − div[Qheat (U, z)], (2.3)

where
⎛ ⎞ ⎛ ⎞
 u
U=⎝ u ⎠, Feq (U) = ⎝ u ⊗ u + P eq Id ⎠.
def def

2 2
ε +  |u| /2 (ε +  |u| /2 + P eq )u

The terms Sgravity (U) and Qheat (U, z) that account respectively for gravity and thermal diffusion effects read:
⎛ ⎞ ⎛ ⎞
0 0
Sgravity (U) = ⎝ g ⎠ , Qheat (U, z) = ⎝ ⎠
def def
0
g · u ϑ grad T
1032 G. FACCANONI ET AL.

where ϑ = zϑl +(1−z)ϑv is the thermal conductivity of the two-phase medium and ϑα is the thermal conductivity
def

of the phase α. The surface tension effects are modeled thanks to the continuum surface force (CSF) model
of [8] which reads here
⎛ ⎞
0  
⎝ ⎠ grad z grad z
Stension (U, z) = T , T=ξ
def def
div
T·u |grad z| |grad z|

where ξ is the capillary coefficient.

The pressure law (τ, ε) → P eq (τ, ε) derives from an equilibrium EOS whose construction will be presented in
Section 4. This EOS accounts for the mass transfer between both phases by imposing local and instantaneous
thermodynamics equilibria in the composition of the medium.

3. Equilibrium EOS: definition

First we shall recall basic notations and assumptions pertaining to homogeneous pure fluids state laws. Then
we will postulate an off-equilibrium state law for a medium composed of a liquid phase and a vapor phase.
Finally, we define equilibrium within this medium that will provide us an equilibrium EOS accounting for mass
transfer between both phases.

3.1. Pure phase EOS

In the following we suppose that sα is a C2 regular function over the cone C (2.1). We note the first and
second derivatives of sα as follows
 
α
def ∂s 
(sα )εα = ∂s ∂εα τ , = ∂ταα  ,
def
(sα )τα
α εα
 
(sα )εα εα = ∂∂εs2α  , (sα )τα τα = ∂∂τs2α  , (sα )τα εα = ∂τ∂α s∂ε
def 2 def 2 def 2
α
α
α τα α εα

and we note the Hessian matrix of sα


(sα )τα τα (sα )τα εα
d sα = .
2 def

(sα )τα εα (sα )εα εα

The temperature Tα , the pressure Pα , the free enthalpy gα and the square of the speed of sound cα are defined
classically by
1 (sα )τα
Tα = , Pα = , gα = εα + Pα τα − Tα sα ,
def def def
(3.1)
(sα )εα (sα )εα
   
∂P  ∂P 
c2α = τα2 Pα
def α  − α  (3.2a)

∂εα τα ∂τα εα

  P
= −τα2 Tα Pα , −1 d2 sα α . (3.2b)
−1
Thermodynamics characterizes the derivatives of sα : first, temperature positivity requires sα to be a strictly
increasing function of εα ; second, a stability assumption is enforced by assuming a definite negative Hessian
matrix d2 sα for sα . Therefore, for all wα ∈ C, we have
⎧
⎪ (s ) > 0, (3.3a)
⎨ α εα
(sα )εα εα (sα )τα τα > ((sα )τα εα ) ,
2
(3.3b)
⎪
⎩
(sα )εα εα < 0, (or equivalently (sα )τα τα < 0). (3.3c)
Let us note that relations (3.3b)–(3.3c) imply that wα → sα is strictly concave but the converse is wrong.
 LIQUID-VAPOR PHASE TRANSITION 1033

3.2. Off-equilibrium two-phase EOS

Following classic thermodynamics (see [9, 28]), we define an entropy σ for the off-equilibrium medium: for all
(wl , wv , yl , yv ) ∈ C 2 × [0, 1]2 ,
σ(wl , wv , yl , yv ) = yl sl (wl ) + yv sv (wv ).
def
(3.4)
The entropy σ can also be expressed thanks to the variables (y, z, ψ, w = (τ, ε)) ∈ [0, 1]3 × C as follows (for the
sake of simplicity we keep the same notation)
⎧
⎨sl (τ,ε)
⎪    if y = 1,
σ(y, z, ψ, τ, ε) = ysl yz τ, ψy ε + (1 − y)sv 1−y
1−z
τ, 1−ψ ε if 0 < y < 1, (3.5)
⎪
⎩
1−y
sv (τ, ε) if y = 0.

3.3. Equilibrium two-phase EOS

Following the lines of [9, 28, 29, 31], for a given global state w =(τ, ε) ∈ C, we can define the equilibrium
def

composition parameters (wl∗ , wv∗ , yl∗ , yv∗ ) ∈ Q(w), in agreement with the second law of Thermodynamics, as
τ∗ ε∗
maximizer of (wl , wv , yl , yv ) → σ and we note z ∗ = y ∗ τl , ψ ∗ = y ∗ εl .
def def

We now assume that the fluid always instantaneously reaches equilibrium. We can introduce an equilibrium
entropy w → seq for the two-phase medium by the following definition: for all w ∈ C,
  

seq (w) = sup σ(wl , wv , yl , yv )  (wl , wv , yl , yv ) ∈ Q(w) (3.6)

or equivalently  
 y∈[0,1]

seq (w) = sup σ(y, z, ψ, w)  z∈[0,1] . (3.7)
ψ∈[0,1]

For the equilibrium two-phase medium, the pressure P eq , the temperature T eq, the chemical potential g eq
and the speed of sound ceq are then defined thanks to the classic formulas

seq
τ 1
P eq = , T eq = , g eq = ε + P eq τ − T eq seq ,
def def def

seq
ε seq
ε

 
eq 
  
eq ∂P  ∂P eq   eq  2 eq P eq
(c ) = τ P
eq 2 2
− = −τ T
2 eq
P , −1 d s . (3.8)
∂ε τ ∂τ ε −1

It was shown in [31] that the optimization problem (3.6) or (3.7) is equivalent to perform an inf-convolution
between two convex functions. This result guarantees that w → seq is always concave and therefore that the
equilibrium speed of sound ceq satisfies (ceq )2 ≥ 0. Let us emphasize that this condition is not sufficient for
ensuring the hyperbolicity of the Euler system (2.3) (see for example the equilibrium p-system studied in [10,11]).
Nevertheless, we will present in Section 4.2 additional arguments that prove that (ceq )2 > 0 and therefore that
the Euler system (2.3) is strictly hyperbolic (excluding of course the cases in which vacuum is present).
We now introduce another assumption that pertains to the thermodynamic properties of the liquid and the
vapor at equilibrium (see e.g. [3, 9, 28, 44]), namely

0 < y∗ < 1 ⇒ {τl∗ < τv∗ , ε∗l < ε∗v , sl (wl∗ ) < sv (wv∗ )} . (3.9)

In the sequel, we shall assume that the EOS of the liquid and the vapor phases verify hypothesis (3.9). The
strict inequality in assumption (3.9) means that we consider only first order phase transition. The assumed
ordering of the liquid and vapor properties is satisfied for most liquid-vapor systems.
1034 G. FACCANONI ET AL.

Remark 3.1. Hypothesis (3.9) implies that the system cannot reach the critical point. Indeed at the critical
point one necessarily has τl∗ = τv∗ , ε∗l = ε∗v and sl (wl∗ ) = sv (wv∗ ).

Assumptions (3.3) and hypothesis (3.9) provide a simple characterization of the solution of the optimization
problem (3.6).

Theorem 3.2 (extremum principle). Let (wl , wv , yl , yv ) → σ be the off-equilibrium entropy of definition (3.4).
For a given state w ∈ C,
(1) there is at least one state (wl∗ , wv∗ , yl∗ , yv∗ ) ∈ Q(w) solution of the optimization problem (3.6);
(2) a state (wl∗ , wv∗ , yl∗ , yv∗ ) ∈ Q(w) is a solution of the optimization problem (3.6) if and only if one of the
following systems is satisfied:
 ∗
yl = 1, yv∗ = 0,
either (3.10)
sl (w) > sv (w);
 ∗
yl = 0, yv∗ = 1,
either (3.11)
sl (w) < sv (w);

0 < yl∗ , yv∗ <
  1,  
either 1 Pl gl ∗ 1 Pv gv ∗ (3.12)
Tl Tl Tl (wl ) = Tv , Tv , Tv (wv ).
, ,

The state (wl∗ , wv∗ , yl∗ , yv∗ ) will then be referred to as an equilibrium state.
When 0 < yl∗ < 1 (then 0 < yv∗ < 1), we will say that the equilibrium state w is a saturated state.
(3) if w ∈ C is a saturated state then there is a unique solution of the optimization problem (3.6).
(4) if w ∈ C is not a saturated state, i.e. yv∗ = 1 (resp. yl∗ = 1) then wv∗ (resp. wl∗ ) defines an unique solution
for (3.4) up to the arbitrary choice of wl∗ (resp. wv∗ ) for the vanished phase.

Proof. See [4, 18, 29]. 

4. Equilibrium EOS: construction

In the present section we show that the optimization procedure involved in (3.6) reads as a direct geometrical
construction of the concave hull of the function w → max{sl (w), sv (w)}. Moreover, this interpretation will
provide us a new equivalent definition for the saturated states that boils down to a single nonlinear scalar
equation. This approach has been exploited in [10, 11, 18, 19, 21, 34] within a numerical simulation framework.

4.1. Properties of the equilibrium EOS

Following [9–11, 18, 34] we recall a first geometrical result that characterizes the solution of (3.6).

Proposition 4.1 (bitangent plane). For α = l, v let Sα be the surface defined by the graph of w → sα in the
(w, s) space. Given a thermodynamical state w ∈ C, if (wl∗ , wv∗ , yl∗ , yv∗ ) ∈ Q(w) maximizes (wl , wv , yl , yv ) → σ
and if 0 < yl∗ < 1 (equivalently 0 < yv∗ < 1), then there exists a unique plane, called “bitangent plane”, tangent
to the surface Sl at the point (wl∗ , s∗l = sl (wl∗ )) and to the surface Sv at the point (wv∗ , s∗v = sv (wv∗ )).
def def

∗
Proof. For α = l, v, let Pα be the tangent plane to the surface Sα at the point (wα , s∗α ). We have to prove that
the plane Pl and the plane Pv are the same. This boils down  to show  Pl and Pv are parallels and that
 that
Pl ∩ Pv = ∅. Because of (3.1), the equation of Pα reads s = PTαα (wα ∗
) τ + T1α (wα ∗
) ε + Tgαα (wα∗
). Thus Pl and
∗ ∗ ∗ ∗ gl ∗ gv ∗
Pv are parallels if and only if Pl (wl ) = Pv (wv ) and Tl (wl ) = Tv (wv ). Moreover, Tl (wl ) = gv (wv ) is equivalent
to w = (0, 0) ∈ Pl ∩ Pv = ∅. 
 LIQUID-VAPOR PHASE TRANSITION 1035

Consequently solving (3.6) leads either to yl∗ ∈ {0, 1} either to determine the bitangent plane to Sl and Sv .
We now recall a second classic result (see [9, 32]) that connects the optimization problem (3.6) and the
bitangent plane construction of Proposition 4.1. An alternative equivalent way, based on the inf-convolution,
for constructing the equilibrium entropy w → seq is presented in [30].

Proposition 4.2 (concave hull). Let S be the graph of the equilibrium entropy w → seq in the (w, s) space.
Then
(1) the surface S is the concave hull of the set

{(w, s) ∈ C × R | s ≤ max [sl (w), sv (w)]};

(2) for every saturated state w ∈ C, the surface S contains a segment r(w) passing through the point (w, seq (w)).
Along this segment the pressure P eq , the temperature T eq and the chemical potential g eq are constant.

In general w → seq is C1 but not C2 [5]. In the following, we assume that it is piecewise C2 in the sense that
the set of all saturated states (defined by (3.12)) is a C2 manifold with a boundary which is a C1 closed loop
curve.

Theorem 4.3 (existence and uniqueness of the segment r(w)). For α = l, v, let Sα be the graph of the phasic
entropy wα → sα in the (w, s) space and let w ∈ C be a saturated state (defined in Thm. 3.2). Then there exists
a unique couple of points

M∗l =(wl∗ , s∗l = sl (wl∗ )) ∈ Sl ,

def def

M∗v =(wv∗ , s∗v = sv (wv∗ )) ∈ Sv ,

def def

such that the point M =(w, seq (w)) belongs to the line segment
def

r(w) =(M∗l , M∗v ) = {yM∗l + (1 − y)M∗v | y ∈ [0, 1]}.

def

Proof. The existence of the segment r(w) follows from Proposition 4.2. We prove uniqueness: each sα is strictly
concave and increasing with respect to the variables τα and εα . Then the mapping (wl , wv ) → (P eq , T eq ) is
one-to-one. If there is another segment r̃(w) = ((w̃l∗ , s̃∗l ), (w̃v∗ , s̃∗v )) such that (w, seq (w)) ∈ r(w) ∩ r̃(w), since
∗ ∗
(P eq , T eq ) are constant along r(w) and along r̃(w), we have wα = w̃α and consequently r(w) = r̃(w). 

4.2. Phase-change equation

We now build a so-called “phase-change equation” which is our key ingredient in the computation of the
equilibrium entropy.
Consider a fixed couple w =(τ, ε). Let (wl∗ , wv∗ , yl∗ , yv∗ ) ∈ Q(w) be the maximizer of σ that corresponds to a
def

saturated state, i.e. 0 < y < 1. In this case, by Theorem 3.2 (τl∗ , ε∗l , τv∗ , ε∗v , y ∗ ) is also the unique solution of
∗

the system (4.1)–(4.2):

⎧
⎪ τ = yτl + (1 − y)τv , (4.1a)
⎪
⎪
⎪
⎨ ε = yεl + (1 − y)εv ,
⎪ (4.1b)
Pl (τl , εl ) = Pv (τv , εv ), (4.1c)
⎪
⎪
⎪
⎪ T (τ , ε ) = Tv (τv , εv ), (4.1d)
⎪
⎩ l l l
gl (τl , εl ) = gv (τv , εv ), (4.1e)

y ∈ (0, 1). (4.2)

1036 G. FACCANONI ET AL.

Introducing the common values (P, T ) of the pressure and temperature, relations (4.1c)–(4.1d) allow to obtain
the following equivalent system
⎧
⎨ τ = yτl (P, T ) + (1 − y)τv (P, T ),
⎪ (4.3a)
ε = yεl (P, T ) + (1 − y)εv (P, T ), (4.3b)
⎪
⎩
gl (P, T ) = gv (P, T ), (4.3c)

y ∈ (0, 1) (4.4)

where the unknows are (T, P, y). If we know P and T , we can compute y by

τ − τv (P, T )
y(P, T ) = (4.5)
τl (P, T ) − τv (P, T )

and eliminate y in system (4.3). So, solving system (4.3) is equivalent to solve
⎧
⎨ τ − τv (P, T ) ε − εv (P, T )
= , (4.6a)
τl (P, T ) − τv (P, T ) εl (P, T ) − εv (P, T )
⎩
gl (P, T ) = gv (P, T ). (4.6b)

The equation (4.6b) defines the well-known coexistence curve P → T sat (P ) [9,28]. In full generality, the solution
of (4.6b) is multivalued. Here we consider the physically relevant branch. If we note classically

ταsat (P ) = τα (P, T sat (P )), εsat

α (P ) = εα (P, T α = l, v,
def def sat
(P )),

then, given (τ, ε) ∈ C, solving (4.6a)–(4.6b) is equivalent to seeking P as the solution of the following scalar
nonlinear equation
τ − τvsat (P ) ε − εsat
v (P )
= sat · (4.7)
τl (P ) − τv (P )
sat sat εl (P ) − εsat
v (P )

In the sequel, the equation (4.7) will be also referred to as the “phase-change equation” (a similar equation was
proposed in [49]).
If we note P ∗ the solution of the phase-change equation (4.7), then the solution (τl∗ , ε∗l , τv∗ , ε∗v , y ∗ ) of the
system (4.1) reads

τl∗ = τlsat (P ∗ ),
ε∗l = εsat ∗
l (P ),
τv∗ = τvsat (P ∗ ),
ε∗v = εsat ∗
v (P ),
y ∗ = y(P ∗ , T sat (P ∗ )).

Remark 4.4. In some cases it can be most useful to compute T → P sat (T ) instead of P → T sat (P ). The
strategy proposed above is the same except that we replace P by T , P ∗ by T ∗ and T sat (P ) by P sat (T ).

We can now summarize the construction of the equilibrium entropy seq in the following definition.

Definition 4.5 (equilibrium entropy construction). The overall procedure for computing the equilibrium EOS
lies in the following alternative: for all w ∈ C,
 LIQUID-VAPOR PHASE TRANSITION 1037

(1) if there exists a solution to (4.7), denoted P ∗ , and if 0 < y ∗ < 1 (where y ∗ = y(P ∗ , T sat (P ∗ ))) then w is a
def

saturated state and we set

y eq (w) = y ∗ ,
z eq (w) = y ∗ τl∗ /τ,
ψ eq (w) = y ∗ ε∗l /ε,
seq (w) = y ∗ sl (wl∗ ) + (1 − y ∗ )sv (wv∗ ),
P eq (w) = Pl (wl∗ ) = Pv (wv∗ ),

where τl∗ = τl (P ∗ , T sat (P ∗ )), ε∗l = εl (P ∗ , T sat (P ∗ )), τv∗ = τv (P ∗ , T sat (P ∗ )) and ε∗v = εv (P ∗ , T sat (P ∗ ));
def def def def

(2) otherwise, if (4.7) has no solution or if the corresponding y ∗ = y(P ∗ , T sat (P ∗ )) is outside the range (0, 1),
def

(2a) if sl (w) > sv (w) then w is a liquid state, therefore we set

y eq (w) = 1,
z eq(w) = 1,
ψ eq (w) = 1,
seq (w) = sl (w),
P eq (w) = Pl (w);

(2b) if sl (w) < sv (w) then w is a vapor state, therefore we set

y eq (w) = 0,
z eq(w) = 0,
ψ eq (w) = 0,
seq (w) = sv (w),
P eq (w) = Pv (w).

4.3. Example of equilibrium states and phase-change equation for two stiffened gas laws
We present in this section the detailed construction of the phase-change equation (4.7) when both phases are
modeled by a stiffened gas EOS. Let us first recall a few properties of this particular state law. The complete
form of the stiffened gas EOS reads

(τ, ε) → s = cv ln(ε − q − πτ ) + cv (γ − 1) ln τ + m (4.8)

where parameters cv > 0, γ > 1, π > 0, q and m are constants describing thermodynamical properties of
the fluid. Remark that the domain of definition of s is not the usual cone C but rather the set τ > 0 and
ε − q − πτ > 0. The case of a perfect gas EOS is recovered by setting π = 0 and q = 0.
The classic definitions (3.1) provide the following expressions for the temperature T , the pressure P and the
Gibbs potential g:
ε − q − πτ
(τ, ε) → T = ,
cv
ε − q − πτ
(τ, ε) → P = (γ − 1) − π,
τ 
m  
(τ, ε) → g = q + (ε − q − πτ ) γ − − ln (ε − q − πτ )τ (γ −1) .
cv
1038 G. FACCANONI ET AL.

The Grüneisen coefficient Γ and the square of the speed of sound c are given by

Γ = − τ T (P sεε − sετ ) = γ − 1 > 0,

def

   
∂P  ∂P 
c =τ P
2 def 2
− = γ(γ − 1)(ε − q − πτ ) = γ(P + π)τ = γΓ cv T > 0.
∂ε  τ ∂τ  ε

By a simple change of variables we also have

(P, T ) → s = cv γ ln T − cv Γ ln(P + π) + m + cv γ ln cv + cv Γ ln Γ,
   Γ 
Γ
(P, T ) → g = cv T γ − ln (cv T )γ − T m + q,
P +π
P + πγ
(P, T ) → ε = cv T + q,
P +π
T
(P, T ) → τ = cv (γ − 1) ·
P +π

For the sake of simplicity we note q  = m + cv γ ln cv + cv Γ ln Γ as in [39, 50].

def

For all w ∈ C the value seq (w) is computed by solving numerically the phase-change equation (4.7) according
to Definition 4.5. In order to obtain this equation we have to compute the four functions

ταsat (P ) = τα (P, T sat (P )), εsat

α (P ) = εα (P, T α = l, v.
def def sat
(P )),

To achieve this task, we compute the coexistence curve P → T sat by solving equation (4.6b), namely

gl (P, T ) = gv (P, T ).

If it happens that the solution is multivalued, we discard the non-physical branch of solution.
With the stiffened gas EOS’s, two cases have to be considered separately:
• Case I: ql = qv . If and cvl γ l = cvv γ v , solving (4.6b) we obtain
 1/B
P → T sat (P ) = A (P + πl )cvl Γl (P + πv )−cvv Γv

where
 
ql − qv
A = exp 1 − > 0, B = cvl γ l − cvv γ v .
def def

Let us note that for two-perfect gas we recover the usual linear dependence between T sat and P . We recall
that hypothesis (3.9) implies some restrictions on the coefficients. An example of a stiffened gas that verifies
all hypothesis of the previous sections for pressures greater than 6.7 × 104 Pa is shown in Table 1 and
Figure 1. If cvl γ l = cvv γ v , we obtain a degenerate case where P sat is a constant defined by

(P sat + πl )cvl Γl
= exp(ql − qv ).
(P sat + πv )cvv Γv

• Case II: ql = qv . We were not able to find any explicit expression for the coexistence curve. Therefore it
is difficult to process the phase-change equation (4.7) for determining explicitly the equilibrium states. This
case appears in [39] for the approximation of the states laws of water and dodecane. We have proposed
in [19] and in Chapter 6 of [18] one possible mean for overcoming this difficulty: the key idea is to compute
P → Tsat (P ) as a simple and convenient approximation of the function P → T sat (P ). To achieve this task,
 LIQUID-VAPOR PHASE TRANSITION 1039

Table 1. Example of parameters for two stiffened gas law that provides an explicit expression
for the coexistence curve. These values are used in the tests of Section 8. This choice guarantees
that the volume and the internal energy of the liquid are lower than the volume and the internal
energy of the vapor while heaving a speed of sound not too large (which implies a not too strict
CFL condition for the computational tests).

Phase cv (J/(kg K)) γ π (Pa) q (J/kg) q  (J/(kg K))

Liquid 800 1.4 105 0 1000
Vapor 1100 1.3 0 0 0

100 0.14 1e+05

sat
T (P) τsat
v (P)
95 τsat
l (P)
0.12 1e+05
90

85 0.1 9e+04

80
0.08 8e+04
75

70 0.06 7e+04

65
0.04 6e+04
60 εsat
v (P)
εsat
l (P)
55 0.02 5e+04
2e

3e

4e

5e

6e

7e

8e

9e

1e
2e

3e

4e

5e

6e

7e

8e

9e

1e

2e

3e

4e

5e

6e

7e

8e

9e

1e
+0

+0

+0

+0

+0

+0

+0

+0

+0
+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0

+0
5

6
5

6
(a) P → T sat (P ) (b) P → τlsat (P ) (bottom) and P → (c) P → εsat
l (P ) (bottom) and P →
τvsat (P ) (top) εsat
v (P ) (top)

Figure 1. Temperature and phasic volumes and internal energies at saturation as functions of
the pressure for the stiffened gas law of Table 1.

first we discretize an interval of pressure [Pmin , Pmax ] and we compute (numerically but with an arbitrary
precision) Tsat at each point of the discretization as the solution of equation (4.6b). Thus we obtain a set of
points A = (Pi , T sat (Pi ))i . Then we can define P → Tsat (P ) thanks to a least square approximation of A .
Finally, instead of solving the phase-change equation (4.7) we solve the following approximate phase-change
equation:
τ − τv (P, Tsat (P )) ε − εv (P, Tsat (P ))
= ·
τl (P, Tsat (P )) − τv (P, Tsat (P )) εl (P, Tsat (P )) − εv (P, Tsat (P ))
This approximation is reasonable since, in the end, the phase-change equation (exact or approximate) is
always solved numerically. We already tested this approach for numerical examples in [18, 19, 21].
Remark 4.6 (tabulated data). The case of more general EOS, including non-analytic forms of the EOS, like
tabulated data, is addressed in [20, 21] and Chapter 7 of [18]. Again the main idea is to build an approximate
phase-change equation. In order to obtain this approximate equation, we need to evaluate the four functions
ταsat (P ) = τα (P, T sat (P )), εsat
α (P ) = εα (P, T α = l, v.
def defsat
(P )),
We consider a set of experimental data values ταsat (Pi ) and εsat
α (Pi ) for a discrete set of pressure values Pi ∈
[Pmin , Pmax ] (see e.g. [41] for an example of tabulated EOS). Then, we can define
ταsat (P ), εsat
α (P ) α = l, v, for P ∈ [Pmin , Pmax ],
thanks to least square approximations. Finally, instead of solving the phase-change equation (4.7) we solve the
following approximate equation
τ − τvsat (P ) ε − εsat
v (P )
= ·
τlsat (P ) − τvsat (P ) εsat
l (P ) − εsat
v (P )

A similar construction of equilibrium EOS is proposed in [30] too.

1040 G. FACCANONI ET AL.

5. Properties of the equilibrium phase transition model

In this section we recall two properties of the convective part of the system (2.3), which corresponds to
the Euler system equipped with the liquid-vapor thermodynamical equilibrium EOS, pertaining to the well-
posedness of the system. For the sake of completeness, proofs of these theorems are provided in Appendix 9 at
page 1050 and in Appendix 9 at page 1052 (these results are announced in [2]).
The first theorem is a technical result that provides informations about the first and second derivatives of
the equilibrium EOS w → seq .

Theorem 5.1. Let w ∈ C be a saturated state (defined in Thm. 3.2) and let r(w) =(M∗l , M∗v ) be the seg-
def

ment associated with w (the segment r(w) and points M∗α are defined in Thm. 4.3). Then, for every point
M =(w, seq (w)) of r(w), we have:
def

(A) the entries of the Hessian matrix d2 seq , evaluated at w, verify

∗ ∗ ∗ ∗
τ ε (εl − εv ) + sτ τ (τl − τv ) = 0,
seq eq
(5.1a)
∗ ∗ eq ∗ ∗
εε (εl − εv ) + sτ ε (τl − τv ) = 0,
seq (5.1b)
τ τ sεε = (sτ ε ) ,
seq eq eq 2
(5.1c)
 ∗ 2
εl − ε∗v
seq
ττ = seq
εε ; (5.1d)
τl∗ − τv∗

τ τ < 0,
seq (5.2a)
seq
εε < 0, (5.2b)
seq
τε > 0; (5.2c)

(B) the pressure P eq , evaluated at w, verifies

(ε∗l − ε∗v )
P eq > − · (5.3)
(τl∗ − τv∗ )

Theorem 5.1 shed light on some important properties of the Hessian matrix of the equilibrium entropy seq
that departs from the classic pure fluid framework. Indeed, the EOS w → seq verify the inequality (3.3c) but
does not verify the inequality (3.3b). More precisely,
• for a pure state w ∈ C, d2 seq is a negative definite matrix, namely:

vT · d2seq (w) · v < 0 ∀v = (0, 0),

which implies that

τ τ (w) < 0,
seq seq
εε (w) < 0, seq
τ τ (w)sεε (w) > (sτ ε (w)) ;
eq eq 2

• for a saturated state w ∈ C, d2 seq is no longer negative definite, it is merely nonpositive and d2 seq = ( 00 00 ),
namely: 
vT · d2seq (w) · v ≤ 0 ∀v ∈ R2 ,
∃! v(w) = (0, 0) such that (v(w))T · d2seq (w) · v(w) = 0.

τ τ (w) < 0, sεε (w) < 0, sτ τ (w)sεε (w) = (sτ ε (w)) and moreover we have sτ ε (w) > 0.
This implies that seq eq eq eq eq 2 eq

Theorem 5.1 allows us to prove that the Euler system equipped with the equilibrium EOS is strictly hyperbolic.
This property is important for the well-posedness of the system and its numerical resolution.
 LIQUID-VAPOR PHASE TRANSITION 1041

Theorem 5.2 (hyperbolicity). For smooth solutions, the compressible Euler system equipped with the liquid-
vapor equilibrium EOS w → seq reads ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎛ ⎞ u   0
   0
⎜ 1 ∂P eq  u 1 ∂P∂ε  ⎟
eq
∂t ⎝u⎠ + ⎝ ∂  ⎠ ∂x ⎝u⎠ = ⎝0⎠
ε
ε 0 P eq
u ε 0 (5.4)

1 seq
with = , P eq = τeq ·
τ sε
Then, for all w ∈ C, the square of the speed of sound ceq , defined by (3.2a), is strictly positive and therefore
ceq , defined by (3.8), is real and the system (5.4) is strictly hyperbolic.

Remark 5.3. As a corollary of our Theorem 5.2, the MTT-equilibrium system defined in [10] or the Euler
system provided with the EOS defined in [4, 29, 31, 34, 40, 47, 55] are strictly hyperbolic.
Remark 5.4. We emphasize that the positivity of the speed of sound does not seem totally trivial. First,
the function w → seq is concave (but not strictly concave) and its Hessian matrix is not negative definite.
Concavity provides that c2 ≥ 0, but we do not have a priori informations that ensures that c2 = 0. Second, for
some systems as in [10,11,26] the loss of strict concavity of the entropy indeed leads to a zero speed of sound: in
this case hyperbolicity is lost because the system eigenvalues are real but the Jacobian matrix of the flux is no
longer diagonalizable. This situation leads to ill-posed problems (see for example [42], p. 362) where uniqueness
is lost within the classical class of entropy solutions.

Remark 5.5. Theorem 5.1 implies that: (1) the flux function in the Euler equations is only piecewise regular;
(2) the functions w → (P eq , T eq )(w) are continuous and piecewise C1 regular: their derivatives are discontinuous
when the fluid changes from a pure phase state y ∈ {0, 1} to a saturated state y ∈ (0, 1); (3) the speed of sound
w → ceq is piecewise continuous: it is discontinuous along the saturation curve, namely across y ∈ {0, 1} regions
and 0 < y < 1 regions.

6. Relaxation approach
The definition of the equilibrium entropy seq as the maximizer of the mixture entropy σ suggests to read the
convective part of the system (2.3), which consists of the Euler equations closed by the EOS w → seq , as the limit
of a relaxed augmented Euler system equipped with the EOS (y, z, ψ, w) → σ (see for example [7, 12, 13, 43]).
This procedures helps to break the implicit definition of the equilibrium EOS and enables to use of simple
discretization strategies. For example, in [29, 31] the authors proposed to consider the system
⎧
⎪
⎪∂t U + div[F(U, y, z, ψ)] = 0,
⎪
⎨∂ y + u · grad y = (y − y eq )μ ,
t y
(6.1)
⎪
⎪ ∂t z + u · grad z = (z − z eq)μz ,
⎪
⎩∂ ψ + u · grad ψ = (ψ − ψ eq )μ ,
t ψ

with
⎛ ⎞
u
F(U, y, z, ψ), = ⎝ u ⊗ u + P Id ⎠,
def

2
((ε + |u| /2) + P )u


∂σ  zα PTαα
def ∂τ ε,z,y,ψ α=l,v
P (w, y, z, ψ) = ∂σ  = 
 ψα T1α
∂ε τ,z,y,ψ
α=l,v

and y , z , ψ
eq eq eq
defined in Definition 4.5. The coefficients μy , μz and μψ are positive relaxation parameters.
1042 G. FACCANONI ET AL.

Formally, when μy , μz , μψ → +∞ we obtain

(y, z, ψ) → (y eq , z eq , ψ eq )(w),
σ(y, z, ψ, w) → seq (w),
F(U, y, z, ψ) → Feq (w)

and the solution U of (6.1) tends to the solution of the convective part of (2.3).
When the mass fraction satisfies 0 < y < 1, we remark that it is also possible to consider the following
alternative augmented system ⎧
⎪
⎪ ∂t U + div[F(U, y, z,
 ψ)] = 0,
⎪
⎪
⎪
⎨∂t y + u · grad y = Tgll − Tgvv μy ,
  (6.2)
⎪
⎪ ∂t z + u · grad z = PTvv − PTll τ μz ,
⎪
⎪  
⎪
⎩∂t ψ + u · grad ψ = 1 − 1 εμψ ,
Tl Tv

and, in this case, for smooth solutions we have the following entropy evolution equation
 2  2  2
gl gv Pv Pl 1 1
∂t (σ) + div(σu) = μy  − + μz − + μψ ε − ≥ 0.
Tl Tv Tv Tl Tl Tv

In the regime μy , μz , μψ → +∞, one retrieves the solution of the convective part of (2.3) for y ∈ (0, 1).
In order to approximate the solution of the convective part of the system (2.3), following similar lines as [10,18]
we consider another augmented system based on the five-equation system with the isothermal closure presented
in [1, 36]. Using (6.3) instead of (6.1) allows us to re-use the code developed in [1, 36] and to modify it in a
non-intrusive manner thanks to an additional module that computes the projection onto the equilibrium states.
The system reads ⎧
⎪
⎨∂t U + div[G(U, y, z)] = 0,
∂t y + u · grad y = (y − y eq )μy , (6.3)
⎪
⎩∂ z + u · grad z = (z − z eq)μ ,
t z

where
⎛ ⎞
u 
G(U, y, z) = ⎝ u ⊗ u + Π Id ⎠ , Π(w, y, z) = zα Pα , Tl = Tv .
def def

2
(ε + |u| /2) + Π)u α=l,v

The solution of (6.3) formally tends to the solution of the convective part of (2.3) in the limit regime μy , μz →
+∞.

Remark 6.1. The system (6.3) can be considered as a partial equilibrium of (6.1) in the formal limit
μψ → +∞. In this case one enforces ψ = ψ eq (w), which is equivalent to Tl = Tv for y ∈ (0, 1) and implies
P (w, y, z, ψ eq (w)) = Π(w, y, z).

Remark 6.2. The non-dissipative structure of both systems (6.1) and (6.3) can be derived thanks to a least
action principles (see [10, 11, 18]).

7. Numerical scheme
We present in this section the numerical strategy we implemented in order to solve the equilibrium system (2.3)
with heat diffusion, gravity and surface tension effects. Following the analysis of Section 6, we chose a relaxation
 LIQUID-VAPOR PHASE TRANSITION 1043

approach that consists in approximating the solution of the following system

⎧
⎪
⎨∂t U + div[G(U, y, z)] = Stension (U, z) + Sgravity (U) − div[Qheat (U, z)]
∂t (y) + div(yu) = (y − y eq )μy , (7.1)
⎪
⎩∂ z + u · grad z = (z − z eq )μ ,
t z

in the limit regime μy , μz → +∞.

Let (U, y, z)n be the discretized state variable at time t = tn . The overall numerical solver is composed of
the following steps:
Step I. Convection. Update (U, y, z)n to (U, y, z)n+I by solving the convective terms of the system (7.1),
i.e. Stension (U, z) = 0, Sgravity (U) = 0, Qheat (U, z) = 0, μy = 0 and μz = 0. In our numerical tests, this
step is achieved thanks to the second-order explicit finite volume Roe-type solver proposed in [1, 36]. For
stability purpose the time step is constrained by a CFL condition (see [1, 36] for details);
Step II. Capillarity effects. Update (U, y, z)n+I to (U, y, z)n+II by solving ∂t U = Stension (U, z) thanks to
the MAC algorithm proposed in [8];
Step III. Gravity effects. Update (U, y, z)n+II to (U, y, z)n+III by integrating ∂t U = Sgravity (U) with an
explicit centered discretization;
Step IV. Thermal diffusion. Update (U, y, z)n+III to (U, y, z)n+IV by solving ∂t U = − div[Qheat (U, z)]
with a standard implicit finite volume scheme;
Step V. Projection onto the equilibrium states. Update (U, y, z)n+IV to (U, y, z)n+1 by computing the
formal limit μy , μz → +∞. This boils down to set

n+1 = n+IV ,
εn+1 = εn+IV ,
un+1 = un+IV ,
y n+1 = y eq (1/n+1 , εn+1 ),
⎧
⎪
⎨0, if y n+1 = 0,
n+1
z = y n+1 n+1 τvn+1 , if 0 < y n+1 < 1,
⎪
⎩1, if y n+1 = 1.

More precisely, we seek P ∗ by means of a dichotomy algorithm as the solution of the phase-change equa-
tion (4.7) with τ = 1/n+1 and ε = εn+1 , namely here:

n+1 − τv (P )
1 sat
εn+1 − εsatv (P )
= ·
τlsat (P ) − τvsat (P ) εsat
l (P ) − εsat
v (P )

Then, following definition 4.5, we set

⎛ ⎞n+1 ⎛ ⎞eq
P P  
⎝y ⎠ ⎝ ⎠ 1 n+1
= y ,ε .
z z n+1

This completes the definition of (U, y, z)n+1 .

Remark 7.1. The functions (τ, ε) → P ∗ and (τ, ε) → y ∗ can be computed once and for all prior to the start of
any flow simulation. Let us note that although the pressure P eq is merely continuous and piecewise C1 regular
(cf. Rem. 5.5), this scheme did not generate spurious oscillations in the solutions of our tests.
1044 G. FACCANONI ET AL.

L1 =1 m

Wall
T =T0
P =P sat (T0 )
= (P sat (T0 )T0 )
liq

L2 =1 m
Piston
T =T0

Wall
P =P sat (T0 )
=vap (P sat (T0 )T0 )

x2
x1
Wall

Figure 2. Geometry of the test “8.1. compression of vapor bubble”.

8. Numerical simulations
The present tests challenge the ability of our approach to reproduce cavitation phenomena and nucleation
phenomena in pool boiling type conditions. As the last test involves gravity, surface tension and thermal diffusion,
in order to obtain not exceedingly long calculation time, we choose to carefully fit the state laws and the physical
parameters. Indeed, the CFL constraint imposed by the explicit scheme of the step I does not allow to use laws
fitted on real water tabulated data: the driving phenomena for such tests are thermal diffusion and gravity
and are too small compared to the speed of sound appearing in the CFL constraint. However we emphasize
that there is no technical difficulty in using our approach with real fluid laws, even tabulated data (see for
example the tests that reproduce cavitation phenomena in [18, 19]). Theses improvements will be presented in
forthcoming papers.

8.1. Compression of vapor bubble

Our model is tested here on the case of vapor bubble compression (see Fig. 2). We consider a 1 m side-
length 2D square domain discretized over a 400 × 400 cell mesh. Vapor bubble is surrounded by liquid phase
in the center of the domain. The radius is initially r = 0.15 m. Both EOS used are “Stiffened Gas” type whose
coefficients are given in Table 1. The initial pressure is fixed to P0 = 13 046.85692 Pa and the fluid is initially at
rest (u = 0) in the whole domain. Both phases are supposed to be at saturation at t = 0, therefore the densities
3 3
for the liquid and the vapor are respectively liq = 35.32714277 kg/m , vap = 3.953593004 kg/m . We suppose
the left boundary to be a piston. We proceed by duplicating the value of , y, z, P and transverse velocity from
the last cell close to the boundary into to the fictitious cell. Then we set the normal velocity to up . We chose
here up = 1.0 m/s. Other boundary conditions are reflective walls.
Figure 3 shows the volume fraction, the mass fraction, the density, the pressure and the temperature for time
varying from t = 0 ms to t = 100 ms. The color field corresponds to the phases: light color around z = y = 0
(vapor) and dark around z = y = 1 (liquid). From the top to the bottom the following sequence can be seen:
the moving piston generates a pressure wave that compresses the vapor. As the pressure increases the vapor
starts to condensate and the bubble shrinks.
 LIQUID-VAPOR PHASE TRANSITION 1045

Figure 3. Time evolution of the volume fraction (first column), the mass fraction (second
column), the density (third column), the pressure (fourth column) and the temperature (last
column) of the test “8.1. compression of vapor bubble”.
1046 G. FACCANONI ET AL.

L1 =1 m

Wall
T =T0
P =P sat (T0 )
= (P sat (T0 )T0 )
liq

Shock at t = 0

L2 =1 m
T =T0
Wall

Wall
P =P sat (T0 )
= (P sat (T0 )T0 )
vap

x2
x1
Wall

Figure 4. Geometry of the test “8.2. liquefaction and vaporisation caused by a shock and its
bounce”.

8.2. Liquefaction and vaporisation caused by a shock and its bounce

Our model is tested here on the case of a phase change caused by a shock and its rebounding (see Fig. 4 for
the initial configuration). We consider a 1 m side-length 2D square domain discretized over a 200×200 cell mesh.
Vapor bubbles in the right of the domain are surrounded by liquid phase. The radius of each bubble is initially
r = 0.04 m. Both EOS used are “Stiffened Gas” type whose coefficients are given in Table 1. Both phases are
supposed to be at saturation at t = 0 and the thermodynamical conditions are as in the test “compression of
vapor bubble”. The fluid is initially at rest (u = 0) in the whole domain. To mimic an initial shock we impose a
moving piston boundary condition (cf. test 8.1) with velocity up = 100 m/s only for the first time step. In the
subsequent time step we enforce reflective wall boundary condition. Other boundary conditions are reflective
walls.
Figure 5 shows the volume fraction, the mass fraction, the density, the pressure and the temperature for time
varying from t = 0 ms to t = 100 ms. The color field corresponds to the phases: light color around z = y = 0
(vapor) and dark around z = y = 1 (liquid). The shock compresses the vapor bubbles which liquefy as in the
test “compression of vapor bubble”. But after the bounce of the shock on the right wall, the pressure decreases
and the liquid vaporizes at the original location of the bubbles.

8.3. Bubbles nucleation, detachment and rise

To study the ability of the model to deal with two-phase flows with phase change, we study numerically the
nucleation and the motion of bubbles inside a closed box under the action of capillary, gravity and thermal
flux imposed. We consider a fluid whose liquid phase and vapor phase are modeled by two stiffened gas EOSs.
The thermodynamics parameters are given in Table 1. The gravity acceleration is g = (0, −100) m/s2 . The
thermal conductivity are ϑl = 67.8 W/(m K) and ϑv = 4.24 W/(m K) and the surface tension is ξ = 7 N/m.
The computational domain is a square of 1 m side-length initially filled with the liquid phase (see Fig. 6). The
liquid is initially at rest (u = 0) with uniform temperature T0 = 5.25 K. The initial pressure is given by the
hydrostatic pressure with a top reference pressure P ref = 6.8 × 103 Pa > P sat (T0 ) ≈ 6.73 × 103 Pa. At the top,
we keep both pressure and temperature constant at P = P ref and T = T ref = 1.0 K. At the bottom we consider
 LIQUID-VAPOR PHASE TRANSITION 1047

Figure 5. Time evolution of the volume fraction (first column), the mass fraction (second
column), the density (third column), the pressure (fourth column) and the temperature (last
column) of the test “8.2. liquefaction and vaporisation caused by a shock and its bounce”.
1048 G. FACCANONI ET AL.

L1 =1 m
Pressure and temperature imposed
P = P ref , T = T ref

Wall, Null heat flux

Wall, Null heat flux

Liquid at

L2 =1 m
g
T = T0 ,
hydro
P = Piso−T

x2
x1
Wall, heat flux imposed
Φs

Φi

Figure 6. Geometry of the test “8.3. bubbles nucleation, detachment and rise”.

wall boundary condition for the convection operator and we impose the following thermal flux profile:
1
Φ(x1 , x2 = 0, t) = ((Φs + Φi ) + (Φs − Φi ) cos(4π(x1 − 0.5)))
def

with Φi = 0 W/m2 and Φs = 20 Wm2 . We impose a homogeneous Neumann boundary conditions for both
convection and thermal diffusion operators at the left and right boundary. For our test, the domain is discretized
with a 100 × 100 cell mesh.
Figure 7 displays the results obtained for the volume fraction, the mass fraction, the density, the pressure
and the temperature. We observe that nucleation of vapor bubbles occurs at the point of maximum heat flux at
the bottom boundary. The bubble of vapor grows due to the phase change, then it is lifted off the boundary by
gravity effects. One can notice that after the first bubble takes off, a portion of vapor remains at the bottom.
As the bubble rises it reaches a region where the liquid is cooler. The vapor is then cooled by thermal diffusion
and the fluid liquefies. But very quickly, the nucleation sites at the bottom starts to spread and merge at the
bottom of the domain.

9. Conclusion and perspectives

We investigated the simulation of dynamical liquid-vapor phase change in compressible flows. Our model
considers both phase to be compressible fluids equipped with their own EOS. Local and instantaneous inter-
phase equilibria are assumed which provides the definition of an equilibrium EOS for the two-phase medium
by means of an equilibrium entropy law. We reviewed the connection between this equilibrium entropy and the
concave hull of the pure phases entropy law. This allowed us to prove that even though the equilibrium entropy
is not strictly concave, the resulting equilibrium system is strictly hyperbolic. The definition of the equilibrium
entropy is obtained thanks to an implicit nonlinear system of equations that involves the phasic variables.
Nevertheless, we showed that it is possible to reduce this system to a nonlinear scalar equation (4.7) based
on the saturation law of the pure phases, which are common experimental data. We exploited this property
in our numerical algorithm. A two-step relaxation strategy was used to decouple the approximation of the
convection operator from the evaluation of the equilibrium EOS. While hydrodynamics is resolved in the first
 LIQUID-VAPOR PHASE TRANSITION 1049

Figure 7. Time evolution of the volume fraction (first column), the mass fraction (second
column), the density (third column), the pressure (fourth column) and the temperature (last
column) of the test “8.3. bubbles nucleation, detachment and rise”.
1050 G. FACCANONI ET AL.

step by means of a simple two-phase model, the computation of the inter-phase equilibria is confined within
the last step that consists in projecting the variables onto the equilibrium states. We proposed two-dimensional
numerical simulations of boiling phenomena involving Stiffened Gas EOS for the pure phases, surface tension,
gravity and thermal diffusion. The results showed that the model can reproduce nucleation mechanisms, in
the sense that vapor can be created in an initially pure liquid domain. Even thought theses results are only
qualitative they are important toward the study of the bubble growth dynamics.
The present work can be enriched in several ways. From a modeling point of view, additional physical effects
may be taken into account such as fluid viscosity and contact angle to model the transition from nucleate boiling
to film boiling. From a numerical point of view, the simulation of boiling-type tests with realistic physical laws
involves large time-scale processes compared to the sound velocity of the medium. This limitation may be
circumvented by implementing implicit numerical methods for the convection operator.

Appendix A. Proof of Theorem 5.1

Proof of (5.1). The jump of specific volume, energy and entropy (postulated at hypothesis 3.9) implies that for
every saturated state w

0 < y ∗ < 1, τl∗ < τv∗ , ε∗l < ε∗v , s∗l =(sl (wl∗ )) < s∗v =(sv (wv∗ )).
def def

In this case, τα∗ ,ε∗α and y ∗ are such that w = y ∗ wl∗ + (1 − y ∗ )wv∗ and seq (w) = y ∗ s∗l + (1 − y ∗ )s∗v .
def seq
Along the segment r(w), the pressure P eq = sτeq and the temperature T eq = s1eq
def

ε ε
are constant. Then, the
directional derivatives read
 eq 
P
0=d · (wl∗ − wv∗ ) = seq ∗ ∗ eq ∗ ∗
τ ε (εl − εv ) + sτ τ (τl − τv ),
T eq
 
1
0=d · (wl∗ − wv∗ ) = seq ∗ ∗ eq ∗ ∗
εε (εl − εv ) + sτ ε (τl − τv ),
T eq
which proves (5.1a) and (5.1b). Then (5.1c) follows by taking the determinant of the singular Hessian matrix.
Eventually (5.1d) is a simple consequence of (5.1a) and (5.1b). 
Proof of (5.2). By construction w → seq is concave. This implies that seq τ τ ≤ 0 and sεε ≤ 0. However w → s
eq eq

is not strictly concave. Since we proved the inequalities (5.1), it is sufficient to prove that seqττ (w)
= 0 for all
saturated state w.
We proceed by contradiction. Let w be a saturated state such that seq τ τ (w) = 0. By equalities (5.1b)–(5.1c),
the Hessian matrix is null at w.
We note (P , T , g) =(Pα , Tα , gα )(w) for α = l, v (we know that P eq , T eq and g eq are constant along the
def

segment r(w)).
We consider a regular C2 curve inscribed on the graph S of seq and parametrized as follows

t ∈ [−1, 1] → (w, γ = seq (w))(t)

def

dw∗ (t=0) ∗
such that w(0) = w and α
dt = 0 with t ∈ [−1, 1] → wα (t) the ends of the line segment t ∈ [−1, 1] →
r(w(t)).
For the sake of simplicity we note
dseq d2 seq
, d2 seq (w(t)) = ·
def def
dseq (w(t)) =
dw dw2
We have
d2 seq d2 w(t) dw T 2 eq dw(t)
γ  (t) = = ds eq
(w(t)) + d s (w(t)) ·
dt2 dt2 dt dt
 LIQUID-VAPOR PHASE TRANSITION 1051

Then, for t = 0, we have

d2 w(0)
w(0) = w, d2seq (w) = ( 00
00 ) , γ  (0) = dseq (w) ·
dt2
We now proceed to another computation of γ  (0). By the implicit function theorem, in a vicinity of 0, there
exist C2 functions t → yα∗ and t → wα
∗
such that
 
w(t) = yα∗ (t)wα
∗
(t), γ(t) = yα∗ (t)s∗α (wα
∗
(t)),
α=l,v α=l,v

where yl∗ = y ∗ et yv∗ = 1 − y ∗ . We note

dsα d2 s∗α
ds∗α = dsα (wα
∗
, d2 s∗α = d2 sα (wα
∗
·
def def def def
(t)) = (t)) = 2
dwα dwα

We have
d2 w   d2 y ∗ dyα∗ dwα
∗ 2 ∗

α ∗ ∗ d wα
= w +2 + yα (A.1)
dt2 dt2 α dt dt dt2
α=l,v

and  T !
 d2 yα∗ ∗ dyα∗ ∗ dwα ∗ 2 ∗ ∗ ∗
∗ d wα dwα dw α
γ  (t) = s α + 2 ds α + y ∗
α ds α + yα∗ d2 s∗α ,
dt2 dt dt dt2 dt dt
α=l,v

s∗α ∗
sα (wα ∗
(0)) = dseq (w(0)) (namely Pα (wα∗
(0)) = P (w(0)) and
def
where = ).
Since, by Proposition 4.2, dsα (wα
∗
Tα (wα (0)) = T (w(0))) we obtain
  dy ∗ ∗ 2 ∗
   dy ∗ dw∗ 2 ∗

α ∗ dwα ∗ ∗ d wα α α ∗ d wα
2 dsα + yα dsα eq
= ds (w) 2 + yα ·
dt dt dt2 dt dt dt2
α=l,v α=l,v

By relation (A.1), we have then at t = 0

  dy ∗ dw∗ 2 ∗
 2  d2 y ∗
α α ∗ d wα eq d w α
eq
ds (w) 2 + yα = ds − ds∗α wα
∗
dt dt dt2 dt2 dt2
α=l,v α=l,v

which implies

d2 w(0) d2 y ∗ ∗   dw∗ T dwα∗

α
γ  (0) = dseq (w) 2
− 2
(s l − ds ∗ ∗
w
l l − s ∗
v + ds ∗ ∗
w
v v ) + yα∗ d2 s∗α
dt dt dt dt
α=l,v

that reads    T
d2 w(0) d2 y ∗ gl gv  dwα∗
dwα∗

γ (0) = ds (w) eq
− − + yα∗ d2 s∗α ,
dt2 dt2 Tl Tv dt dt
α=l,v
gα
as s∗α − ds∗α wα
∗
= s∗α − PTαα τα∗ − 1 ∗
Tα εα = Tα .
Since Tgαα = Tg , this implies
 T
d2 w(0)  ∗ dwα∗
dwα∗
γ  (0) = dseq (w) + yα d2s∗α ·
dt2 dt dt
α=l,v
1052 G. FACCANONI ET AL.

Finally we have
 T
d2 w(0) d2 w(0)  ∗ dwα∗
dwα∗
dseq (w) 2
= γ  (0) = dseq (w) + yα d2s∗α ·
dt dt2 dt dt
α=l,v

Consequently, we deduce
  ∗
T  ∗

dwα dwα
0= yα∗ d2 s∗α · (A.2)
dt dt
α=l,v

dw∗ (t=0)
As α
dt = 0 and the Hessian matrices d2 s∗α of the phasic entropies sα are negative definite, (A.2) is
impossible.
∗ ∗ ∗ ∗
Therefore seq
τ ε = 0 and since sgn(εl − εv ) = sgn(τl − τv ), thanks to relation (5.1b) we obtain the inequality

ε∗l − ε∗v
seq
τ ε = −sεε
eq
> 0. 
τl∗ − τv∗

Proof of (5.3). The inequality involving P eq follows from the definition of the Gibbs potential. Indeed, along
the segment r(w) we have

0 = gl − gv = (ε∗l − ε∗v ) + P eq (τl∗ − τv∗ ) − T eq(s∗l − s∗v ).

Granting the assumptions T eq > 0, τl∗ < τv∗ and s∗l < s∗v , we obtain that

(ε∗l − ε∗v ) ∗ ∗
eq (sl − sv ) (ε∗l − ε∗v )
P eq = − + T > − ·
(τl∗ − τv∗ ) (τl∗ − τv∗ ) (τl∗ − τv∗ )

This concludes the proof. 

Proof of Theorem 5.2

In one space dimension, the Jacobian matrix of the flux Feq has three eigenvalues

λ1 = u − ceq , λ2 = u, λ3 = u + ceq ,
def def def

with the corresponding three right eigenvectors

⎛ ⎞ ⎛ 
∂P eq 
⎞ ⎛ ⎞
2 ∂ε 2
⎜ 0  ⎟
r1 = ⎝ −ceq  ⎠ , ⎠, r3 = ⎝ ceq  ⎠ .
def def def
r2 = ⎝
eq 
P eq − ∂P
∂  P eq
ε

Therefore, we only have to check that (c ) > 0 which immediately ensures that the three eigenvalues are
eq 2

distincts and the three eigenvectors are linearly independent. As P eq = seq τ /sε , the square of the speed of sound
eq

verifies (3.2b), namely   eq

 eq  seqεε seq
τ ε P
(c ) = −τ T
eq 2 2 eq
P , −1 . (A.3)
seq s eq
ετ τ τ −1
We distinguish two cases:
Pure phase states. (y ∈ {0, 1}): if w is a pure phase state then seq (w) ≡ sα (w) with α = l if y = 1 and
α = v if y = 0. Therefore, using (3.3) in (A.3), we obtain that the entropy w → seq satisfies

τ 2 T eq eq eq 2
(ceq )2 > − (P sετ − seq
τ τ ) ≥ 0.
seq
ττ
 LIQUID-VAPOR PHASE TRANSITION 1053

Saturated states. (0 < y < 1): in Theorem 5.1 we proved that if w is a saturated state then w → seq does
not satisfies (3.3b). However, taking into account (5.1c) we obtain from (A.3)
1
(ceq )2 = −τ 2 T eq (P eq seq
ετ − sτ τ ) ,
eq 2
seq
ττ

and thanks to the relation (5.1a), we finally have

 2
(seq )2 ε∗l − ε∗v
(c ) = −τ
eq 2 2
T eq ετ P + ∗
eq
·
seq
ττ τl − τv∗

Thanks to the inequalities (5.2) and (5.3), this shows that (ceq )2 > 0.
We can conclude that for all w ∈ C, (ceq (w))2 is positive, which implies that ceq (w) is real and = 0 and the
system (5.4) is strictly hyperbolic.

References
[1] G. Allaire, S. Clerc and S. Kokh, A five-equation model for the simulation of interfaces between compressible fluids. J. Comput.
Phys. 181 (2002) 577–616.
[2] G. Allaire, G. Faccanoni and S. Kokh, A strictly hyperbolic equilibrium phase transition model. C. R. Acad. Sci. Paris, Sér. I
344 (2007) 135–140.
[3] K. Annamalai and I.K. Puri, Advanced thermodynamics engineering. CRC Press (2002).
[4] Th. Barberon and Ph. Helluy, Finite volume simulations of cavitating flows. Comput. Fluids 34 (2005) 832–858.
[5] J. Benoist and J.-B. Hiriart-Urruty, What is the subdifferential of the closed convex hull of a function? SIAM J. Math. Anal.
27 (1996) 1661–1679.
[6] S. Benzoni Gavage, Stability of multi-dimensional phase transitions in a Van der Waals fluid. Nonlinear Anal. 31 (1998)
243–263.
[7] F. Bouchut, Nonlinear stability of finite volume methods for hyperbolic conservation laws and well-balanced schemes for
sources. Frontiers in Mathematics. Birkhäuser Verlag, Basel (2004).
[8] J.U. Brackbill, D.B. Kothe and C. Zemach, A continuum method for modeling surface tension. J. Comput. Phys. 100 (1992)
335–354.
[9] H.B. Callen, Thermodynamics and an introduction to thermostatistics. John Wiley & sons, 2nd edition (1985).
[10] F. Caro, Modélisation et simulation numérique des transitions de phase liquide-vapeur. Ph.D. thesis, École Polytechnique
(2004). [Link]
[11] F. Caro, F. Coquel, D. Jamet and S. Kokh, A simple finite-volume method for compressible isothermal two-
phase flows simulation. International Journal on Finite Volumes (2006). [Link]
[Link].
[12] G. Chen, C.D. Levermore and T.-P. Liu, Hyperbolic conservation laws with stiff relaxation terms and entropy. Comm. Pure
Appl. Math. 47 (1992) 787–830.
[13] F. Coquel and B. Perthame, Relaxation of energy and approximate Riemann solvers for general pressure laws in fluids dynamics.
SIAM J. Numer. Anal. 35 (1998) 2223–2249.
[14] J.-M. Delhaye, M. Giot and M.L. Riethmuller, Thermohydraulics of two-phase systems for industrial design and nuclear
engineering. Hemisphere Publishing Corporation (1981).
[15] J.-M. Delhaye, M. Giot, L. Mahias, P. Raymond and C. Rénault, Thermohydraulique des réacteurs. EDP Sciences (1998).
[16] V.K. Dhir, Boiling heat transfer. Ann. Rev. Fluid Mech. 30 (1998) 365–401.
[17] J.E. Dunn and J. Serrin, On the thermomechanics of interstitial working. Arch. Rational Mech. Anal. 88 (1985) 95–133.
[18] G. Faccanoni, Étude d’un modèle fin de changement de phase liquide-vapeur. Contribution à l’étude de la crise d’ébullition.
Ph.D. thesis, École Polytechnique, France (2008). [Link]
[19] G. Faccanoni, S. Kokh and G. Allaire, Numerical simulation with finite volume of dynamic liquid-vapor phase transition, Finite
Volumes for Complex Applications V. ISTE and Wiley (2008) 391–398.
[20] G. Faccanoni, G. Allaire and S. Kokh, Modelling and numerical simulation of liquid-vapor phase transition, in Conf. Proc. of
EUROTHERM-84, Seminar on Thermodynamics of Phase Changes, Namur (2009).
[21] G. Faccanoni, S. Kokh and G. Allaire, Approximation of liquid-vapor phase transition for compressible fluids with tabulated
EOS. C. R. Acad. Sci. Paris Sér. I 348 (2010) 473–478.
[22] H. Fan, One phase Riemann problem and wave interactions in systems of conservation laws of mixed type. SIAM J. Math.
Anal. 24 (1993) 840–865.
[23] H. Fan, Traveling waves, Riemann problems and computations of a model of the dynamics of liquid/vapor phase transitions.
J. Differ. Equ. 150 (1998) 385–437.
1054 G. FACCANONI ET AL.

[24] H. Fan and M. Slemrod, The Riemann problem for systems of conservation laws of mixed type, in Conf. Proc. on Shock Induced
Transitions and Phase Structure in General Media Institute of Mathematics and its Applications. Minneapolis (1990) 61–91.
[25] C. Fouillet, Généralisation à des mélanges binaires de la méthode du second gradient et application à la simulation numérique
directe de l’ébullition nuclée. Ph.D. thesis, Université Paris 6 (2003).
[26] E. Godlewski and N. Seguin, The Riemann problem for a simple model of phase transition. Commun. Math. Sci. 4 (2006)
227–247.
[27] H. Gouin, Utilization of the second gradient theory in continuum mechanics to study the motion and thermodynamics of
liquid-vapor interfaces. Physicochemical Hydrodynamics – Interfacial Phenomena B 174 (1987) 667–682.
[28] W. Greiner, L. Neise and H. Stöcker, Thermodynamics and statistical mechanics. Springer (1997).
[29] Ph. Helluy, Quelques exemples de méthodes numériques récentes pour le calcul des écoulements multiphasiques. Mémoire
d’habilitation à diriger des recherches (2005).
[30] Ph. Helluy and H. Mathis, Pressure laws and fast Legendre transform. Math. Models Methods Appl. Sci. to appear.
[31] Ph. Helluy and N. Seguin, Relaxation models of phase transition flows. ESAIM: M2AN 40 (2006) 331–352.
[32] J.-B. Hiriart-Urruty and C. Lemaréchal, Fundamentals of convex analysis. Grundlehren Text Editions, Springer-Verlag, Berlin
(2001).
[33] D. Jamet, O. Lebaigue, N. Coutris and J.-M. Delhaye, The second gradient method for the direct numerical simulation of
liquid-vapor flows with phase change. J. Comput. Phys. 169 (2001) 624–651.
[34] S. Jaouen, Étude mathématique et numérique de stabilité pour des modèles hydrodynamiques avec transition de phase. Ph.D.
thesis, Université Paris 6, France (2001).
[35] S. Jin and C.D. Levermore, Numerical schemes for hyperbolic conservation laws with stiff relaxation terms. J. Comput. Phys.
126 (1996) 449–467.
[36] S. Kokh, Aspects numériques et théoriques de la modélisation des écoulements diphasiques compressibles par des méthodes de
capture d’interfaces. Ph.D. thesis, Université Paris 6 (2001).
[37] D.J. Korteweg, Sur la forme que prennent les équations des mouvements des fluides si l’on tient compte des forces capillaires
par des variations de densité. Arch. Néer. Sci. Exactes Sér. II 6 (1901) 1–24.
[38] P.G. LeFloch, Hyperbolic systems of conservation laws. Birkhäuser Verlag, Basel (2002).
[39] O. Le Métayer, J. Massoni and R. Saurel, Elaborating equations of state of a liquid and its vapor for two-phase flow models.
Int. J. Thermal Sci. 43 (2004) 265–276.
[40] O. Le Métayer, J. Massoni and R. Saurel, Modelling evaporation fronts with reactive Riemann solvers. J. Comput. Phys. 205
(2005) 567–610.
[41] E.W. Lemmon, M.O. McLinden and D.G. Friend, Thermophysical properties of fluid systems, in WebBook de Chimie NIST,
Base de Données Standard de Référence NIST Numéro 69, National Institute of Standards and Technology, edited by P.J.
Linstrom and W.G. Mallard. Gaithersburg MD, 20899, [Link]
[42] R.J. LeVeque, Finite Volume methods for hyperbolic problems. Cambridge University Press, Cambridge. Appl. Math. (2002).
[43] T.P. Liu, Hyperbolic conservation laws with relaxation. Commun. Math. Phys. 108 (1987) 153–175.
[44] T. Matolcsi, On the classification of phase transitions. Z. Angew. Math. Phys. 47 (1996) 837–857.
[45] R. Menikoff and B. Plohr, The Riemann problem for fluid flow of real materials. Rev. Mod. Phys. 61 (1989) 75–130.
[46] S. Nukiyama, The maximum and minimum values of the heat Q transmitted from metal to boiling water under atmospheric
pressure. Int. J. Heat Mass Transfer 9 (1966) 1419–1433. (English translation of the original paper published in J. Jpn Soc.
Mech. Eng. 37 (1934) 367–374).
[47] F. Petitpas, E. Franquet, R. Saurel and O. Le Métayer, A relaxation-projection method for compressible flows. II. Artificial
heat exchanges for multiphase shocks. J. Comput. Phys. 225 (2007) 2214–2248.
[48] P. Ruyer, Modèle de champ de phase pour l’étude de l’ébullition. Ph.D. thesis, École Polytechnique (2006). [Link].
[Link]/Theses/Files/[Link].
[49] R. Saurel, J.-P. Cocchi and P.-B. Butlers, Numerical study of cavitation in the wake of a hypervelocity underwater projectile.
J. Propuls. Power 15 (1999) 513–522.
[50] R. Saurel, F. Petitpas and R. Abgrall, Modelling phase transition in metastable liquids: application to cavitating and flashing
flows. J. Fluid Mech. 607 (2008) 313–350.
[51] M. Shearer, Admissibility criteria for shock wave solutions of a system of conservation laws of mixed type. Proc. R. Soc.
Edinb. 93 (1983) 133–244.
[52] M. Slemrod, Admissibility criteria for propagating phase boundaries in a van der Waals fluid. Arch. Rational Mech. Anal. 81
(1983) 301–315.
[53] L. Truskinovsky, Kinks versus shocks, in Shock induced transitions and phase structures in general media, edited by R. Fosdick
et al. Springer Verlag, Berlin (1991).
[54] P. Van Carey, Liquid-vapor phase-change phenomena. Taylor and Francis (1992).
[55] A. Voß, Exact Riemann solution for the Euler equations with nonconvex and nonsmooth equation of State. Ph.D. thesis,
RWTH-Aachen (2004). [Link]