On A Model of The Dynamics of Liquid/Vapor Phase Transitions

SIAM J. APPL. MATH.
c 2000 Society for Industrial and Applied Mathematics

Vol. 60, No. 4, pp. 1270–1301
ON A MODEL OF THE DYNAMICS OF LIQUID/VAPOR PHASE

TRANSITIONS∗
Downloaded 12/31/12 to 129.173.72.87. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
HAITAO FAN†
Abstract. A model for the liquid/vapor phase transitions in a shock tube is studied. Mathe-
matical analysis for the one-dimensional isothermal case is carried out. A sufficient condition for the
existence of traveling waves is given. These traveling waves represent liquefaction and evaporation
shocks. The nonexistence of some of these traveling waves when the shock speed is smaller than
some number is proved. Major one-dimensional wave patterns observed in actual experiments with
retrograde fluids are also observed in solutions of Riemann problems. A rough estimate of the dif-
ference of the calculated pressure in the solution of the Riemann problem and that in experimental
data due to the modeling is given.
Key words. liquid/vapor phase transition, traveling wave, Riemann problem
AMS subject classifications. 35K57, 35L65, 35L67, 35O30, 76N10, 76T05
PII. S0036139998343204
1. Introduction. Consider a pure fluid that exhibits liquid/vapor phase changes.

The pressure function p(ρ, T ) for such a fluid with fixed temperature below critical
temperature has the shape depicted in Figure 1.1, where ρ is the density. The re-
gion ρ < α corresponds to vapor, while ρ > β corresponds to liquid. The line joining
(m, p(m, T )) and (M, p(M, T )) is called the Maxwell line where two equilibrium phases
can coexist. The regions m < ρ < α and β < ρ < M are called metastable regions.
The spinodal region ρ ∈ [α, β] is a highly unstable region where the fluid, if it can
enter this region, will quickly decompose into vapor or liquid or their mixture. Thus,
the pressure curve p(ρ, T ) in this region cannot, at least for now, be measured.
In this paper, we shall concentrate on the dynamic flows involving liquid/vapor
phase transitions of retrograde fluids, i.e., fluids with large heat capacities, in shock
tubes. There were numerous experimental and computational studies of dynamics of
phase transitions in shock tubes. Works on condensation in expansion waves in shock
tubes, using regular fluids with inert carrier gases, were initiated and developed by
Wegener et al. [WLe, WLu, WW] and notable contributions were made by Sisilian
and Glass [SG], Wu [Wu], and Bauschdorff [Ba]. Many of these papers contained
models describing the conservation of mass, momentum, and energy coupled with
some complicated droplet growth models that govern the phase changes. These com-
plicated droplet initiation and growth equations tend to be difficult when subjected to
mathematical analysis. Experimental and computational studies on phase transitions
in shock tubes using retrograde (i.e., with large heat capacity) fluids were carried out
by authors of [GTC, TCK, TCMKS], where they observed a number of interesting
wave patterns. For details of the definition of retrograde fluids, see [TCMKS].
As shocks or rarefaction waves pass through fluid in the shock tube, phase changes
are induced. For example, as a compression shock passes through saturated vapor of
retrograde fluid, both pressure and temperature will rise. Due to the large heat
capacity, the temperature rise is relatively small and so is the equilibrium pressure
∗ Received by the editors August 10, 1998; accepted for publication (in revised form) July 2, 1999;
published electronically March 23, 2000. This research was supported by NSF grant DMS-9705723.
http://www.siam.org/journals/siap/60-4/34320.html
† Department of Mathematics, Georgetown University, Washington, DC 20057 (fan@math.
georgetown.edu).
1270
DYNAMICS OF LIQUID/VAPOR PHASE TRANSITIONS 1271
Fig. 1.1.
pe = p(m(T ), T ). Then the pressure after the shock is greater than pe = p(m(T ), T )
and condensation will occur. Condensation induced by shocks tends to produce fine
liquid drops well mixed with the remaining vapor. The passage of the waves is much
faster than the process in which these fine liquid drops collide to become large drops.
For this reason we shall assume in this paper, for simplicity, that vapor and liquid are
always finely mixed. We further assume that the flow is homogeneous. Similar to the
systems for multiphase reactive flows, we can derive a model describing such flow in
the one-dimensional case as
ρt + (ρu)x = 0,
(ρu)t + (ρu2 + p(ρ, λ, T ))x = uxx ,
(1.1)
Et + (ρuE + up − uux − κTx )x = 0,
(λρ)t + (λρu)x = w1 + µλxx ,
where ρ is the density, u the velocity, T the temperature, E the energy per volume,
λ the mass density fraction of vapor in the fluid, p the pressure, and , κ, and µ
are positive constants representing viscosity, heat conduction, and diffusion of vapor
in fluid, respectively. The function w1 is the reaction rate function for vapor phase
governing the growth of vapor in the fluid. It typically contains two parts, one for the
creation of nuclei of a new phase and the other for the subsequent growth of these
nuclei:
(1.2) w1 = wgrowth + wnucleation .
The function w1 can be very complicated, and there are still several competing theories
about these functions; see [KG] and [Sp] for a review on this subject. The modeling
of evaporation process is even more complicated; see [Wa]. We shall discuss wgrowth
and the pressure function p(ρ, λ) in section 2.
The derivation of (1.1) and (1.2) necessarily involves many assumptions. For
system (1.1) and (1.2) to be a valid model for the dynamics of liquid/vapor phase
1272 HAITAO FAN
(a)
(b)
Fig. 1.2. Conceptual representation of one-dimensional wave splitting in a liquid-vapor sys-

tem. All waves are traveling to the right. Direction of the impulsive piston motion is indicated by
broad-banded arrows, fluid motion by the thin black arrows. (a) Rarefaction wave leading to liquid-
evaporation splitting. FW = forerunner wave, EW = evaporation wave. (b) Compression wave
leading to vapor-condensation splitting. FS = forerunner shock, CD = condensation discontinuity.
transitions, it should at least exhibit wave patterns observed in actual experiments.

It would be even better if it offers a hope that qualitative and quantitative agreement
between the model and experiments can be achieved by more accurate functions for
p and w1 , etc., or by introducing even more variables if necessary.
In this paper, we shall investigate whether system (1.1) and (1.2) can exhibit
wave patterns observed in shock tube experiments on retrograde fluids. We shall do
so through the study of the traveling waves and solutions of Riemann problems of
system (1.1) and (1.2).
We recall major one-dimensional wave patterns observed in shock tube experi-
ments on retrograde fluids summarized in [TCK] and [TCMKS]. The apparatus is a
tube with a piston on one side, say the left, and the other side is either open or closed.
By compressing or withdrawing the piston, phase transition can be induced. Figure
1.2 is taken from [TCMKS] which summarizes the wave patterns observed when the
piston is compressed and withdrawn.
(a) Withdrawing the piston from liquid to send an expansion wave into the liquid.
This wave then splits into a forerunner rarefaction wave, which sends the liquid into
the metastable region, followed by a slower moving evaporation shock; see Figure
1.2(a). The upstream state of the evaporation shock is overheated liquid.
(b) Compressing the piston into the vapor to send a compression shock into the
vapor; see Figure 1.2(b). If the shock strength is not too strong, then this shock will
split into a forerunner shock and a slower moving condensation discontinuity across
which vapor changes to liquid/vapor mixture or liquid. The slower moving conden-
sation discontinuity is subsonic on both sides of the shock. By further increasing
the shock strength, the condensation discontinuity will move faster, and eventually
the condensation discontinuity and the forerunner shock merge if the compression
strength is strong enough.
When the upstream state is an equilibrium liquid/vapor mixture, then there is

no wave splitting.
(c) Withdrawing the piston from the equilibrium liquid/vapor mixture will create
a rarefaction shock. The downstream state of the shock is overheated vapor. In fact,
given rarefaction initial data, the liquid/vapor system will sharpen up to form a rar-
efaction shock across which phase transition occurs. This phenomenon was mentioned
in [TCK, TCMKS] although no experiment was carried out to demonstrate it.
Much of the mathematical analysis on systems similar to (1.1) has been mostly
restricted to the case of premixed ideal gases with Arrhenius-type reactions where the
reaction rate functions are typically of the form w = kρλ exp(−E/RT ), where k > 0
and E > 0 are constants. These choices of pressure and reaction rate functions are
used in combustion in ideal gases. There are few works on (1.1) with other types
of pressure functions and rate functions which are suitable for dynamics of phase
transitions. Rabie, Fowles, and Fickett [RFF] studied a model of dynamics of phase
transitions similar to (1.1), but without the diffusion term in the reaction equation.
They used a simple rate law
∂λ λ − λe
(1.3a) =− χ(p ≥ pig ),
∂t γ
where

0 if ρ > β,
ρ−α
(1.3b) λe = β−α if β ≥ ρ ≥ α,
1 if ρ < α
and χ is the characteristic function, γ is the typical reaction time, and pig the ignition
pressure. The part χ(p ≥ pig ) is a technical assumption to make traveling waves
possible when upstream is metastable liquid. They studied the phase plane for the
traveling wave equations of (1.1) with (1.3). They also studied the piston problem
with initial values in the metastable liquid region. They found a double wave structure
in the solutions. However, the choice of the reaction rate equation (1.3), while it is
good for studying explosive evaporation, has many limitations: the ignition pressure
pig needs artificial adjustments for different Riemann data. For general initial data,
pig will be difficult to choose. Under the rate law (1.3), metastable states will settle
into equilibrium exponentially fast which defies the fact that the metastable state can
exist for a relatively long period of time. Also, the rate law precludes the creation of
metastable states. The investigation of the effect of nucleation is not possible under
(1.3) since no nuclei are required to start the phase change in (1.3). The absence of
the diffusion term µλxx removes the possibility of moving phase boundary induced by
the close neighboring nuclei of another phase.
The dynamics of liquid/vapor phase transitions has also been studied through
the investigation of the p-system of conservation laws with the van der Waals type
of pressure functions, with which the system is of hyperbolic-elliptic mixed type; see
[Sl, Sh, Hs, Fan1]. We shall see that some wave patterns similar to those observed
in these studies are also observed in (1.1). However, the system (1.1) is closer to
actual experiments; for example, it allows the mixture of liquid and vapor. Moreover,
system (1.1), with assumption (1.5) below, is of hyperbolic type which is easier to
handle mathematically than hyperbolic-elliptic mixed type systems of conservation
laws. The case where the pressure function is that of equilibrium is studied in [MP],
which also found the splitting of the shock in the compression case.
1274 HAITAO FAN
Recently, Fan [Fan2] studied the traveling waves and Riemann problems for the
isothermal case (i.e., temperature=constant) of (1.1) in Lagrange coordinates:
vt − ux = 0,
ut + p(v, λ)x = uxx ,
(1.4)
1
λt = w(v, λ) + µλxx ,
γ
where v is the specific volume. Constant temperature can be achieved either when
the shock tube is put into a heat bath with very large heat conductivity or when the
heat capacity of the fluids goes to infinity. The latter approximates retrograde fluids,
whose heat capacities are very large. The pressure p(v, λ) is assumed to satisfy
(1.5) pv < 0, pλ > 0, and pvv > 0.
When the right-hand side of (1.4) is 0, the system (1.4) is a strictly hyperbolic system
of conservation laws. In the hope of making the system (1.1) mathematically simple
enough but still preserving the basic wave patterns of the flow observed in experiments
and allowing a fair amount of mathematical analysis as well as exhibiting interesting
phenomena, we took
pe − p
(1.6) wgrowth = λ(λ − 1)
γpe
in [Fan2]. The constant γ > 0 represents the typical reaction time. More details
about our choice of (1.6) are given in section 2.
We note that the wnucleation in (1.2) is not equal to 0 when p = pe , which
does not allow equilibrium points for traveling wave equations. This is the typical
“cold boundary” difficulty. For fluids in the metastable region with pressure not far
from equilibrium, it takes some time for these terms to have a sizable effect while the
passage of fluid dynamic waves is relatively faster. In other words, the term wnucleation
is relatively small in metastable regions. Thus, we took in [Fan2]
(1.7) w1 (ρ, λ) = wgrowth
in our studies of traveling waves in the isothermal case.
Traveling waves of (1.4) are solutions of
− cv − u = 0,
− cu + p = u ,
(1.8)
− cλ = aw(v, λ) + bλ ,
(v, u, λ)(±∞) = (v± , u± , λ± ), (v , u , λ )(±∞) = (0, 0, 0),
where a = /γ, b = µ/, and “ ” denotes d/dξ with ξ = (x − ct)/. The system (1.8)
can be rewritten as
− cv = c2 (v − v− ) + p − p− ,
(1.9) − cλ = aw(v, λ) + bλ ,
(v, λ)(±∞) = (v± , λ± ), (v , λ )(±∞) = (0, 0).
Clearly, the Rankine–Hugoniot condition
− c(u+ − u− ) − (v+ − v− ) = 0,
(1.10) − c(v+ − v− ) + (p+ − p− ) = 0,
w(v± , λ± ) = 0, λ± ∈ [0, 1]
is necessary for (1.8) and hence (1.9) to have solutions. From above, the speed of a
traveling wave is
p+ − p −
(1.11) c2 = − .
v+ − v−
In [Fan2], we proved the following results.

Theorem 1.1. Assume λ± = 0 or 1,
(1.12a) (v+ − v− )(λ+ − λ− ) > 0,
(1.12b) (v+ − v− )(p+ − pe ) ≥ 0,
and that there is no other equilibrium point of (1.9) with v value between v− and v+ .
If the speed

p+ − p−
c= −
v+ − v−
satisfies
(1.13) c2 > 4ab max(p(v+ , 1) − pe , p(v− , 1) − pe ),
then there is a monotone solution of (1.9). Furthermore, if c2 < 4ab(p+ − pe ), then

there is no solution of (1.9) satisfying λ ∈ [0, 1].
We note that by the definition of λ, the range of λ must be included in [0, 1] to
be physically meaningful. Although Theorem 1.1 is stated for the case c ≥ 0, it can
be easily converted to the case c < 0 by a transformation ξ → −ξ in (1.4), or by
interchanging (v− , λ− ) and (v+ , λ+ ) in Theorem 1.1. In [Fan2], we also studied the
admissibility of traveling waves and solutions of some Riemann problems. We showed
that system (1.4) with (1.5) and (1.7) can exhibit many wave patterns observed in
actual experiment on retrograde fluids in shock tubes.
The method for establishing the existence part of Theorem 1.1 is to prove the
existence of solutions of a singularly perturbed (1.9)
− cv = c2 (v − v− ) + p − p− + ηv ,
(1.9 ) − cλ = aw(v, λ) + bλ ,
(v, λ)(±∞) = (v± , λ± ), (v , λ )(±∞) = (0, 0)
on finite intervals by using Leray–Schauder degree theory, and then to extend the
existence result to R. Finally, we let η → 0+ to obtain the existence of solutions of
(1.9).
However, the above result on the existence of traveling waves of (1.4) does not
cover the important case where the upstream state of a shock is an equilibrium mixture
of liquid and vapor. This case is important because many actual experiments involve
equilibrium mixture. Also, condition (1.13) is too restrictive. In this paper, we shall
improve the earlier result Theorem 1.1 to relax the condition (1.13) and to allow the
upstream state of the shock to be an equilibrium mixture.
Our results on the existence of traveling waves are roughly as follows: if the
slope of the line connecting (v− , p(v− , λ− )) and (v+ , p(v+ , λ+ )) is greater than some
number, then there is a solution of (1.9) with speed c > 0.
1276 HAITAO FAN
(a) (b)
Fig. 1.3.
Precise statements of the main results of this paper are as follows.

Theorem 1.2. (i) Assume
(1.14a) (v+ − v− )(λ+ − λ− ) > 0,
(1.14b) (v+ − v− )(p+ − pe ) ≥ 0,
(1.14c) c2 + pv± < 0,
If the speed c of traveling waves satisfies

(1.15) c ≥ ω ∗ := 2 ab|p− − pe |,
then there is a traveling wave solution of (1.9). If the speed c satisfies

(1.16) c < 2 ab|p+ − pe |,
then there is no traveling wave of (1.9) satisfying λ ∈ [0, 1].

(ii) Assume
(1.14a ) (v+ − v− )(λ+ − λ− ) > 0,
(1.14b ) (v+ − v− )(p− − pe ) ≤ 0,
(1.14c ) c2 + pv± < 0,
If

(1.15 ) −c ≥ ω ∗ := 2 ab|p+ − pe |,

(1.16 ) −c < 2 ab|p− − pe |,

Corollary 1.3. (i) Let λ− = 0 or 1. For each (v− , λ− ), if ab > 0 are small
enough, then there are (v+ , λ+ ) such that (v± , λ± ) satisfy (1.14) and there are solu-
tions to the traveling wave equation (1.9) with speed c > 0 for such (v± , λ± ).
(ii) Let λ+ = 0 or 1. For each (v+ , λ+ ), if ab > 0 are small enough, then there
are (v− , λ− ) such that (v± , λ± ) satisfy (1.14) for which there are solutions to the
traveling wave equation (1.9) with c < 0.
From the above results, √ we see that the lower bound of the speed of a phase
boundary is proportional to ab, where ab = µ/γ is the product of diffusion and
typical reaction speed. This confirms the intuition that the faster the reaction speed,
the faster the phase boundary moves; and that the larger the diffusion, the more
nuclei of one phase are spread into the fluid of another phase and hence the faster the
phase transition occurs and hence the faster the phase boundary propagates.
Although some solutions of (1.4) in the → 0+ limit found through Rankine–
Hugoniot diagrams have been known for years, solutions obtained in this way are
nonunique. The result above shows that some of the phase boundaries are not ad-
missible and hence helps in resolving the nonuniqueness. Our method for proving the
existence part of Theorem 1.2 is different from what we used in [Fan2]. Here we first
modify (1.9) so that it becomes the traveling wave equations for a reaction-diffusion
system of partial differential equations. We find that this system is of monotone
type. Traveling waves for monotone systems of reaction-diffusion equations are well
studied. But our system has an extra complexity: that is, the source term in the
reaction-diffusion system also depends on the speed of the traveling wave.
In this paper, we also study Riemann problems which simulate shock tube exper-
iments. We find almost all the one-dimensional wave patterns in shock tube experi-
ments on retrograde fluids listed in [TCK] and [TCMKS] are present in solutions of
Riemann problems of the system (1.4). Thus, system (1.4) and hence (1.1) can model
pattern of waves of liquid/vapor phase transitions in retrograde fluids in shock tubes.
Although we have not carried out a computation based on real data from experi-
ments to access the quantitative agreement between the model and experiments, we
will provide a very rough estimate of the order of magnitude of error in this paper. We
shall see that if both computation and experiment show a wave pattern as depicted
in Figure 1.2(b), the agreement between the calculated pressure and that observed in
experiments can be quite good and the accuracy only depends on that of the state
functions such as pressure and energy functions, etc.
This paper is divided as follows. In section 2, we recall from [Fan2] some details
of the model (1.1), (1.6), and the pressure function. The role of the diffusion term
µλxx in (1.1) is discussed. In section 3, we prove our main result, Theorem 1.2 and
Corollary 1.3. In section 4, we give examples of Riemann problems of (1.4) whose
solutions exhibit wave pattern (a)–(c) listed in section 1. A discussion of possible
differences of computation and experiments due to the model is given. This suggests
that the model can be qualitatively correct at least when calculating the pressure at
three points, the right, middle, and left, as depicted in Figure 1.2(b).
2. A model for the dynamics of phase transitions in liquid/vapor sys-

tems. We start with the pressure function for the liquid/vapor system. We can
observe from experiments the pressure function when the material is in pure vapor
phase and liquid phase. Let p1 (ρ, T ) and p2 (ρ, T ) be the pressure function of pure
vapor phase and liquid phase, respectively, where ρ is the density and T the tem-
perature; see Figure 2.1. The Maxwell equilibrium density m(T ), M (T ) at which
1278 HAITAO FAN
Fig. 2.1.
Fig. 2.2.
p1 (m, T ) = p2 (M, T ) and the liquid and vapor phases can coexist in equilibrium can
be measured in the laboratory. The spinodal region α < ρ < β is highly unstable
where the material instantly decomposes into liquid or vapor or their mixtures. As
a result, the portion of the traditional van der Waals pressure curve within the spin-
odal region, shown as a dashed line in Figure 2.1, cannot be measured in experiments
and is probably artificial. Thus, we shall keep only the observable part and proceed
without specifying p(ρ) for ρ in the spinodal region. We extend each component of
function p(ρ, T ) continuously to an increasing function on R, as shown in Figure 2.2,
to obtain p1 (ρ, T ) and p2 (ρ, T ). In other words, the functions p1 (ρ, T ) and p2 (ρ, T )
are pressure functions for pure vapor and liquid states, respectively. Although the
part of these pressure functions beyond the spinodal limit, i.e., p1 (ρ1 ) for ρ1 > α and
p2 (ρ2 ) for ρ2 < β, cannot be measured and our extensions of the pressure functions
into the spinodal region are somewhat arbitrary, these parts of the pressure functions
do not affect the outcome; for example, the part of the pressure function p1 (ρ) for
ρ > α(T ) does not have much effect since there will not be any vapor there. We shall
see more about this later.
We shall treat liquid and vapor as different species and specify the transition of
phases by a reaction rate equation. We introduce the following notation.

Quantities ai denote the quantity for vapor when i = 1 and that of liquid when
i = 2;
fi (x, t) is the volume fraction of ith species at point x and time t;

ρi (x, t) is the density of ith species at point x and time t;
wi is the reaction rate function denoting the mass of ith species produced per
unit volume per time unit. By the conservation of mass, we have w1 + w2 = 0.
Modifying the systems of equations describing reacting two-phase flows, we derive
the following balance laws to model the dynamics of phase transitions:
ρt + ∇ · (ρu) = 0,

ρi fi
(ρ1 f1 )t + ∇ · (ρ1 f1 u) = w1 + ∇ · µii ∇ ,
(2.1) ρ
(ρu)t + ∇ · (ρ(uu) + P) = 0,
Et + ∇ · (uE + u · P + q) = 0,
where u is the average velocity, E the energy per volume, P the stress tensor, q the
heat flux vector, and µ the diffusion coefficients.
We make the following assumptions which simplify the modeling.
(A1) Vapor and liquid fill up the space, i.e.,
(2.2) f1 + f2 = 1,
(2.3) ρ = ρ1 f1 + ρ2 f2 .
(A2) The temperatures of liquid and vapor components at each point are equal.
(A3) The pressure in the portion occupied by ith component is pi (ρi , T ). Different
components at the same location have the same pressure:
(2.4) p = p1 (ρ1 , T ) = p2 (ρ2 , T ).
(A4) The stress tensor is

2
(2.5) P= p+ 1 − 2 (∇ · u) I − 1 (∇u + ∇uT )
3
and the heat conduction q is given by the Fourier law q = −κ∇T , where constants
1 , 2 > 0, κ > 0 are viscosity and heat conductivity coefficients, respectively.
Under these assumptions, the system describing the dynamics of phase changes
is
ρt + ∇ · (ρu) = 0,

ρi fi
(ρ1 f1 )t + ∇ · (ρ1 f1 u) = w1 + ∇ · µii ∇ ,
ρ
(2.6) (ρu)t + ∇ · (ρ(uu) + P) = 0,
Et + ∇ · (uE + u · P + q) = 0,
p = p1 (ρ1 , T ) = p2 (ρ2 , T ),
ρ = ρ1 f1 + ρ2 f2 .
Once the relations E = E(ρ, ρ1 f1 , T ) and w1 = w1 (ρ, ρ1 f1 , T ) are known, the system
(2.6) forms a closed system of equations. In the one-dimensional case, this system is
1280 HAITAO FAN
reduced to
ρt + (ρu)x = 0,
(ρu)t + (ρu2 + p)x = uxx ,

Et + (ρuE + up − uux − κTx )x = 0,

(2.7) ρ1 f1
(ρ1 f1 )t + (ρ1 f1 u)x = w1 + µ ,
ρ xx
p = p1 (ρ1 , T ) = p2 (ρ2 , T ),
ρ = ρ1 f1 + ρ2 f2 .
In Lagrange coordinates, the corresponding system of equations is
vt − ux = 0,
u
x
u t + px = ,
v x
uu Tx
x
et + (up)x = +κ ,
(2.8) v x v x

λx
λt = w(λ, v, T ) + µ ,
v x
p = p1 (v1 , T ) = p2 (v2 , T ),
v = λv1 + (1 − λ)v2 ,
where λ = ρ1 f1 /ρ is the mass fraction of vapor in liquid/vapor mixture and e =

e(λ, v, T ) is the specific energy and w = w1 /ρ. For the isothermal case, i.e., when
the temperature is constant, which approximates the case where heat capacity is very
large, the above system becomes
vt − ux = 0,
u
x
u t + px = ,
v x
(2.9) λx
λt = w(λ, v) + µ ,
v x
p = p1 (v1 ) = p2 (v2 ),
v = λv1 + (1 − λ)v2 .
∂p1 ∂p2
From the last two equations in (2.9), we find that if ∂v , ∂v < 0, then
∂p p1 (v1 )p2 (v2 )

(2.10) = <0
∂v λp2 (v2 ) + (1 − λ)p1 (v1 )
and
∂p p (v1 )p2 (v2 )(v2 − v1 )
(2.11) = 1 > 0.
∂λ λp2 (v2 ) + (1 − λ)p1 (v1 )
Further calculation shows that

∂2p ∂ 2 pj
(2.12) > 0 if > 0, j = 1, 2.
∂v 2 ∂v 2
The production of vapor (or liquid) phase is through the initiation or annihilation
of nuclei of vapor (or liquid) and the subsequent growth of such nuclei when these
nuclei and elements of the metastable old phase meet. Correspondingly, the produc-
tion rate function of the first species, w, contains two parts: the rate of the creation
of nuclei of vapor phase, winitiation , and that of the growth of such nuclei, wgrowth .
Similar to the theory of chemical kinetics and thermochemistry [see ZBLM], the part
governing the growth of nuclei of vapor phase can be modeled by
wgrowth = kλa (1 − λ)b ,
where a and b are positive constants. In this paper, we shall take
wgrowth = kλ(1 − λ).
Here λ(1 − λ) is the probability for particles of liquid and vapor to collide to make
reactions possible and k is known as the rate constant. In the theory of thermochem-
istry, the rate constant is a function of temperature given by Arrhenius’s law. During
liquid/vapor phase transitions, however, the temperature changes are usually minor,
but the rate of phase change varies significantly according to |p − pe |. Thus, the ex-
pression of the rate constant k should be adjusted according to the following facts:
When pe (T ) − p(λ, v, T ) > 0, where pe (T ) is the equilibrium pressure at temperature
T , liquid tends to evaporate and hence wgrowth ≥ 0 and the larger pe − p, the faster
the evaporation. Similarly, when pe (T ) − p(λ, v, T ) < 0, vapor tends to condense into
liquid and hence wgrowth ≤ 0 and the larger |pe − p|, the faster the condensation.
Consistent with these facts, we shall model the growth of nuclei of vapor by
pe − p
(2.13) wgrowth = λ(λ − 1),
γpe
where γ is taken as a constant.
The function winitiation is neglected in our study of traveling waves and Riemann
problems. Thus, we do not discuss it here. For more information on this function,
see the reviews [Ox, Sp] and references cited therein.
Remark. Now we return to our earlier discussion on the arbitrariness of the
extension of the pressure functions beyond the spinodal point. We claim that the
extension part of the pressure functions does not have much effect in our system. For
example, let’s consider the part of p1 (ρ1 ) in ρ1 > α. In this range, the rate function
w << −1 has very large absolute value which forces the fraction of vapor λ to be
virtually zero which implies that ρ2 = ρ > β and hence the pressure p = p2 (ρ) is well
defined and is not sensitive to the extended part of function p1 for ρ1 > α. Thus the
claim holds.
Remark. The diffusion term λxx has a significant role in system (2.9) and hence
(2.6). The term (p − pe )λ(λ − 1)/γ/pe represents the rate of reaction for liquid
and vapor particles at the same location (x, t) only. It does not cover the reaction
among neighboring particles. The diffusion term µλxx then determines the chance for
neighboring particles to collide. The necessity of this diffusion term can be illustrated
by the following example. Let u = 0, T = constant,

0, x < 0,
λ=
1, x ≥ 0,
and v be such that pressure is a constant and the vapor is supersaturated at initial
time. By Maxwell’s rule, which states that the only stationary phase boundary hap-
pens at the equilibrium pressure, vapor should start to condense at the liquid/vapor
1282 HAITAO FAN
interface which results in the motion of the phase boundary. However, the system
(2.9) with µ = 0 says that the solution is the initial data and hence no phase change
happens. In fact, we can see easily that when µ = 0 and the pressure is not the
equilibrium pressure, the liquid/vapor interface in (2.9) with µ > 0 must move, which
is in agreement with Maxwell’s rule. In most traditional modeling of phase transitions
in shock tube, droplet growth formulae are used which already include the effect of
vapor near a droplet of liquid colliding with the droplet, resulting in the growth of
the droplet. With the use of those droplet growth formulae, the diffusion term λxx is
redundant. However, those droplet growth laws are usually difficult to analyze. Since
we choose the simple form (2.13) for the rate function, the diffusion term is needed.
3. The existence of traveling waves. In this section, we shall provide suf-
ficient conditions for the existence of traveling waves for (1.4) with (1.5), (1.6), and
(1.7).
We modify the system of equations for traveling waves (1.9) as
− cv = c2 (v − v− ) + p − p− + ηv ,
(3.1) − cλ = aw(v, λ) + bλ ,
(v, λ)(−∞) = (v− , λ− ), (v, λ)(∞) = (v+ , λ+ ).
We intend to let the small constant η > 0 go to 0 after we establish the existence of
solutions of (3.1). We see that there is a one-to-one correspondence between solutions
of (3.1) with speeds c > 0 and those with negative speeds through the transformation
ξ → −ξ. For this reason, we shall assume c ≥ 0 in this section.
The system (3.1) ais the same system of equations of traveling waves with speeds
c of the system
vt = ηvxx + c2 (v − v− ) + p − p− ,
(3.2)
λt = bλxx + a(p − pe )λ(λ − 1),
or
(3.2 ) Ut = AUxx + F (U, c)
for short, where U = (v, λ)T , A = diag(η, b), and

c2 (v − v− ) + p − p−
(3.3) F (U, c) = .
a(p − pe )λ(λ − 1)
The Jacobian of F is

c2 + pv pλ
(3.4) ∇F = .
apv λ(λ − 1) a(p − pe )(2λ − 1) + apλ λ(λ − 1)
The off-diagonal elements of ∇F are positive for λ ∈ [0, 1]. We recall that a system
of reaction-diffusion equations
(3.5) Ut = AUxx + G(U )
is called monotone if all off-diagonal elements of ∇G are positive. Thus, the system
(3.2) is a monotone system of parabolic partial differential equations for λ ∈ [0, 1].
Now we consider the eigenvalues of ∇F at (v± , λ± ). By the Rankine–Hugoniot

condition (1.10), we have λ± = 0, 1 or p(v± , λ± ) = pe . When λ± = 0 or 1, the
eigenvalues of ∇F are
(3.6a) α1± = c2 + pv (v± , λ± )
and
(3.6b) α2± = a(p± − pe )(2λ± − 1).
At p = pe , the eigenvalues of ∇F are

(3.6c)
1 2
α1 = c + pv + apλ λ(λ − 1) + (c2 + pv + apλ λ(λ − 1))2 − 4ac2 pλ λ(λ − 1) ≥ 0
2
and
(3.6d)
1 2
α2 = c + pv + apλ λ(λ − 1) − (c2 + pv + apλ λ(λ − 1))2 − 4ac2 pλ λ(λ − 1) ≤ 0.
2
According to the theory of traveling waves of monotone system (3.5), when some of
the eigenvalues of ∇F at U− are positive, there is no traveling wave of (3.3) with
speed c ≥ 0. The eigenvalues of ∇F at U = U− are negative if and only if either
α1− = c2 + pv− < 0,

(3.7)
p− > pe , λ− = 0,
or
α1− = c2 + pv− < 0,
(3.8)
p− < pe , λ− = 1.
We are interested in the existence of traveling waves when there are no equilibrium
points (v ∗ , λ∗ ) of (1.9) with v ∗ between v− and v+ . Then, by the Rankine–Hugoniot
condition (1.10), possibilities for U+ are as follows.
Case 1. Condition (3.7) and one of the following (1a) or (1b) hold:
(1a) p+ > pe , λ+ = 1, v+ > v− , and c2 + pv+ < 0.
(1b) p+ = pe , λ+ ∈ [0, 1], v+ > v− , and c2 + pv+ < 0.
(2a) p+ < pe , λ+ = 0, v+ < v− , and c2 + pv+ < 0.
(2b) p+ = pe , λ+ ∈ [0, 1], v+ < v− , and c2 + pv+ < 0.
(3a) p+ < p− , λ+ = 0, and v+ > v− .
(3b) p+ > p− , λ+ = 0, and v+ < v− .
(4a) p+ < p− , λ+ = 1, and v+ > v− .
(4b) p+ > p− , λ+ = 1, and v+ < v− .
Cases (3b) and (4b) are not possible for traveling waves of nonnegative speed.
Cases (3a) and (4a) correspond to ordinary Lax shock without phase transitions.
Our interest is in Case 1 and Case 2, which involve phase changes. We list cases
corresponding to Case 1 and Case 2 for negative speed.
Case 1 . The conditions
(3.7 ) c2 + pv+ < 0, p+ > pe , λ+ = 0

1284 HAITAO FAN
and either (1 a) p− > pe , λ− = 1, and v− > v+ , or (1 b) p− = pe , λ− ∈ [0, 1], and
v+ < v− hold.
Case 2 . The conditions
(3.8 ) c2 + pv+ < 0, p+ < pe , λ+ = 1
and either (2 a) p− < pe , λ− = 0 and v+ > v− , or (2 b) p− = pe , λ− ∈ [0, 1], and
v+ > v− hold.
In Cases 1 and 2, eigenvalues of ∇F at U− are negative. At U+ , one eigenvalue of
∇F is nonpositive and the other is positive. In other words, U− is a stable equilibrium
of (3.3) while U+ is an unstable equilibrium. Thus, system (3.3) in Cases 1 and 2 is
monostable.
We are interested in monotone traveling waves of (3.4). We say that a vector-
valued function is monotone if every component of the vector-valued function is mono-
tone. When we say a vector q ≥ 0, we mean that each of its components is nonnegative.
An interval [U, W ], where U and W are two vectors of the same size, is the set of all
vectors q satisfying U ≤ q ≤ W .
Theorem 3.1. (see [VVV, Chapter 3, Theorem 4.3]). Let the system (3.5) be
monotone and monostable with U+ being the unstable equilibrium. Assume that a
vector q ≥ 0 exists such that
(3.9) G(rq + U+ ) ≥ 0 for 0 ≤ r ≤ r0 ,
where r0 > 0 is some number. Assume that in the interval [U+ , U− ], the vector-
valued function G(U ) vanishes only at U+ and U− . Suppose the matrix A in (3.5) is
A = diag(a1 , a2 ) with a1 , a2 > 0. Then for all
aj Uj + Gj (U )
(3.10) s ≥ ω := inf sup
U ∈K x,j −Uj
there exists a decreasing traveling wave of (3.5) with speed s ≥ 0, where

K := U ∈ C 2 (R; R2 ) | U is decreasing and lim U (x) = U± .
x→±∞
When s < ω such a solution does not exist.

By performing the transformation
Ū = −U,
Ḡ(Ū ) = −G(−Ū ).
Theorem 3.1 can be rewritten as follows to allow U+ > U− .

THEOREM 3.1 . Let the system (3.5) be monotone and monostable with U+ being
the unstable equilibrium. Assume that a vector q ≤ 0 exists such that
(3.9 ) G(rq + U+ ) ≤ 0 for 0 ≤ r ≤ r0 ,
where r0 > 0 is some number. Assume that G(U ) = 0 in the interval [U− , U+ ].
Suppose the matrix A in (3.5) is A = diag(a1 , a2 ) with a1 , a2 > 0. Then system (3.5)
has an increasing traveling wave with speed s ≥ 0 if and only if
aj Uj + Gj (U )
(3.10 ) s ≥ ω := inf sup ,
U ∈K x,j −Uj
where
K := {U ∈ C 2 (R; R2 ) | U is increasing and lim U (x) = U± }.
x→±∞
We have to bear in mind that we are looking for traveling waves of (3.3) with
speed c > 0 and where F (U, c), unlike G(U ) in (3.5), contains c. Thus, to prove (3.1)
has a solution, it is necessary and sufficient to prove that c ≥ ω(c), where ω(c) is
defined as in (3.10) with G(U ) replaced by F (U, c).
Theorem 3.2. For Cases 1 and 2, if

(3.11) c ≥ ω ∗ := 2 ab|p− − pe |,
then there is a solution of (3.1) for η > 0 small enough. For Case 1 and 2 , if

(3.11 ) −c ≥ ω ∗ := 2 ab|p+ − pe |,
then there is a solution of (3.1) for η > 0 small enough.

Proof. We consider Case 2, that is,
(3.12) c2 + pv± < 0, U+ = (v+ , λ+ ) < (v− , λ− ) = (v− , 1) = U− , p− < p+ ≤ pe ,
where λ+ ∈ [0, 1) and there is no other equilibrium point with v between v± . The
proof for Case 1 is similar.
By assumptions (1.5) and (3.12), we see that there is no equilibrium point in the
interval (U+ , U− ). Now we verify the condition (3.9). In fact, we can take the vector
q = (q1 , q2 )T > (0, 0)T so that
(c2 + pv+ )q1 pv+ q1
(3.13) 0<− < q2 < − .
pλ+ pλ+
With (3.13), we have

d
(3.14) p(rq1 + v+ , rq2 + λ+ ) = pv+ q1 + pλ+ q2 < 0,
dr r=0
which implies that p(rq +U+ ) < p+ < pe for small r > 0. Thus, the second component
of F in (3.3)
(3.15) F2 (rq + U+ , c) = a(p(rq + U+ ) − pe )(rq2 + λ+ )(rq2 + λ+ − 1) > 0
for small r > 0. Similarly, by using Rankine–Hugoniot condition (1.11), we can write
the first component of F as
F1 (rq + U+ , c) = c2 rq1 + p(rq + U+ ) − p+ .
Its derivative satisfies

dF1
= (c2 + pv+ )q1 + pλ+ q2 > 0
dr r=0
under the choice (3.13). This implies that F1 (rq + U+ , c) > 0 for small r > 0. Thus,
condition (3.9) is satisfied by F (U, c) in Case 2.
From Theorem 3.1, we know that (3.2) has a decreasing traveling wave with speed
c > 0 if and only if
aj ρ + Fj (ρ, c)
c ≥ ω(c) = inf sup .
U ∈K x,j −ρj
1286 HAITAO FAN
Since j = 1, 2 in the above, we can consider j = 1 and 2 separately. We start with

j = 2 since in this case
bλ + a(p − pe )λ(λ − 1)

(3.16) ω2 (c) := inf sup
(v,λ)∈K x −λ
does not contain c explicitly. By Lemma 3.3 below, given a function
(3.17) λ(x) ∈ K2 := {λ ∈ C 2 (R; R) | λ(x) is decreasing and lim λ(x) = λ± },

x→±∞
we can select a function v(x) = v(λ(x)) so that
(3.18) p(v(x), λ(x)) ∈ [p− , p+ ].
We note that K2 is the set of functions which are the second component of functions
in K. For such a function U (x) = (v(λ(x)), λ(x)), we have
bλ + a(p− − pe )λ(λ − 1)

(3.19a) ω2 (c) ≤ inf sup =: ω ∗ .
λ∈K2 x −λ
It is clear that ω ∗ is independent of c. Now we consider
ηv + c2 (v − v+ ) + p − p+
(3.19b) ω1 (c) := inf sup .
(v,λ)∈K x −v
Suppose that λn (x), n = 1, 2, 3, . . . is a sequence of functions in K2 satisfying
bλn + a(p− − pe )λn (λn − 1)

(3.20) ω ∗ + 1 > sup → ω∗
x −λn
as n → ∞. Now we need the following lemma whose proof will follow this one.
Lemma 3.3. Let (v± , λ± ) be of Case 2. For each of the functions λn (x), n =
1, 2, 3, . . ., in K2 satisfying (3.20), the equation
(3.21) R := c2 (v − v+ ) + p(v, λn ) − p+ = 0
uniquely defines a function vn (x) such that (vn , λn ) ∈ K. Furthermore, the functions
(vn (x), λn (x)) satisfy p(vn (x), λn (x)) ∈ [p− , p+ ] and
vn
(3.22) 0 ≤ sup ≤C
x −vn
for some constant C > 0 independent of n.
With these functions (vn , λn ), we see that
ηvn
(3.23) ω1 (c) ≤ inf sup = O(1)η.
n x −vn
It is clear that
ω(c) = max(ω1 (c), ω2 (c))
holds. With η > 0 small enough, we have from above (3.19a) and (3.23) that
(3.24) ω(c) = ω2 ≤ ω ∗ .
Therefore, if c ≥ ω ∗ , there is a solution of (3.1) for sufficiently small η > 0.

To complete
the proof of the first statement of the theorem, it remains to prove
that ω ∗ = 2 ab|p− − pe |. To this end, we consider the equation
(3.25) λt = bλxx + a(p− − pe )λ(λ − 1)
and its traveling wave equations
− cλ = bλ + a(p− − pe )λ(λ − 1),

(3.26)
λ(−∞) = 1, λ(∞) = 0.
It can be verified that (3.25) satisfies the conditions of Theorem 3.1. Then, (3.26) has
a monotone solution
if and only if c ≥ ω ∗ . We claim that (3.26) has asolution if and
only if c ≥ 2 ab|p− − pe |. It is clear that this claim implies ω ∗ = 2 ab|p− − pe | in
view of Theorem 3.1. To prove this claim, we observe that (3.26) is equivalent to
λ = q,
(3.27) bq = −cq − a(p− − pe )λ(λ − 1),
(λ, q)(−∞) = (1, 0), (λ, q)(∞) = (0, 0).
The eigenvalues of (3.27) at (λ, q) = (1, 0) are
1
α1− = [−c − c2 − 4ab(p− − pe )] < 0
2b
and
1
α2− = [−c + c2 − 4ab(p− − pe )] > 0,
2b
where we used p− − pe < 0 in Case 2. There is an unstable manifold of (3.27) issued
from (λ = 1, q = 0) into the triangle in (λ, q)-plane:

c
(3.28) D := (λ, q) ∈ R2 : 0 ≤ λ ≤ 1, − λ ≤ q ≤ 0 .
2b
By (3.27), this unstable manifold cannot cross the boundary of D at q = 0, 0 ≤ λ ≤ 1.

c
At the part of the boundary of D, q = − 2b λ, the slope of the unstable manifold can
be calculated from (3.27):
dq c a(p− − pe )λ(λ − 1)
=− −
dλ b bq
c 2a(p− − pe )(λ − 1) c
=− + −
(3.29) 2b c 2b
1 2 c
=− (c − 4ab(p− − pe )(λ − 1)) −
2bc 2b
1 2 c
≤− (c − 4ab|p− − pe |) − ,
2bc 2b
where we used p− − pe < 0 in the last step. If

c ≥ 2 ab|p− − pe |,
1288 HAITAO FAN
then
dq c
≤− ,
dλ 2b
which implies that the unstable manifold cannot cross the boundary of D. Thus, this
unstable manifold hasto enter (λ, q) = (0, 0) as ξ → ∞ and hence there is a solution
of (3.26) when c ≥ 2 ab|p− − pe |. On the other hand, the eigenvalues of (3.27) at
λ = 0 are
1
α= − c ± c2 − 4ab|p− − pe | ,
2b

which are imaginary if c < 2 ab|p− − pe |. Then the stable manifold of (3.27) at
(0, 0) is not monotone
and hence there is no monotone solution of (3.26). In view of
Theorem 3.1, ω ∗ = 2 ab|p− − pe |. This completes the proof of the first statement.
The proof for Case 1 is similar. The only difference is that we use Theorem
3.1 instead of Theorem 3.1. Performing a transform ξ → −ξ, we can convert our
statements for Cases 1 and 2 to those for Cases 1 and 2 .
Proof of Lemma 3.3. First, we prove that for each λ ∈ [λ+ , λ− = 1], the solution
v of (3.21)
(3.30) R(v, λ) := c2 (v − v+ ) + p(v, λ) − p+ = 0
in the range [v+ , v− ] is unique. Indeed, if otherwise, there would be a number λ0 ∈

[λ+ , λ− ] such that R(v, λ0 ) = 0 has two solutions v0 and v1 , with v0 < v1 . Since
(3.31) Rvv = pvv > 0,
it is necessary that
(3.32) Rv (v0 , λ0 ) < 0 and Rv (v1 , λ0 ) > 0
and that
(3.33) Rv (v, λ0 ) > Rv (v1 , λ0 ) > 0 for v− > v > v1 .
Inequality (3.33) then implies that 0 = R(v− , λ− ) > R(v− , λ0 ) > R(v1 , λ0 ) = 0 which
is a contradiction. This contradiction establishes the uniqueness of the solution of
(3.30) in the range v ∈ [v+ , v− ].
The above argument also implies that if R(v, λ) = 0 with λ ∈ [λ+ , λ− ] and
v ∈ [v+ , v− ], then at such (v, λ),
(3.34) Rv = c2 + pv < 0.
For λ(x) ∈ [λ+ , λ− ], we have that
(3.35) R(v+ , λ(x)) > R(v+ , λ+ ) = 0 = R(v− , λ− ) > R(v− , λ(x)).
Thus, there is a unique function v(x) satisfying R(v(x), λ(x)). If λ(x) is in K2 , that
is, λ ∈ C 2 (R; R), λ < 0, λ(x) → λ± as x → ±∞, then
pλ λ
(3.36) v = − < 0,
c2+ pv
where we used (3.34). Further computation shows that v ∈ C 2 (R; R) and v(x) → v±
as x → ±∞. In other words, (v, λ)(x) determined by R = 0 is in K if λ(x) is in K2 .
Observing that the solution of equation R = 0 is the intersection of a straight
line p = −c2 (v − v+ ) + p+ and the curve p = p(v, λ) in (v, p)-plane, we see that the
solution of R(v, λ(x)) = 0 satisfies (3.18),
(3.37) p(v(x), λ(x)) ∈ [p− , p+ ].
Now we consider −vn /vn for function vn satisfying (3.20). We claim ||λn ||∞ is
independent of n. Indeed, if otherwise, there is a subsequence of {λn }, still denoted
by {λn } such that ||λn ||∞ → ∞ as n → ∞. By the fact that λ(x) ∈ [λ+ , λ− ], there is
an interval [yn , xn ] such that
(3.38) lim λn (yn ) = −∞

n→∞
and
(3.39) −λn (x) ≥ −λn (xn ) = 1 for x ∈ [yn , xn ].
By the monotonicity of λ(x), it is clear that
(3.40) |yn − xn | ≤ 1/|λ− − λ+ |.
On the interval [yn , xn ], from (3.20), (3.37), and (3.39), we have
λn
(3.41) = Cn (x) ≤ C,
−λn
where C > 0 is a constant independent of n. Integrating (3.41), we obtain

xn
|λn (yn )| = |λn (xn )| exp C(x)dx ≤C
yn
which contradicts (3.38). Thus, we have
(3.42) ||λn ||∞ ≤ C.
From (3.30), we have

pλ
(3.43) vn = − λ .
c2 + pv n
From (3.34), we know that c2 + pv < 0 for (v, λ) satisfying (3.30) in the range λ ∈
[λ+ , λ− ] and v ∈ [v+ , v− ]. Then there is a constant δ0 such that
(3.44) c2 + pv < −δ0 .
Applying this and (3.42) in (3.43), we obtain
(3.45) ||vn ||∞ ≤ C.

1290 HAITAO FAN
To prove (3.22), we compute from (3.30) to get

vn 1 pλλ (c2 + pv ) λn
= p v + 2p λ + λ + p
vv n vλ n n λ
−vn c2 + pv pλ λn

pλ λn
≤C+ 2
c + pv λn

pλ bλn + a(p− − pe )λ(λ − 1) a(p− − pe )λ(λ − 1)
≤C− −
b(c2 + pv ) −λn −λn

pλ bλn + a(p− − pe )λ(λ − 1)
≤C− 2
≤ C,
b(c + pv ) −λn
where we used (3.20) and the facts that p− < pe , c2 + pv < −δ0 , and λn < 0. The
part (3.22), 0 ≤ supx (−vn /vn ), is a direct consequence of vn < 0 and vn (x) → v±
as x → ±∞.
Theorem 3.4. Assume the data (v± , λ± ) are of one of Case 1 or Case 2.
(i) If the speed c of traveling waves satisfies

(3.46) c ≥ ω ∗ := 2 ab|p− − pe |,

(3.47) c < 2 ab|p+ − pe |,
(ii) For Cases 1 and 2 , if

(3.46 ) −c ≥ ω ∗ := 2 ab|p+ − pe |,

(3.47 ) −c < 2 ab|p− − pe |,
Proof. (i) From Theorem 3.2, we see that under condition (3.46), there is a
monotone solution of (3.1) with sufficiently small η > 0. We denote this solution by
(v η , λη ). We choose arbitrarily a fixed number λ0 between λ± and not equal to λ± .
Since (3.1) is invariant under shifting, ξ → ξ + ξ0 , we can assume λη (0) = λ0 . By
the monotonicity and boundedness of (v η , λη ), there is a sequence {ηn }∞
n=1 such that
ηn → 0+ as n → ∞ and
(3.48) (v, λ)(ξ) := lim (v ηn , ληn )(ξ)
n→∞
exists. The limit (v, λ)(ξ) defined in (3.48) satisfies (1.9)1 and (1.9)2 weakly. A
weak solution of (1.9)1 and (1.9)2 is also a strong solution. It is clear that the limit
(v, λ)(ξ) is monotone and is contained in the interval [(v+ , λ+ ), (v− , λ− )] and satisfies
λ(0) = λ0 . Since the interior of the interval [(v+ , λ+ ), (v− , λ− )] does not contain
equilibrium point of (1.9), the boundary condition (1.9)3 is also satisfied by the limit
(v, λ)(ξ). Thus, (v, λ)(ξ) is a solution of (1.9).
For definiteness, we consider Case 1 in our proof for the second statement in (i).
In this case, condition (3.47) implies that p− > p+ > pe and hence λ+ = 1. The
eigenvalues of (1.9) at (v+ , λ+ ) are
c2 + pv+
α1+ = − >0
−c
and
1
α2+ = − c ± c2 − 4ab|p+ − pe | .
2b
Under condition (3.47), α2+ are complex. Thus, stable manifolds of (1.9) at (v+ , λ+ )
have a part in the region λ > 1 and hence there is no solution of (1.9) satisfying
λ ∈ [0, 1]. The proof for Case 2 is similar.
Statement (ii) can be proved similarly.
Remark. It is clear from (3.24) that ω(c) is determined by the reaction-diffusion
part of the system (1.4). In this sense, we can say that the existence of traveling
waves of (1.4) is primarily determined by (1.4)3 .
Corollary 3.5. (i) Let λ− = 0 or 1. For each (v− , λ− ), if ab > 0 are small
enough, then there are (v+ , λ+ ) such that (v± , λ± ) are of Case 1 or 2 for which there
is a solution to the traveling wave equation (1.9) with speed c > 0.
(ii) Let λ+ = 0 or 1. For each (v+ , λ+ ), if ab > 0 are small enough, then there
are (v− , λ− ) such that (v± , λ± ) are of Case 1 or 2 for which there is a solution to
the traveling wave equation (1.9) with c < 0.
Proof. (i) For definiteness, we consider only Case 1, where λ− = 0 and p− >
p+ ≥ pe . The proof for Case 2 is similar. From Theorem 3.4, we know that if
1 ≥ c−2 4ab(p− − pe ) or equivalently
v+ − v−
(3.49) 1 ≥ 4ab (p− − pe ),
p− − p+
then there is a solution of (1.9). Since p+ > pe , we have v+ ≤ M , the Maxwell point
at which p(M, 1) = pe . We can choose v+ so that p+ is close to pe . Then for small
enough ab > 0, the left-hand side of (3.49) satisfies
v+ − v−
(3.50) 4ab (p− − pe ) ≤ 4ab(M − v− ) ≤ 1
p− − p+
and hence there are solutions of (1.9) for such (v± , λ± ).

(ii) The proof is similar to that of (i).
From Theorem 3.4 or Corollary 3.5, we see that for some (v− , λ− ), there are
infinitely many (v+ , λ+ ) so that the traveling wave equation (1.9) has a solution. If
all these traveling waves are “good,” then solutions of a Riemann problem for (1.4)
in the → 0+ limit will be nonunique. We say that a traveling wave of (1.4) is
metastable if it is stable with respect to perturbations in (u, v) but not in λ. Only
stable or metastable traveling waves can be present in a solution for a sizable period
of time. Thus, only stable or metastable traveling waves of (1.4) are considered as
admissible traveling waves. The stability and admissibility of traveling waves of (1.4)
is left for future investigations.
4. Wave patterns in solutions to some Riemann problems and those
observed in experiments on retrograde fluids. In this section, we shall investi-
gate solutions to Riemann problems of (1.4) with (1.5), (1.6), and (1.7) in the → 0+
limit with scaling γ = a and µ = b. We show that system (1.4) can exhibit major
one-dimensional wave patterns of liquid/vapor phase transitions observed in shock
tube experiments for retrograde fluids, i.e., fluids with large heat capacity. We shall
also discuss the order of magnitude of the possible error between computation and
experimental data due to the modeling.
1292 HAITAO FAN
Major one-dimensional wave patterns observed in shock tube experiments on

retrograde fluids are listed in (a)–(c) in section 1. We are interested in whether these
wave patterns can be observed in solutions of the initial value problem
vt − ux = 0,
ut + p(v, λ)x = uxx ,
a
(4.1) λt = (p − pe )λ(λ − 1) + bλxx ,

(u− , v− , λ− ) if x < 0,
(u, v, λ)(x, 0) =
(u+ , v+ , λ+ ) if x > 0,
when > 0 is small. We denote the solution of (4.1) by (u , v , λ )(x, t). If there is a
sequence n , n = 1, 2, . . ., such that n → 0+ as n → ∞ and the limit
(4.2) (u, v, λ)(x, t) := lim (un , v n , λn )(x, t)

n→∞
exists almost everywhere, then the limit satisfies
vt − ux = 0,
ut + p(v, λ(x, t))x = 0,
(4.3) (p(v, λ) − pe )λ(λ − 1) = 0,

(u− , v− , λ− ) if x < 0,
(u, v, λ)(x, 0) =
(u+ , v+ , λ+ ) if x > 0.
Instead of studying (4.1) for small > 0 directly, we can study solutions of (4.3) that
are → 0+ limits of solutions of (4.1). Thus, we are interested in solutions of (4.3)
that are some type of strong limits of solutions of (4.1) as n → 0 for some sequence
{n }. Also, since only stable or metastable solutions can be observed, we allow only
stable or metastable solutions as admissible. A solution is called metastable if it is
stable with perturbations in u and v only.
Definition 4.1. If a stable or metastable solution (u, v, λ)(x, t) of (4.3) is a
strong limit of (un , v n , λn ), solutions of (4.1) with = n , for some sequence
{n }∞n=1 , n → 0+ as n → ∞, then we call (u, v, λ)(x, t) an admissible solution
of (4.3).
From (4.3)3 , piecewise smooth solutions of (4.3) consist of (a) shock waves with
two sides of the shock satisfying λ = 0, 1 or p = pe , (b) smooth waves with λ ≡ 0 or
1, and (c) smooth isobaric waves where p = pe .
(a) Shock waves. We say that a shock with end states (u± , v± , λ± ) has a shock
profile if the traveling wave equation (1.9) of (4.1) has a solution with (v, λ)(±∞) =
(v± , λ± ).
When λ+ = λ− , there is a phase change across the shock. Sufficient conditions for
the existence of shock profiles for such shocks are given in Theorem 3.4 and Corollary
3.5. The stability or metastability of these profiles is left for future investigations.
For shocks with λ+ = λ− , there is no phase transition involved. From (4.1), we
can see that it is necessary that λ− = λ+ = 0, or 1 in this case. The admissibility for
these nonreacting shocks is determined by Lax’s criterion. Nonreacting shock profiles
satisfying Lax’s criterion are metastable; see [SX].
(b) Nonreacting smooth rarefaction waves. These are the “ordinary” admissible
gas dynamic rarefaction waves with λ ≡ 0 or λ ≡ 1. These nonreacting rarefaction
waves are metastable; see [X].
(c) Isobaric waves. These are smooth solutions of (4.3) with
(4.4) p(v(x, t), λ(x, t)) = pe

for (x, t) in some regions D called isobaric regions. In isobaric regions, (4.3) and (4.4)
yield
u = u(x, t0 ),
(4.5) v(x, t) = vt (x, t0 )(t − t0 ) + v(x, t0 ),
λ = λe (v(x, t)),
where the last equation is the explicit form of (4.4), which exists due to pλ > 0. The
characterization of admissible isobaric solutions of (4.3) is left for future investigations.
From the above discussions, we can define admissible piecewise smooth solutions
as follows.
Definition 4.2. (i) A shock wave solution of (4.3) is admissible if the shock has
a stable or metastable traveling wave profile.
(ii) A piecewise smooth solution of (4.3) is admissible if all its discontinuities
have stable or metastable shock profiles and all smooth parts of the solution are either
nonreacting rarefaction waves or isobaric waves admissible in the sense of Definition
4.1.
The following are examples of Riemann problems of (4.3) exhibiting wave patterns
observed in shock tube experiments on retrograde fluids.
Example 4.1. This example shows that solutions of Riemann problems of (4.3) can
exhibit the wave splitting phenomena as depicted by Figure 1.2(b). The compression
shock is set by the initial condition

(u− > 0, v− , λ− = 0) if x < 0,

(4.6) (u, v, λ)(x, 0) =
(u+ = 0, v+ , λ+ = 1) if x > 0
with v− < v+ and p(v+ , λ+ ) < pe < p(v− , λ− ).

In these initial data, the fluid is stable vapor at rest for x > 0 and stable liquid
for x < 0 moving from left to right.
We assume that pvv > 0. From the classical theory of the Riemann problem of
the p-system for isothermal gas dynamics, we know that the states (u∗ , v∗ , λ∗ = 0)
that can connect (u− , v− , λ− = 0) from the right by backward nonreacting shocks are
determined by

(4.7) S1 : u∗ − u− = − (v∗ − v− )(p− − p∗ ), α ≥ v− > v∗ ,
while those by backward rarefaction waves are given by

v∗
(4.8) R1 : u∗ − u− = −pv (y, 0)dy, v− < v∗ ≤ α.
v−
∗ ∗ ∗
Those states (u , v , λ = 1) that can connect (u+ , v+ , λ+ = 1) from the left by
forward nonreacting shocks and rarefaction waves are determined by

(4.9) S2 : u∗ − u+ = (v ∗ − v+ )(p+ − p∗ ), β ≤ v ∗ < v+
and
v+
(4.10) R2 : u+ − u∗ = −pv (y, 0)dy, v ∗ > v+ ≥ β,
v∗
1294 HAITAO FAN
Fig. 4.1(a). The curves S1 and R1 are the set of points that can be connected to (u− , v− ) from
the right by a backward shock and a backward rarefaction wave, respectively. The curves S2 and
R2 are the set of points that can be connected to (u+ , v+ ) from the left by a forward shock and a
forward rarefaction wave, respectively.
Fig. 4.1(b). BW is a backward nonreacting shock or rarefaction wave. CD is a condensation

discontinuity. FW is a forward nonreacting shock or rarefaction wave.
respectively; see Figure 4.1(a). Then the structure of the general solution of the
Riemann problem with initial data (4.6) is as shown in Figures 4.1(a) and 4.1(b)
which consists of, from left to right, a backward nonreacting shock or rarefaction
wave, BW, followed by a condensation discontinuity, CD, and a faster moving forward
nonreacting shock or rarefaction wave, FW.
Given initial value (4.6), for ab > 0 suitably small, Corollary 3.5 guarantees the
existence of shock profiles connecting (v− , λ− ) to some (v ∗ , λ∗ = 1) with the speed
∗
p − p−
c1 = − ∗ > 0.
v − v−
Assume this shock profile is stable. To see that a wave pattern in Figure 1.2(b) is
present in solutions of Riemann problems of (1.4), we choose v+ > v ∗ and such that
the slope of the chord connecting (v ∗ , p(v ∗ , 1)) and (v+ , p+ ) is larger than c21 ; see
Figure 4.2. Choose u± to satisfy the Rankine–Hugoniot condition
u+ − u− = −c1 (v ∗ − v− ) − c2 (v+ − v ∗ );
then the Riemann problem (4.3) has a solution depicted in Figure 4.3. The pressure
pattern of the solution is the same as depicted in Figure 1.2(b) for any fixed time.
Fig. 4.2.
Fig. 4.3.
This solution consists of a nonreacting gas dynamic shock, FS, followed by a slower
moving liquefaction shock, CD. The wave pattern of this solution is the same as Figure
1.2(b). One may notice that in Figure 1.2(b), it is mostly a mixture of liquid and
vapor after the phase boundary. This is because when vapor condenses to liquid as
it passes through the condensation discontinuity, latent heat is released which causes
temperature to rise. This in turn raises the equilibrium pressure pe (T ) which will
stop the condensation and thus the condensation is partial and a mixture of liquid
and vapor is created. The above solution of the Riemann problem is for the isothermal
case which excludes the mechanism related to temperature change and hence we get
the complete liquefaction shock.
The above example shows that solutions of (1.4) can exhibit the wave pattern
depicted in Figure 1.2(b). Now we consider the possibility of the quantitative agree-
ment of such a Riemann solver and the actual experimental data in order to assess
1296 HAITAO FAN
Table 4.1
M0 p∗(bars) p1 (bars) p∗exp p1,exp split?
1.81 3.646 3.835 3.60 3.80 Y
2.07 4.774 4.999 4.70 5.00 Y

2.13 5.056 5.267 5.00 5.10 Y
2.21 5.444 5.684 5.30 5.70 Y
2.44 6.640 6.896 6.50 7.10 N
2.48 6.860 7.118 6.85 7.60 N
2.72 8.258 8.519 7.85 8.85 N
2.81 8.813 9.075 8.85 10.48 N
the adequacy of the model (2.8), the “complete” version of (1.4). To fix ideas, we
consider only experiments shown in Figure 1.2(b) since it is the one with most data
available from [Gu, TCK, TCMKS]. Large differences between the calculation based
on (2.8) and the experimental data can arise mostly for the following two reasons:
(1) the system (2.8) is inadequate or (2) the state functions such as p(v, λ) and pa-
rameters µ, γ, , α, β, etc., are inaccurate. We see that once we know the solution is
of the two-wave structure depicted in Figure 1.2(b), then the pressure p0 , p∗ , and p1
of the solution (see Figure 4.3) is completely determined by the Rankin–Hugoniot
jump conditions. Rankin–Hugoniot conditions, stating the conservation of mass, mo-
mentum, and energy across shocks, depend only on state functions. Thus, if both
the solution of (2.8) and the experiment with the same initial data show the two-
wave structure depicted in Figure 1.2(b), then inaccuracy can arise only from reason
(2). This indicates the possibility of quantitative accuracy of (2.8) by using more
accurate state functions and parameters. Table 4.1 is Table 7.3 of [Gu] showing the
calculated pressure p∗ and p1 from Rankin–Hugoniot conditions with Hobbs pressure
function [Hobbs] with 64 terms, and with the corresponding pressure data observed
in experiments. The fluid used in experiments is 2,2,4-trymethylpentane (iso-octan,
C8 H18 ). In the table, the pressure and temperature upstream are p0 = 1.1 bars and
T0 = 110o C; the Mach number of incident shock is M0 ; subscript exp means the data
are from actual experiments, while all other data are computed.
From the above reasoning, we see that difference between the calculation based
on (2.8) and experiments due to reason (1) only occurs when the solution of (2.8) does
not show the two-wave structure consisting of three almost constant pieces while the
experimental data do. We claim that the difference caused by this reason is no more
than O(1)|pβ − pe | under the assumption that p1 (v.T, λ) and p1 (v.T, λ) are convex in
v, where pβ is the pressure at the vapor spinodal limit. Indeed, under the convexity
assumption on p1 and p2 with respect to v, an admissible solution of the Riemann
problem with p− < pe > p+ and λ− = 0 < λ+ = 1, modeling the incident liquefaction
shock propagating into stable vapor, consists of a backward nonreacting shock or
rarefaction wave, a forward phase boundary, and a faster running forward nonreacting
shock, as shown in Figure 4.1(b). Since the pressure p∗ and actual pressure pexp are
in the range pe ≤ p∗ ≤ pβ and the Rankin–Hugoniot jump conditions depend on p∗
continuously, the errors in pressure data |p∗ − p1 | and |pexp − p∗ | are of the order
O(1)|pβ − pe |. The same is true for all other quantities on two sides of the forerunning
nonreacting shock and the trailing condensation discontinuity decided by Rankin–
Hugoniot conditions. Of course, the above estimates are very rough. More detailed
estimates and computations are left for future research.
Example 4.2. In this example, we intend to simulate Figure 1.2(a) where the
piston is withdrawn, causing depressurization to induce evaporation. Let initial data
Fig. 4.4.
be

(u− < 0, v− , λ− = 1) if x < 0,

(4.11) (u, v, λ)(x, 0) =
(u+ = 0, v+ , λ+ = 0) if x > 0
with v− > v+ and p(v+ , λ+ ) > pe > p(v− , λ− ).
In these initial data, the fluid is stable liquid at rest for x > 0 and stable vapor
moving to the left for x < 0. Let v− be such that there is an admissible shock that
connects (v− , λ− = 1) to (v ∗ , λ∗ = 0) with speed c1 > 0 and p∗ < pe and hence
v ∗ > v+ ; see Figure 4.4.
There is a nonreacting smooth rarefaction wave connecting v ∗ to v+ . Suppose u±
are chosen to satisfy
v+
u+ − u− = −c1 (v ∗ − v− ) + −p (v, 0)dv.
v∗
Then the solution of Riemann problem (4.3) with such Riemann initial data consists
of a forward rarefaction wave FW and a slower forward evaporation discontinuity ES,
as shown in Figure 4.5. The structure of the solution is the same as observed in
experiments on retrograde fluids depicted in Figure 1.2(a).
Example 4.3. The first part of phenomenon (c) in section 1 states that withdraw-
ing the piston from an equilibrium mixture of vapor and liquid creates a rarefaction
shock. This kind of shock data are λ− = 1, λ+ ∈ (0, 1), p− < pe , and p+ = pe . System
(1.4) has such shocks as solutions since the existence of these rarefaction shocks is
guaranteed by Theorem 3.4 and Corollary 3.5.
The second part of phenomenon (c) says that given a rarefaction initial data,
where particles of the fluid are flowing away from each other, the system will sharpen
up to form a (rarefaction) evaporation shock. This phenomenon is worth our attention
since in classical shock theory of gas dynamics, rarefaction initial data lead to smooth
rarefaction wave, not shocks.
Since the Riemann problem is not suitable for studying sharpening up of the
solution, we shall numerically verify that this phenomenon is also present in system
(1.4) in Eulerian coordinates in this example.
1298 HAITAO FAN
Fig. 4.5.
First we nondimensionalize the variables with respect to the critical state ρc ,

pc and a typical length L, such as the diameter of the tube, as follows: ρ = ρ̄/ρc ,
m = ρu = m̄/(ρc pc )1/2 , p = p̄/pc , T = T̄ /Tc , t = t̄(pc /ρc )1/2 /L, and x = x̄/L,
where variables with bars denote variables with dimensions and those without bars
are dimension-free variables. The standard van der Waals pressure function p for the
above dimension-free variables is
8T ρ
(4.12) h(ρ) = − 3ρ2 .
3−ρ
In our computation, we take the temperature to be T = 7.5/8. Equilibrium pressure
pe (T ) can be computed from (4.12) by the Maxwell equal area rule. The pressure
functions p1 and p2 used in the computation are

h(ρ)
if ρ ≤ α,
p1 (ρ) = β−3
h 3−α (3 − ρ) + 3 + h(α) − h(β) if ρ > α
and

h(ρ) if ρ ≥ β,
p2 (ρ) = h(β) α
h(α) h β ρ if ρ < β.
The reaction rate function w1 is taken as

(4.13)
peq − p
w1 = λ(1 − λ)ρ
γpeq

peq − p c4 p2eq
+ χ(0 ≤ λ ≤ 1) ρ c1 χ(p < pβ ) + c2 χ(p > pα ) + c3 exp − ,
peq (p − peq )2
where γ is the typical reaction time. The parameters in (1.4) used in the computations
are = 0, γ = 0.005, µ = 2γ/5, c1 = c2 = 1/γ c3 = 1, and c4 = 0.1. The grid steps
are ∆x = 0.01 and ∆t = 0.002.
Fig. 4.6. Initial data for Example 4.3. The fluid is stable vapor at x = 0 and stable liquid
at x = 2 and smoothly varying in between. The three horizontal lines are, from high to low, the
upper spinodal limit pressure p(v = β, λ = 1), the equilibrium pressure pe , and liquid spinodal limit
pressure p(α, λ = 0). Thus, initial data is rarefying in the sense that the fluid on the left flows to
the left, as indicated by the negative momentum, while that on the right flows to the right.
Fig. 4.7. At time t = 1.332, the solution sharpens up from the smooth initial data shown in
Figure 4.6(a) to a rarefaction shock.
1300 HAITAO FAN
The numerical method we used is as follows. We discretize the diffusion terms by

central difference. Since we expect that the speed of the phase boundary is sensitive
to diffusion terms as indicated by Theorem 1.2, the numerical viscosity should be
made at least one order smaller than the diffusion terms. For this reason, we used
the third-order WENO method [JS] to discretize the left-hand side of (1.4). Iteration
scheme in time is the third-order Runge–Kutta method, also from [JS].
The initial data are shown in Figure 4.6, where the fluid is stable vapor at x = 0
and stable liquid at x = 2 and smoothly varying in between. The three horizontal
lines in Figures 4.6 and 4.7 are, from high to low, the upper spinodal limit pressure
p(v = β, λ = 1), the equilibrium pressure pe , and liquid spinodal limit pressure
p(α, λ = 0). The fluid on the left flows to the left, as indicated by the negative
momentum, while that on the right flows to the right. Thus, it is a rarefaction initial
data. As shown in Figure 4.6(a), the initial data are quite smooth. Figure 4.6(b)
shows that at time t = 1.332, the solution sharpens up to a rarefaction shock. Thus,
system (1.4) also exhibits the wave pattern (c).
REFERENCES
[Ba] D. Bauschdorff, Carrier gas effects on homogeneous nucleation of water vapor in a

shock tube, Phys. Fluids, 18 (1975), pp. 529–535.
[FD] W. Fickett and W. C. Davis, Detonation, University of California Press, Berkeley,
CA, 1979.
[Fan1] H. Fan, One-phase Riemann problem and wave interactions in systems of conservation
laws of mixed type, SIAM J. Math. Anal., 24 (1993), pp. 840–865.
[Fan2] H. Fan, Traveling waves, Riemann problems and computations of a model of the dy-
namics of liquid/vapor phase transitions, J. Differential Equations, 150 (1998), pp.
385–437.
[Gu] S. C. Gulen, A Study of Shock Splitting and Liquefaction Shocks in Fast Phase Changes
in Retrograde Fluids, Ph.D. dissertation, Rensselaer Polytechnic Institute, Troy,
NY, 1992.
[GTC] S. C. Gulen, P. A. Thompson, and H. J. Cho, An experimental study of reflected
liquefaction shock waves with near critical downstream states in a test fluid of
large molar heat capacity, J. Fluid Mech., 277 (1994), pp. 163–196.
[Hobbs] D. E. Hobbs, A Virial Equation of State Utilizing the Corresponding State Principle,
Ph.D. thesis, Rensselaer Polytechnic Institute, Troy, NY, 1983.
[Hs] L. Hsiao, Uniqueness of admissible solutions of Riemann problem of systems of con-
servation laws of mixed type, J. Differential Equations, 86 (1990), pp. 197–233.
[JS] G.-S. Jiang and C. W. Shu, Efficient implementation of weighted ENO schemes, J.
Comput. Phys., 1261 (1996), pp. 202–228.
[KG] S. Kotake and I. I. Glass, Flows with nucleation and condensation, Progr. Aerospace
Sci., 19 (1981), pp. 129–196.
[MP] R. Menikoff and B. J. Plohr, The Riemann problem for fluid flow of real materials,
Rev. Modern Phys., 61 (1989), pp. 75–130.
[Ox] D. W. Oxtoby, Homogeneous nucleation: Theory and experiment, J. Phys. Condens.
Matter, 4 (1992), pp. 7627–7650.
[RFF] R. L. Rabie, G. R. Fowles, and W. Fickett, The polymorphic detonation, Phys.
Fluids, 22 (1979), pp. 422–435.
[Sh] M. Shearer, Nonuniqueness of admissible solutions of Riemann initial value problem
for a system of conservation laws of mixed type, Arch. Rational Mech. Anal., 93
(1986), pp. 45–59.
[SG] J. P. Sisilian and I. I. Glass, Condensation of water vapor in rarefaction waves: I.
Homogeneous nucleation, AIAA J., 14 (1976), pp. 1731–1737.
[Sl] M. Slemrod, Admissibility criterion for propagating phase boundaries in a van der
Waals fluid, Arch. Rational Mech. Anal., 81 (1983), pp. 301–315.
[Sp] G. S. Springer, Homogeneous nucleation, Adv. in Heat Transfer, 14 (1978), pp. 281–
345.
[SX] A. Szepessy and Zhouping Xin, Nonlinear stability of viscous shock waves, Arch.
Rational Mech. Anal., 122 (1993), pp. 53–103.
[TCK] P. A. Thompson, G. C. Carofano, and Y.-G. Kim, Shock waves and phase changes
in a large-heat-capacity fluid emerging from a tube, J. Fluid Mech., 166 (1986), pp.
57–92.
[TCMKS] P. A. Thompson, H. Chaves, G. E. A. Meier, Y.-G. Kim, and H.-D. Speckmann,
Wave splitting in a fluid of large heat capacity, J. Fluid Mech., 185 (1987), pp.
385–414.
[VVV] A. I. Volpert, V. A. Volpert, and V. Volpert, Traveling Wave Solutions of
Parabolic Systems, Transl. Math. Monogr. 140, AMS, Providence, RI, 1994.
[Wa] G. B. Wallis, One-Dimensional Two Phase Flow, McGraw–Hill, New York, 1969.
[WLe] P. P. Wegener and C. F. Lee, Condensation by homogeneous nucleation of H2 O,
C6 H6 , CCl4 and CCl3 F in a shock tube, J. Aerosol. Sci., 14 (1983), pp. 29–37.
[WLu] P. P. Wegener and G. Lundquist, Condensation of water vapor in the shock tube
below 150 K, J. Appl. Phys., 22 (1951), p. 233.
[Wu] B. J. C. Wu, Analysis of condensation in the centered expansion wave in a shock tube,
in Condensation in High Speed Flows, A. A. Pouring, ed., ASME, New York, 1977,
pp. 73–82.
[WW] P. P. Wegener and B. C. J. Wu, Gas dynamics and homogeneous nucleation, in
Nucleation Phenomena, A. C. Zettlemoyer, ed., Elsevier, New York, 1977, pp. 325–
417.
[X] Z. P. Xin, Asymptotic stability of rarefaction waves for 2 × 2 viscous hyperbolic con-
servation laws, J. Differential Equations, 72 (1988), pp. 45–77.
[ZBLM] Ya. B. Zeldovich, G. I. Barenblatt, V. B. Librovich, and G. M. Makhviladze,
The Mathematical Theory of Combustion and Explosions, Consultants Bureau, New
York, 1985, p. 4.

On A Model of The Dynamics of Liquid/Vapor Phase Transitions

Uploaded by

Copyright:

Available Formats

On A Model of The Dynamics of Liquid/Vapor Phase Transitions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

On A Model of The Dynamics of Liquid/Vapor Phase Transitions

Uploaded by

Copyright:

Available Formats

SIAM J. APPL. MATH.

c 2000 Society for Industrial and Applied Mathematics

ON A MODEL OF THE DYNAMICS OF LIQUID/VAPOR PHASE

Key words. liquid/vapor phase transition, traveling wave, Riemann problem

AMS subject classiﬁcations. 35K57, 35L65, 35L67, 35O30, 76N10, 76T05

1. Introduction. Consider a pure ﬂuid that exhibits liquid/vapor phase changes.

Fig. 1.2. Conceptual representation of one-dimensional wave splitting in a liquid-vapor sys-

transitions, it should at least exhibit wave patterns observed in actual experiments.

When the upstream state is an equilibrium liquid/vapor mixture, then there is

In [Fan2], we proved the following results.

(1.12a) (v+ − v− )(λ+ − λ− ) > 0,

(1.12b) (v+ − v− )(p+ − pe ) ≥ 0,

(1.13) c2 > 4ab max(p(v+ , 1) − pe , p(v− , 1) − pe ),

then there is a monotone solution of (1.9). Furthermore, if c2 < 4ab(p+ − pe ), then

Precise statements of the main results of this paper are as follows.

(1.14a) (v+ − v− )(λ+ − λ− ) > 0,

(1.14b) (v+ − v− )(p+ − pe ) ≥ 0,

(1.14c) c2 + pv± < 0,

then there is a traveling wave solution of (1.9). If the speed c satisﬁes

then there is no traveling wave of (1.9) satisfying λ ∈ [0, 1].

(1.14a ) (v+ − v− )(λ+ − λ− ) > 0,

(1.14b ) (v+ − v− )(p− − pe ) ≤ 0,

(1.14c ) c2 + pv± < 0,

then there is a traveling wave solution of (1.9). If the speed c satisﬁes

then there is no traveling wave of (1.9) satisfying λ ∈ [0, 1].

2. A model for the dynamics of phase transitions in liquid/vapor sys-

phases by a reaction rate equation. We introduce the following notation.

fi (x, t) is the volume fraction of ith species at point x and time t;

(2.4) p = p1 (ρ1 , T ) = p2 (ρ2 , T ).

(A4) The stress tensor is

(ρu)t + (ρu2 + p)x = uxx ,

In Lagrange coordinates, the corresponding system of equations is

where λ = ρ1 f1 /ρ is the mass fraction of vapor in liquid/vapor mixture and e =

∂p p1 (v1 )p2 (v2 )

Further calculation shows that

(3.2 ) Ut = AUxx + F (U, c)

for short, where U = (v, λ)T , A = diag(η, b), and

(3.5) Ut = AUxx + G(U )

Now we consider the eigenvalues of ∇F at (v± , λ± ). By the Rankine–Hugoniot

(3.6a) α1± = c2 + pv (v± , λ± )

(3.6b) α2± = a(p± − pe )(2λ± − 1).

At p = pe , the eigenvalues of ∇F are

α1− = c2 + pv− < 0,

(3.7 ) c2 + pv+ < 0, p+ > pe , λ+ = 0

(3.8 ) c2 + pv+ < 0, p+ < pe , λ+ = 1

(3.9) G(rq + U+ ) ≥ 0 for 0 ≤ r ≤ r0 ,

there exists a decreasing traveling wave of (3.5) with speed s ≥ 0, where

When s < ω such a solution does not exist.

Theorem 3.1 can be rewritten as follows to allow U+ > U− .

(3.9 ) G(rq + U+ ) ≤ 0 for 0 ≤ r ≤ r0 ,

then there is a solution of (3.1) for η > 0 small enough.

(3.12) c2 + pv± < 0, U+ = (v+ , λ+ ) < (v− , λ− ) = (v− , 1) = U− , p− < p+ ≤ pe ,

(3.15) F2 (rq + U+ , c) = a(p(rq + U+ ) − pe )(rq2 + λ+ )(rq2 + λ+ − 1) > 0

F1 (rq + U+ , c) = c2 rq1 + p(rq + U+ ) − p+ .

Its derivative satisﬁes

Since j = 1, 2 in the above, we can consider j = 1 and 2 separately. We start with

bλ + a(p − pe )λ(λ − 1)

(3.17) λ(x) ∈ K2 := {λ ∈ C 2 (R; R) | λ(x) is decreasing and lim λ(x) = λ± },

we can select a function v(x) = v(λ(x)) so that

(ρu)t + (ρu2 + p)x = uxx ,

(4.2) (u, v, λ)(x, t) := lim (un , v n , λn )(x, t)