J. Math. Pures Appl.

104 (2015) 801836

Contents lists available at ScienceDirect

Journal de Mathmatiques Pures et Appliques

www.elsevier.com/locate/matpur

Two-velocity hydrodynamics in uid mechanics: Part II

Existence of global -entropy solutions to the compressible
NavierStokes systems with degenerate viscosities
Didier Bresch a, , Benot Desjardins b , Ewelina Zatorska c,d,e
a
LAMA UMR5127 CNRS, Universit de Savoie Mont-Blanc, 73376 Le Bourget du Lac, France
b
Fondation Mathmatique Jacques Hadamard and CMLA ENS Cachan, 94235 Cachan Cedex, France
c
Centre de Mathmatiques Appliques, cole Polytechnique, 91128 Palaiseau Cedex, France
d
Institute of Mathematics, Polish Academy of Sciences, ul. niadeckich 8, 00-656 Warszawa, Poland
e
Institute of Applied Mathematics and Mechanics, University of Warsaw, ul. Banacha 2,
02-097 Warszawa, Poland

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

Article history: This paper addresses the issue of global existence of the so-called -entropy solutions
Received 14 November 2014 to the NavierStokes equations for viscous compressible and barotropic uids
Available online 6 May 2015 with degenerate viscosities. We consider the three dimensional space domain with
periodic boundary conditions. Our solutions satisfy the weak formulation of the mass
MSC:
35Q30 and momentum conservation equations and also a generalization of the BD-entropy
76N10 identity called: -entropy. This new entropy involves a mixture parameter (0, 1)
35D30 between the two velocities u and u + 2() (the latter was introduced by the rst
two authors in Bresch and Desjardins (2005) [5]), where u is the velocity eld
Keywords: and is a function of the density  dened by  (s) =  (s)/s. As a byproduct
Hypocoercivity of the existence proof, we show that two-velocity hydrodynamics (in the spirit of
Compressible NavierStokes S.C. Shugrin, 1994) is a possible formulation of a model of barotropic compressible
Augmented system ow with degenerate viscosities.
Two-velocity hydrodynamics 2015 Elsevier Masson SAS. All rights reserved.
-Entropy

r s u m

Cet article concerne lexistence globale de solution -entropique des quations de

NavierStokes pour un uide un visqueux compressible barotrope avec viscosits
dgnres pour un domaine tri-dimensionnel priodique. Les solutions satisfont
la formulation faible des quations de conservation de la masse, des quations
de conservation des moments et galement une BD-entropie gnralise appele :
entropie. Cette nouvelle entropie fait intervenir un paramtre de mlange (0, 1)
entre deux vitesses u et u + 2() (la dernire vitesse a t introduite par les deux
premiers auteurs dans Bresch et Desjardins (2005) [5]), o u est le champ de vitesse
et est une fonction de la densit  dnie par  (s) =  (s)/s.
2015 Elsevier Masson SAS. All rights reserved.

* Corresponding author.
E-mail address: didier.bresch@univ-smb.fr (D. Bresch).

http://dx.doi.org/10.1016/j.matpur.2015.05.004
0021-7824/ 2015 Elsevier Masson SAS. All rights reserved.
802 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

1. Introduction

In 20062007, the rst two authors introduced the concept of global weak solutions to the NavierStokes
equations for compressible barotropic uids degenerate viscosities satisfying the energy inequality and an
extra mathematical entropy called BD entropy, [79]. The BD entropy identity was derived in [5] for
() = 2( () ()) as a generalization of the work of the rst two authors and C.K. Lin [6], where
() =  and () = 0. It involves an energy related to the velocity u + 2() where is a function of
the density  dened by  (s) =  (s)/s. It is worth to note that such quantity appears also in study of
inviscid systems, see [3] where () = log  and in the works by E. Nelson (see for example [34]) related to
kinematics of Markovian motion with interesting discussion using two velocities formulation. There, quantity
/ stands for velocity required for the particle to counteract osmotic eects (osmotic velocity) and the
current velocity.
Including a drag force in the momentum equation or an additional singular pressure terms, stability
of global weak solutions to the barotropic compressible NavierStokes equations with density dependent
viscosities satisfying the extra BD entropy was proven in [9,7] and with D. Grard-Varet in [10],
see also [6]. Stability of global weak solutions satisfying the extra BD entropy without these extra terms
was obtained by A. Mellet and A. Vasseur in [31]. They showed additional estimate on the velocity eld
but the construction of approximate solutions satisfying the energy, the BD entropy and the MelletVasseur
estimate together was so far an open problem. Very recently A. Vasseur and C. Yu have proved in [38]
that given a global weak solution to the compressible NavierStokes equations with turbulent drag terms
and appropriate capillary term satisfying the energy and the BD entropy estimates, it is possible to pass
to the limit and get a weak solution to the compressible NavierStokes system. They constructed smooth
multipliers allowing to get the MelletVasseur estimate uniformly with respect to the drag coecient and
to pass to the limit in the capillary terms. The uniform control allows to suppress the drag terms letting
the drag coecients tend to zero. This result, when coupled with our present paper gives the rst complete
existence result to the compressible NavierStokes equations with general degenerate viscosities with no
extra terms. Here we construct the weak solution to the system similar to the considered in [38] modulo
capillarity terms. It was however shown in [11] that this construction is compatible with the quantum

capillary term of the form ( / ) appearing in the ghost system [28], see also [21] for the study on
quantum viscous NavierStokes system and [4] for a full range of compatible capillary terms.
Concerning construction of approximate solutions with singular pressure, drag terms or capillarity terms,
the authors gave some hints in [8] for the general setting () = 2( () ()). This has been fully
developed in [42,32,33] in the case of a linear viscosity () = . However, even for this case, the procedure
seems to be very complex. The approximation involves: i) regularization of the mass equation, ii) an extra
regularizing terms in the momentum equation inspired by those given in [8], iii) high-order generalized
capillarity term, and iv) regularization of the velocity u + 2().
In the present paper, we propose to come back to the compressible NavierStokes equations with
degenerate viscosities in the general setting with simpler construction of approximate solutions. We propose
a very natural concept of global -entropy solutions based on a generalized BD entropy. The global
weak solution of the compressible NavierStokes equations which satises the BD entropy is also a
-entropy solution for all 0 < < 1, however, the starting points for the proofs of existence are rather
dierent. Through -entropy we introduce a two-velocity hydrodynamical formulation of the compressible
NavierStokes equations with degenerate viscosities. Our construction of the approximate solutions is based
on an augmented approximate scheme using this two-velocity structure. Similar approximate scheme has
been introduced for the zero Mach number system in the rst part of the present series [11]. For readers
convenience, since lack of divergence-free condition gives rise to several new terms and since the -entropy
and the limit process changes, we repeat the whole construction from [11]. Consistent numerical schemes
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 803

for compressible ows with degenerate viscosities based on the augmented system is the purpose of the
forthcoming paper [12].
By using augmented systems, we extend an approach developed for inviscid compressible systems with
dispersion (see for instance S. Benzoni, R. Danchin, S. Descombes [2] and F. Bthuel, R. Danchin,
P. Gravejat, J.-C. Saut, D. Smets [3]). Lately, similar structure was also used by P. Noble and
J.-P. Vila [36] to study the stability of various approximations of the one-dimension EulerKorteweg
equations (dispersive system). They introduced an additional unknown (the gradient of a function of
the density) in order to rewrite the system into a hyperbolic system perturbed by a second order skew
symmetric term. Derivation of relevant numerical scheme using an additional unknown for multi-dimensional
EulerKorteweg system (dispersive system) is purpose of the paper [4].
In our construction, the only regularization of the continuity equation is via regularization of the new
velocity w = u + 2() and no further viscous approximation is needed. Parabolicity of the equation
for the density written in this form has been recently observed and studied by B. Haspot in [22] with
the change of variables v = u + (). This property was then used to obtain result concerning global
controllability for the shallow-water system with two control forces, see [16].
The change of unknowns from the work of B. Haspot corresponds to = 1/2 from our denition of
-entropy solution. The 1/2-entropy solution has been obtained recently by M. Gisclon and I. Violet [21],
for () =  and () = 0, starting from the quantum compressible NavierStokes equations studied by
A. Jengel in [25] (extended in [15] and [24]) with an additional singular pressure and letting the scaled
Planck constant vanish. Their proof strongly relies on the Bohm potential identity and therefore works
only for () =  in the multi-dimensional space case. See also result by B. Haspot concerning Korteweg
systems in [23] and references cited therein.
Interestingly enough, it turns out that our two-velocity formulation is linked to [37] and [20]: we introduce
a mixture parameter (0, 1) which combines two velocity vector elds u and u + 2() sharing the
same reference density. The augmented system used to construct approximate solution is also similar to the
one from [37] where the main velocity u + 2() and a drift 2() correspond to the same reference
density . The reader is also referred to the paper by E. Feireisl and A. Vasseur [19] where they study
a compressible system with two-velocities proposed by H. Brenner.
The generalized BD entropy (-entropy) reects some non-linear hypocoercivity property of the
nonlinear compressible NavierStokes equations under the aforementioned relation between and . For an
introduction to hypercoercivity, the interested reader is referred to the paper by K. Beauchard and
E. Zuazua [1] and the book by C. Villani [40] (and references cited therein) which describe its link to
global existence around equilibrium and large-time behavior (see [14] by R. Danchin for an application to
uid mechanics). Readers interested in entropy for nonlinear partial dierential equations are also referred
to [13]. Our generalized BD entropy (-entropies) may be seen as a nonlinear version of the identity that was
proven by A. Matsumura and T. Nishida on the linearized compressible system around the equilibrium
(eq , ueq ) = (1, 0). For the readers convenience, we revisit hypocoercivity on linearized compressible
NavierStokes systems in the last two sections of the paper (barotropic and heat-conducting case).
The purpose of the last section is also to present a thermodynamically consistent two-velocity model
with heat conductivity in the spirit of the work of S.M. Shugrin [37]. Each of the velocity vector elds
u and 2() is associated with dierent density (1 ) and , respectively. The main objective is
not to prove global existence of solutions rigorously but to show that the two-velocity hydrodynamics is
consistent with the study performed in the rst part of the present series [11] for the zero Mach number
system. Our formulation uses the generalized -temperature which is not a priori the usual temperature.
In this sense, our formulation and the usual heat-conducting compressible NavierStokes equations are
not equivalent. However, the formal low-Mach number limit for such two-velocity system with generalized
temperature gives exactly the augmented system from [11] used to obtain the existence result. This is a
kind of consistency between the approaches and the systems.
804 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

2. The barotropic NavierStokes system

Two compressible uid models with degenerate viscosity and pressure depending only on the density
(barotropic ows) will be considered.
1) Compressible NavierStokes with singular pressure. First the compressible NavierStokes equations for
compressible and barotropic uids write as follows:

t  + div (u) = 0,
in (0, T ) , (1)
t (u) + div(u u) div (2()D(u)) (()div u) + p() = 0,

where D(u) = 12 (u + t u), (div(u u)j = i (ui uj ) and is a periodic box = T3 . The pressure p is
singular close to zero density

 c1  1 for   ,
p () = (2)
c2  1 for  > 
+

with ci > 0, + > 1 and > 0. A more precise estimation of will be given below.
The viscosity coecients (), () satisfy the BreschDesjardins relation introduced in [5]

() = 2( () ()). (3)

This system is completed with initial data

|t=0 = 0 , (u)|t=0 = m0 = 0 u0 . (4)

The pressure (2) was introduced in [9]. Note that the singular part is active only for density close to
vacuum and that the usual power law equation of state is recovered far from vacuum. The physical relevance
of the compressible NavierStokes equations is very questionable in regions where density is close to zero:
the medium is not only unlikely to be in a liquid or gas state (elasticity and plasticity has to be considered
for such solid materials, for which by the low densities may lead to negative pressures), but also the rareed
regime of vanishing densities violates the assumptions on the mean free path of particles suitable for uid
models. The negativity of the cold pressure implied by the above assumptions for low densities may also
be interpreted as some articial way to get close to a solid state in tension. This is exactly the idea in the
work [26] where intermolecular forces, namely long-range attractive van der Waals and short-range Born
repulsive intermolecular forces, are considered.
From the mathematical point of view, the singular pressure (2) has some stabilizing properties. They were
used in [32] to investigate the model of compressible mixture. Steady compressible NavierStokes system
with pressure singular at vacuum was also investigated by M. asica [30]. For studies on doubly singular
pressure in the context of mixtures we refer to paper by E. Feireisl, Y. Lu and J. Mlek [18]. See also
the work by G. Kitavsev, P. Laurenot, B. Niethammer [26] where they study lubrication equations
in the presence of strong slippage.

2) Compressible NavierStokes with a turbulent drag term. We discuss, in Section 4, how to handle the
compressible NavierStokes equations with turbulent drag term (r1 |u|u, r1 > 0) and standard gamma-type
pressure law, namely

t  + div (u) = 0,
(5)
t (u) + div(u u) div (2()D(u)) (()div u) + r1 |u|u + p() = 0
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 805

with the usual pressure law p() = a with > 1. The assumptions on the viscosity in this section will be
similar to the one introduced in [31] with the extra control () c2/3+1/3 for (1/, 1) when  1.
This additional assumption is introduced because the energy and BD entropy are mixed in the -entropy
(less information is available compared to [10]). In this case, we will only focus on the extra terms that
have to be included in the augmented system and show how the -entropy changes. The existence result
and asymptotic limit with respect to smoothing parameters , n and will not be aected when is xed.
The asymptotic limit when tends to 0 is then the same as in [10].

2.1. Denition of a global -entropy solution of (1)(4)

Denition 1. Let be such that 0 < < 1, the couple of functions (, u) is called a global -entropy solution
to system (1)(4) if the following properties are satised:

The mass equation is satised in the following sense

  
 t dx u dx = 0 (0) dx (6)

for all Cc ([0, T ) ).

The momentum equation is satised in the following sense

  
u t dx (u u) : dx + 2()D(u) : dx

  
+ ()div u div dx p()div dx = 0 u0 (0) dx (7)

for all (Cc ([0, T ) ))3 .

Moreover (, u) satises, for all t [0, T ], the following -entropy estimates

  

|u + 2()|2 |2()|2
sup  + (1 ) (t) dx + e()(t) dx
t[0,T ] 2 2

T  T 
 ()p ()
+ 2 2
()|A(u)| dx ds + 2 ||2 dx ds

0 0

T  

+ 2(1 ) ()|D(u)|2 dx + ( () ())|div u|2 dx ds

0
  
|u + 2( )|
0 0
|2(0 )|2
2
 0
+ (1 ) dx + 0 e(0 ) dx (8)
2 2

with  (s) =  (s)/s, A(u) = 12 (u t u) and the internal energy e() dened by

2 de()
= p().
d
806 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

Remark 1. Note that

 
|u + 2()|2 |2()|2
 + (1 ) (t) dx
2 2

 
|u|2 |u + 2()|2
=  (1 ) + (t) dx. (9)
2 2

That means that the -entropy demonstrates two-velocity structure in the usual compressible NavierStokes
system. For an introduction to the two-velocity hydrodynamics and thermodynamics we refer to [37] and [20].
Indeed, using the identity

u + 2() = (1 )u + (u + 2())

we see that (9) is the kinetic energy of a two-uid mixture having the velocities u and u + 2() and
playing the role of the mass fraction.

Remark 2. Equality (8) is a generalization of the BD entropy obtained by D. Bresch and B. Desjardins
in the case = 1. More precisely, they formally derived the following identity
   
d |u + 2()|2 d  ()p ()
 dx + e() dx + 2()|A(w)|2 dx + 2 ||2 dx = 0. (10)
dt 2 dt 

If a global solution satises the BD entropy and the standard energy balance
   
d |u|2 d
 dx + () dx + 2()|D(u)|2 dx + ()|div u|2 dx = 0,
dt 2 dt

it also satises the -entropy estimate for all 0 < < 1. It suces to use the identity

|u + 2()|2 + (1 )|u|2 = |u + 2()|2 + (1 )|2()|2 ,

and therefore to add times the BD entropy to (1 ) times the energy. A global weak solution satisfying
the BD entropy is therefore a -entropy solution for all 0 < < 1. The converse is however not clear.

2.2. Main results

The initial data (4) are assumed to satisfy

(0 )
0 0, 0 L1 (), 0 e(0 ) L1 (),  L2 (), (11)
0

|m0 |2 m0
= 0, a.e. on {x : 0 (x) = 0},  L2 (), (12)
0 0

and we consider the periodic boundary conditions

= T3 .
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 807

2.2.1. Compressible NavierStokes equations with singular pressure

In what follows we will make some assumptions concerning the viscosity coecients and the pressure.
We assume that (), () are C 1 ([0, )) such that  () c > 0, (0) = 0, the following relation is satised

() = 2( () ()).

Moreover, there exist positive constants c0 , c1 ,  ,

+ 1
m > 3/4, 2/3 < n < , (13)
2
such that:

for all s <  , (s) c0 sn and 3(s) + 2(s) sn ,

sm sm
for all s  , c1 sm (s) and c1 sm 3(s) + 2(s) .
c1 c1

The pressure (2) is a C 1 ([0, )) of  and we assume that

2n(3m 2)
+ > 1, > 1. (14)
4m 3

Our main results reads as follows:

Theorem 1. Assume that 0 < < 1 is xed. If the initial data, viscosity coecients and satisfy all the
assumptions above with a singular pressure given by (2) then there exists a global in time weak -entropy
solution to (1) in the sense of Denition 1.

Remark 3. Assumptions (13) and (14) are required to pass to the limit in the convective term. More precisely
one needs to ensure that
1
 2 u Ls (0, T ; Lr ())

with r, s > 2: the details of this estimate will be given at the end of Section 3.4. The assumption on
3() + 2() is used to get estimates from (8). In the course of the proof, we combine BD relation
() = 2(() ()) with the following equality:
 2
 1  1
|D(u)| = D(u) divu I + |divu|2 .
2 
3 3

2.2.2. Compressible NavierStokes equations with a turbulent drag term

In this part, we assume the following hypotheses on viscosities () and () that may be found for
instance in [31]: let (), () be C 1 ([0, )) such that there exists a positive (0, 1) with

 () , (0) 0, (15)

and
1  1
| ()| (), () 2() + 3() (). (16)

The following relation (introduced by the two rst authors) is also assumed

() = 2( () ()). (17)

808 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

As stressed in [31], the hypotheses above imply that

C2/3+/3 () C2/3+1/(3) when  1, (18)

C2/3+1/(3) () C2/3+/3 when  1. (19)

The interested reader is referred to [31] for more details to see where such hypotheses are used for stability
purposes.
In this part, we add the following hypothesis to deal with the turbulent drag term because, compared
to [10], the -entropy mixes the usual energy estimate and the BD-entropy (less information is available):

() C2/3+1/(3) , with (1/, 1) when  1. (20)

Theorem 2. Assume that 0 < < 1 is xed. Let us assume (15)(20) and (11)(12) be satised with a
pressure p() = a with > 1, then there exists global in time weak -entropy solution to (5) in the similar
sense to Denition 1 (with the extra term coming from the turbulent drag and the power-law pressure).

Remark 4. The drag term gives an extra information on |u|2 needed to pass to the limit in the convective
term without having the MelletVasseur estimate. When no turbulent drag term is included an extra
assumption is required to get the MelletVasseur estimate, namely, if 3, it is also assumed that

()
lim inf >0
+ /3+

with > 0.

Remark 5. The system of compressible NavierStokes equations with BD structure with dierent boundary
conditions was investigated in [10]. The authors considered the Dirichlet condition for the momentum

u| = 0

as well as the Navier type condition

u n| = 0, ()(D(u)n) | = ()u | ,

together with additional boundary condition on the density, namely

()() n| = 0.

This type of boundary conditions could perhaps give transmission boundary conditions between the two
quantities u + 2() and 2() and helps to conclude in bounded domains.

2.3. Change of variables and -entropy

We generalize to the compressible framework some ideas developed recently in [11] for low Mach number
systems with heat conductivity eects. More precisely, let us dene the following velocity eld generalizing
the one introduced in the BD entropy estimate

w = u + 2() (21)
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 809

with

 ()
 () = . (22)


Note that BD entropy, as introduced in [5] by the rst two authors, corresponds to = 1. Assuming that
solutions to (1) are smooth enough, it can be shown that w satises the following evolution equation

t (w) + div (u w) 2(1 )div(()D(w)) 2div(()A(w))


+ 4(1 )div(()()) ( () 2( () () div u) + p() = 0. (23)

Let us now write the equation satised by (, w, ()). We get the system

t  + div (w) 2() = 0,

t (w) + div (u w) 2(1 )div (()w) 2div(()A(w))

+ 4(1 )div(()2 ()) ( () 2( () ()) div u) + p() = 0,
t (()) + div(u ()) 2div(()2 ()) + div(()t w)
+ (( () ())div u) = 0,
w = u + 2(). (24)

Taking the scalar product of the equation satised by w with w, the scalar product of the equation satised
by () with 4(1 )() and adding the resulting expressions we get the -entropy

  
d |w|2 |2()|2 d
 + (1 ) dx + e() dx
dt 2 2 dt

   
()p ()
+ 2 ()|A(w)|2 dx + 2 ||2 dx


 

+ 2(1 ) 2
()|D(u)| dx + ( () ())|div u|2 dx = 0, (25)

where we used assumption (3) to write


(() 2( () ())div u div w + 4(1 )( () ())div u div() dx


= 2(1 ) ( () ())|div u|2 dx (26)

recalling that u = w 2().

Remark 6. Relation (25) generalizes the one obtained in [25] (see also [21]) for quantum NavierStokes
equations, it is enough to take = 1/2 and = log . However, in contrast to aforementioned papers,
we do not need the Bohm potential formula to conclude.
810 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

3. Construction of solution

Following the idea developed recently by D. Bresch, V. Giovangigli and E. Zatorska in [11] for
the low Mach system:

t  + div(u) = 0,
t (u) + div(u u) + = 2div(()D(u)) + (()div u),
div u = 2() (27)

with an increasing function of  and 0 < < 1 xed, we construct the approximate solution to system (1)
using an augmented approximate system. More precisely, we introduce a new unknown v, which is not yet
known to satisfy v = 2(). Our aim will be to nd a solution (, w, v) of the following system:

t  + div(w) 2() = 0,
t (w) + div ((w 2()) w) ((() 2( () ())) div(w v))
2(1 )div(()D(w)) 2div(()A(w)) + p() = 2(1 )div(()v),
t (v) + div((w 2()) v) 2div(()v) + 2(( () ())div (w v))
= 2div(()t w). (28)

Remark 7. To see the link between the above system and the one from [37], one should take the main
velocity equal to w = u + 2() and the drift equal to v = 2().

Our idea of construction of solution is based on the following -entropy equality:

   
d |w|2 |v|2 d
 + (1 ) dx + e() dx + 2(1 ) ()|D(w) v|2 dx
dt 2 2 dt

 
+ 2 ()|A(w)|2 dx + 2(1 ) ( () ())|div(w v)|2 dx

  
()p ()
+2 ||2 dx = 0, (29)


that holds for any suciently smooth solution of (28). In order to build such a solution for > 0 given,
we need to go through several levels of approximations. For example, to build a solution of the nonlinear
parabolic equation for , some assumptions are required on the coecients. Two smoothing parameters
> 0 and > 0 (denoting standard mollication with respect to t and x) are therefore introduced in all
the transport terms, so that the approximate system can be rewritten as

t  + div([w] ) 2div ([ ()] ) = 0,

t (w) + div(([w] 2[ ()] ) w) ((() 2( () ())) div(w v))
2(1 )div(()D(w)) 2div(()A(w)) + 2s w div((1 + |w|2 )w) + p()
= 2(1 )div(()v),
t (v) + div(([w] 2[ ()] ) v)
2div(()v) + 2 (( () ())div (w v))
= 2div(()t w). (30)
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 811

Compared to [11], we have an extra term in the continuity equation because w is no longer divergence
free. In addition, a smoothing high-order derivative term 2s w with s 2, depending on small parameter
> 0 has to be introduced to control large spatial variations of w, because div w is not a priori bounded in
L1 (0, T ; L ())3 . Such bound will be required to be able to have bounds on the density. We will also need to
show that v = 2() at some point of the construction process and the second term in the regularization
process will be helpful.
In order to solve (30) for given , , > 0, the equations for the two velocities (w, v) are projected onto
nite dimensional spaces (Xn , Yn ) as is classically done when using FaedoGalerkin approximation. The rst
step is to prove global in time existence of solutions when n, , and are xed, then to pass to the limit
0 and after to let n . Next, we let 0 in order to prove that v = 2(). At the end, we combine
the equation satised by w and v to get an equation satised by u (multiply the equation satised by v by
and subtract from the equation for w). The last limit passage 0 is performed already for the system
written in terms of u which gives the weak formulation of the original system (1). This last step follows the
same lines as the proof presented by the two rst authors for the heat-conducting case, see [9]. It essentially
uses the presence of singular pressure, which allows to stabilize the system close to vacuum or the presence
of the turbulent drag term to have enough control on the velocity eld. More complicated construction in
the case of NavierStokes type model for compressible mixture with viscosity () =  can be found in the
PhD thesis of the third author [41], see also [32,33].

3.1. Existence of solutions for the full approximation

Below we present the basic level of approximation procedure.

1. The continuity equation is replaced by its regularized version

t  + div([w] ) div([ ()] ) = 0,

(0, x) = [0 ] , (31)

where , denote the standard regularizations with respect to time and space. Note that similar double
regularization has been recently used in [29] and discussed in the book by J.L. Vzquez [39].
2. The momentum equation is replaced by its FaedoGalerkin approximation with additional regularizing

term 2s w div((1 + |w|2 )w)]

 
w( ) dx (([w] 2[ ()] ) w) : dx dt
0
 
+ 2(1 ) ()D(w) : dx dt + 2 ()A(w) : dx dt
0 0

+ 2(1 ) ( () ()) div w div dx dt
0
 
2(1 ) ()v : dx dt 2(1 ) ( () ())div v div dx dt
0 0
 

p()div dx dt + s w s + (1 + |w|2 )w : dx dt
 0 0

= (w) dx
0
(32)

812 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

satised for any [0, T ] and any test function Xn , where Xn = span{i }ni=1 and {i }
i=1 is an
orthonormal basis in (W 1,2 ())3 with i (C ())3 for all i N .
3. The FaedoGalerkin approximation for the articial equation

 
v( ) dx (([w] 2[ ()] ) v) : dx dt
0
 
+ 2 ()v : dx dt + 2 ( () ())div v div dx dt
0 0
 
2 ( () ())div w div dx dt 2 ()t w : dx dt
0 0

= (v)0 dx (33)

satised for any [0, T ] and any test function Yn , where Yn = span{ i }ni=1 and {i }
i=1 is an
orthonormal basis in (W 1,2 ())3 with i (C ())3 for all i N .

Existence of solutions to the continuity equation. For xed w C([0, T ]; Xn ) we solve the continuity
equation, which is now quasi-linear parabolic equation with smooth coecients. Thus, application of classical
existence theory of Ladyzenskaja, Solonnikov and Uralceva [27] (see for example Theorem 10.24 from [17],
which is a combination of Theorems 7.2, 7.3 and 7.4 from [27]) yields the following result:

Theorem 3. Let (0, 1) and suppose that the initial condition 0 C 2+ () is such that 0 < r 0 R
and it satises the periodic boundary conditions. Then problem (31) possesses a unique classical solution 
from the class
 
 C([0, T ]; C 2+ ()) C 1 ([0, T ] ),
V[0,T ] = (34)
t  C /2 ([0, T ]; C()) ,

and satisfying classical maximum principle

0 < r (t, x) R. (35)

Moreover, the mapping w  (w) maps bounded sets in C([0, T ]; Xn ) into bounded sets in V[0,T ] and is
 
continuous with values in C [0, T ]; C 2+ () , 0 <  < < 1.

Local existence of solutions to the Galerkin approximations. Here we proceed as in the analogous proof
performed in [11] for w being the divergence-free vector eld. In what follows, we will show that the integral
equations (32) and (33) possess the unique solution on possibly short time interval via xed point argument.
More precisely, we will prove that there exists time T = T (n) and (w, v) C([0, T ]; Xn ) C([0, T ]; Yn )
satisfying (32), (33). To this purpose let us rewrite these equations as a xed point problem

t t
(w(t), v(t)) = M(t) PXn (w)0 + K(w)(s)ds , N(t) PYn (v)0 + L(v)(s)ds
0 0

= T [w, v](t) (36)

 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 813

where  = (w) is a solution to the continuity equation with w given, Xn is identied with Xn , so the
symbol , (Xn ,Xn ) denotes the action of a functional from Xn (= Xn ) on the element from Xn , similarly
for , (Yn ,Yn ) , and

M(t) : Xn Xn , M(t) [] dx = , (Xn ,Xn ) , , Xn ,


N(t) : Yn Yn , N(t) [] dx = , (Yn ,Yn ) , , Yn ,

PXn , PYn denote the projections of L2 () onto Xn , Yn , respectively, and K(w), L(v) are two operators
dened as follows:

K: Xn Xn ,
 

K(w), (Xn ,Xn ) = (([w] 2[ ()] ) w) : dx 2(1 ) ()D(w) : dx


2(1 ) ( () ()) div w div dx

 
2 ()A(w) : dx + 2(1 ) ()v : dx

 

+ 2(1 ) ( () ())div v div dx + p()div dx



w + (1 + |w| )w : dx,
s s 2

L: Yn Yn ,
 
L(v), (Yn ,Yn ) = (([w] 2[ ()] ) v) : dx 2 ()v : dx

 
2 ( () ())div v div dx + 2 ( () ())div w div dx


+2 ()t w : dx.

Since (t, x) is bounded from below by a positive constant, we have

1
M(t) L(Xn ,Xn ) , N(t) L(Yn ,Yn ) .
r
Moreover

M1 (t) M2 (t) L(Xn ,Xn ) + N1 (t) N1 (t) L(Yn ,Yn ) c(n, r1 , r2 )1 2 L1 () , (37)

and by the equivalence of norms on the nite dimensional space we prove that

K(w)Xn + L(v)Yn c(r, R, L2 () , wXn , vYn ). (38)

814 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

Next, we consider a ball B in the space C([0, ]; Xn ) C([0, ]; Yn ):

 
BM, = (w, v) C([0, ]; Xn ) C([0, ]; Yn ) : wC([0, ];Xn ) + vC([0, ];Yn ) M .

Using estimates (37), (38), (34) and (35), one can check that T is a continuous mapping of the ball BM,
into itself and for suciently small = T (n) it is a contraction. Therefore, it possesses a unique xed point
which is a solution to (32) and (33) for T = T (n).

Global existence of solutions. In order to extend the local in-time solution obtained above to the global
in time one, we need to nd uniform (in time) estimates, so that the above procedure can be iterated.
First let us note, that w, v obtained in the previous paragraph have better regularity with respect to time.
It follows by taking the time derivative of (36) and using the estimates (34), (35), that

(w, v) C 1 ([0, ]; Xn ) C 1 ([0, ]; Yn ).

This is an important feature since now we can take time derivatives of (32) and (33) and use the test
functions = w and = v, respectively. We then obtain
  
d |w|2 d
 dx + e() dx + 2(1 ) ()|D(w)|2 dx
dt 2 dt

 
+ 2 ()|A(w)|2 dx + 2(1 ) ( () ()) (div w)2 dx

 
2(1 ) ()v : w dx 2(1 ) ( () ())div v div w dx

   
()p ()
+ 2 ||2 dx + p()div ([w] w) dx



s 2 
+ | w| + (1 + |w|2 )|w|2 dx = 0, (39)

and
  
d |v|2
 dx + 2 ()|v|2 dx + 2 ( () ())(div v)2 dx
dt 2

 

2 ( () ())div w div v dx 2 ()t w : v dx = 0. (40)

Therefore, multiplying (40) by (1 ) and adding it to (39), we obtain

   
d |w|2 |v|2 d
 + (1 ) dx + e() dx + 2(1 ) ()|D(w) v|2 dx
dt 2 2 dt

 
+ 2(1 ) ( () ()) (div w div v)2 dx + 2 ()|A(w)|2 dx

 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 815

 
 ()p () s 2 
+ 2 ||2 dx + | w| + (1 + |w|2 )|w|2 dx



= p()div ([w] w) dx. (41)

Integrating the above estimate with respect to time, using the Hlder and the Gronwall inequalities,
we obtain uniform estimate for w and v necessary to repeat the procedure described in the previous
paragraph. Thus, we obtain a global in time unique solution (, w, v) satisfying Eqs. (31), (32), (33).

Uniform estimates. Below we present uniform estimates that will allow us to pass to the limit with and
n respectively. First observe that multiplying continuity equation (31) by  and integrating by parts with
respect to x gives
   
1 d
| |2 dx + 2 [ ( )] | |2 dx = div[w] | |2 dx div[w] L () | |2 dx.
2 dt

Integrating this equality with respect to time and using the uniform bound of div[w] in L1 (0, T ; L ())
provides the following estimates:

 L (0,T ;L2 ()) +  [ ( )]  L2 (0,T ;L2 ()) c. (42)

Moreover, the standard maximum principle gives boundedness of  from above and below. Indeed,
multiplying Eq. (31) by


= max(0, r  ) = min(0, R  ),
and +

respectively we obtain also using the bound on [w] that

0 < r  (t, x) R . (43)

Next, using (43) and integrating (41) with respect to time we see that for 0 < < 1 we have

w L (0,T ;L2 () + w L2 (0,T ;H 2s ()) + w L4 (0,T ;L4 ())

+ v L (0,T ;L2 ()) + v L2 (0,T ;H 1 ()) c. (44)

In the above estimate the constant c is uniform with respect to all approximation parameters except .

3.2. Passage to the limit with respect to and with respect to n

Passage to the limit 0. On a nite dimensional subspace all the norms are equivalent, therefore the
space compactness of w and v is automatic. In fact, for n xed we also know that t w is bounded in
L2 (0, T ; Xn ), thus up to the extraction of a subsequence, w w strongly in L2 (0, T ; Xn ) and the same
can be deduced for v . The biggest problem is thus to pass to the limit in the term

[ ( )]  w when 0 (45)

which requires the strong convergence of the density and the weak convergence of the gradient of density.
Observe that higher order estimates for  cannot be used here due to -regularization of the coecient
816 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

 ( ), however the information from (43) and (42) is enough. Indeed, using (43) and the assumption that
 () c > 0, estimate (42) implies that up to a subsequence

  weakly in L2 (0, T ; W 1,2 ()) when 0,

moreover t  L2 (0, T ; W 1,2 ()) and  L ((0, T ) ), thus the AubinLions lemma implies strong
convergence of 

  strongly in Lp ((0, T ) ) when 0, p < .

This justies the passage to the limit in (45). Therefore, one is able to pass to the limit 0 in both
velocity equations (32) and (33). The limit functions (, w, v) = (n , wn , vn ) satisfy the following system
of equations:

the momentum equation


t (n wn ) (t), (Xn ,Xn ) ((n [wn ] 2(n )) wn )(t) : dx

 
+ 2(1 ) (n )D(wn )(t) : dx + 2 (n )A(wn )(t) : dx


+ 2(1 ) ( (n )n (n )) div wn (t) div dx

 
2(1 ) (n )vn (t) : dx 2(1 ) ( (n )n (n ))div vn (t) div dx

 

p(n )(t)div dx + s wn (t) s + (1 + |wn |2 )wn (t) : dx = 0 (46)

satised for all Xn with t [0, T ],

the auxiliary equation for vn

t (n vn ) (t), (Yn ,Yn ) ((n [wn ] 2(n )) vn )(t) : dx

 
+ 2 (n )vn (t) : dx + 2 ( (n )n (n ))div vn (t) div dx

 
2 ( (n )n (n ))div wn (t) div dx 2 (n )t wn (t) : dx = 0 (47)

is satised for all Yn with t [0, T ].

However, so far we only know that the approximate continuity equation (31) is satised in the sense of
distributions

T T T
t n , (W 1,2 (),W 1,2 ()) dt n [wn ] dx dt + (n ) dx dt = 0 (48)
0 0 0
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 817

for any test function from L2 (0, T ; W 1,2 ()). But on the other hand, we know that in the sense of
distributions

= t n div ( (n )n ) = n div[wn ] [wn ] n L2 ((0, T ) ),

if wn L (0, T ; L ()). Indeed, note that we have

n div[wn ] L2 (0,T ;L2 ()) n L (0,T ;L ()) div[wn ] L2 (0,T ;L2 ()) , (49)
[wn ] n L2 (0,T ;L2 ()) [wn ] L (0,T ;L ()) n L2 (0,T ;L2 ()) , (50)

and the r.h.s. of the above inequalities is bounded at the level of Galerkin approximations, therefore also

 (n ) L2 ((0, T ) ).

Taking the product of and  (n ) we obtain


 (n ) (t n (n )) dx c(t) L1 ((0, T )),
2

and the above integral gives rise to estimates

  
 (n )(t n )2 dx + 2  (n )((n ))2 dx 2  (n )t n (n ) dx

  
d
=  (n )(t n )2 dx + 2  (n )((n ))2 dx + |(n )|2 dx c(t). (51)
dt

Note that we have passed to the limit with in the previous paragraph.
Note that this estimate asks for L ((0, T ) ) bound for wn , which is possible only at the level of
Galerkin approximation. Nevertheless, regularity (51) allows us to rewrite (31) as

t (n ) + div([wn ] (n )) 2 (n )(n ) + ( (n )n (n )) div[wn ] = 0. (52)

Multiplying the above equation by () and integrating by parts we obtain

 
d
|(n )| dx + 2
2
 (n )|(n )|2 dx
dt

 

= (n )n div[wn ] (n ) dx + [wn ] (n )(n ) dx.

Note that for any f H 2 () L () we have the following realization of the GagliardoNierenberg
inequality

f 2L4 () cf L2 () f L () , (53)

see for example [35]. Therefore, applying this inequality with f = (n ) and using the uniform bound for
wn L2 (0, T ; H 1 ()) and -dependent bound in L4 (0, T ; L4 ()) following from (44), we obtain

(n )L (0,T ;L2 ()) + 21/2 (n )L2 ((0,T )) c 0n H 1 () , R, , , (54)
818 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

and thus, coming back to (52) and using (43) we easily check the uniform (with respect to n and ) estimate

n L (0,T ;L2 ()) + n L2 ((0,T )) c. (55)

Passage to the limit n . As in the previous paragraph the biggest problem is to identify the limit of
the term

 (n )n wn , (56)

and in the convective term

n wn wn . (57)

Having obtained estimate (55) we can estimate the time-derivative of gradient of n . Indeed taking gradient
of (31) we obtain

t n = (n div[wn ] ) ([wn ] n ) (n ) L2 (0, T ; W 1,3/2 ()).

Note, however, that unlike in the analogous step of [11] the above estimate is no longer uniform with respect
to . Now, applying the AubinLions lemma for n we obtain

n  strongly in L2 (0, T ; L2 ()),

therefore due to (43) we also have

1 1
n  and strongly in Lp (0, T ; Lp ()) (58)
n 

for p < and

n wn w weakly in Lp1 (0, T ; Lq1 ()) Lp2 (0, T ; Lq2 ()), (59)

where p1 < 2, q1 < 6, p2 < , q2 < 2. These convergences justify the limit passage in (56).
To justify passage to the limit in (57) we rst estimate

 (n wn )  3 n L (0,T ;L2 ()) wn L2 (0,T ;L6 ())
L2 (0,T ;L 2 ())

+ wn L2 (0,T ;L2 ()) n L (0,T ;L ()) . (60)

We can also estimate the time derivative of momentum, from (46) we obtain
 T 
  
 
sup  t (n wn ) dx dt
1  
0
 T 
  

= sup  (( [w ] 2( )) w ) : dx dt 
 n n n n 
1  
0
 T   T 
     
   
+ 2(1 )  (n )D(wn ) : dx dt + 2  (n )A(wn ) : dx dt
   
0 0
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 819

 T 
  
 

+ 2(1 )  ( (n )n (n )) div wn div dx dt

 
0
 T   T 
     
   
+ 2(1 )  
(n )vn : dx dt +   p(n )div dx dt
   
0 0
 T 
  
 
+ 2(1 )  ( (n )n (n ))div vn div dx dt
 
0
 T   T 
     
    
+  (w ) dx dt + 
s s   (1 + |w| )w : dx dt ,
2
(61)
   
0 0

where the norm  denotes the norm in the space WT := L2 (0, T ; H 2s ()) L4 (0, T ; W 1,4 ()) with s 2.
Let us estimate the convective term
 T 
  
 
 ((n [wn ] 2 (n )n ) wn ) : dx dt


 
0
T 
L6 () Rwn 2 12 + c(, R)n L6 () wn  3 dt
L 5 () L2 ()
0
c(, R, )L2 (0,T ;W 2,2 ()) ,

that is bounded for s 2. For the highest order terms we have

 T 
  
 
 s wn s dx dt wn L2 (0,T ;H 2s ()) L2 (0,T ;H 2s ()),
 
0

and
 T 
   T  
 

 2 
(1 + |w| )w : dx dt L4 () wn 3L4 () + wn  43 dt
  L ()
0 0
 
L4 (0,T ;L4 ()) w3L4 (0,T ;L4 ()) + w 43 4 .
L (0,T ;L 3 ())

The other terms from (61) are less restrictive, therefore

t (n wn ) WT c, (62)

where WT denotes the dual space of WT dened above. Collecting (59), (60), (62) and applying the
AubinLions lemma to n wn , we prove that

n wn w strongly in Lp (0, T ; Lp ())

for some p > 1 and therefore thanks to (58) and (44)

wn w strongly in Lp (0, T ; Lp ()) (63)

for 1 p < 4. In particular, convergence of (57) is proved.

820 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

For future purposes we now estimate the time derivative of v. We use (47) to obtain
 T 
  
 
sup  t (n vn ) dx dt
1  
0
 T 
  

= sup  ((n [wn ] 2 (n )n ) vn ) : dx dt
1  
0
 T   T 
     
   
+ 2  (n )vn : dx dt + 2  ( (n )n (n ))div vn div dx dt
   
0 0
 T   T 
     
   
+ 2  ( (n )n (n ))div wn div dx dt + 2  (n )t wn : dx dt
   
0 0

for L4 (0, T ; W 1,4 ()). We will only estimate the convective term since it is most restrictive
 T 
  
 
 ((n [wn ] 2 (n )n ) vn ) : dx dt


 
0

T

L4 () Rwn L4 () vn L2 () + c(, R)n L4 () vn L2 () dt
0
 1

c(, R)L4 (0,T ;W 1,4 ()) vn L2 (0,T ;L2 ()) wn L4 (0,T ;L4 ()) + n L2 2 (0,T ;L2 ())

thus

t (n vn )  4 4 c. (64)
L 3 (0,T ;W 1, 3 ())

Hence, the limit functions (, w, v) = ( , w , v ) fulll

the continuity equation

t  + div ( [w ] ) 2( ) = 0 (65)

a.e. in (0, T ) ,
the momentum equation


t ( w ) , (W ,W ) (( [w ] 2( )) w ) : dx dt
0
 
+ 2(1 ) ( )D(w ) : dx dt + 2 ( )A(w ) : dx dt
0 0

+ 2(1 ) ( ( ) ( )) div w div dx dt
0
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 821


2(1 ) ( )v : dx dt
0

2(1 ) ( ( ) ( ))div v div dx dt
0
 

p( )div dx dt + s w s + (1 + |w |2 )w : dx dt = 0 (66)
0 0

for all W with [0, T ],

the auxiliary equation for v

t ( v ) , 4 4
(L 3 (0, ;W 1, 3 ()),L4 (0, ;W 1,4 ()))
 
(( [w ] 2( )) v ) : dx dt + 2 ( )v : dx dt
0 0

+ 2 ( ( ) ( ))div v div dx dt
0
 

2 ( ( ) ( ))div w div dx dt 2 ( )t w : dx dt = 0, (67)
0 0

for L4 (0, ; W 1,4 ()), [0, T ].

3.3. Passage to the limit tends to 0 and identication of v with 2()

The aim of this paragraph is to let 0 in Eqs. (65), (66) and (67). This limit passage can be performed
exactly as n presented above. The only dierence is that after this step we may drop the additional
equation for v thanks to identication v = 2(). Below we present the details of this reasoning.
Note that the coecients of the quasi-linear parabolic equation (65) (i.e. [w ] ) are suciently regular
and the maximum principle (43) holds uniformly with respect to all approximation parameters except .
Therefore, the classical theory of Ladyzenskaja, Solonnikov and Uralceva [27] (Theorems 7.2, 7.3 and 7.4
from [27]) can be applied to show further regularity of  , we have in particular

t  C([0, T ]; C()),  C([0, T ]; C 2 ()). (68)

Let us now rewrite (65) using (22) as

t  + div ( ([w ] 2 ( ))) = 0

and therefore multiplying the above equation by  () we obtain

t ( ) + div (( )([w ] 2 ( ))) + ( ( ) ( )) (div[w ] 2( )) = 0.

Dierentiating it with respect to space one gets in the sense of distributions

t ( v ) + div( ([w ] 2 ( )) v ) + 2 (( ( ) ( )) (div[w ] div v ))

+ 2div((( )t [w ] ) 2div(( )v ) = 0 (69)
822 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

where by v we denoted 2( ). Note that due to particular case of the GagliardoNirenberg interpolation
inequality (53) and (43) we know that  is bounded

 L4 (0,T ;L4 ()) c,

uniformly with respect to . One can thus estimate the convective term of (69) in L2 (0, T ; W 1,2 ())
uniformly with respect to . Indeed, we now w is uniformly bounded in L4 (0, T ; L4 ()) with respect to
and therefore
 T 
  
 
sup  ( ([w ] 2 ( )) v ) : dx dt
1  
0

c(R)L2 (0,T ;L2 ())  L4 (0,T ;L4 ()) [w ] L4 (0,T ;L4 ()) +  L4 (0,T ;L4 ())

for L2 (0, T ; W 1,2 ()) (uniformly with respect to ), which justies that

T
t ( v ), (L2 (0,T ;W 1,2 ()),L2 (0,T ;W 1,2 ()))  ([w ] 2 ( )) v : dx dt
0

T T

2 ( ( ) ( ))div[w ] div dx dt + 2 ( ( ) ( ))div v div dx dt
0 0

T T
2 ( )t [w ] : dx dt + 2 ( )v : dx dt = 0 (70)
0 0

is satised for any L2 (0, T ; W 1,2 ()).

Remark 8. As we noticed in (68) the regularity of  is in fact much higher and allows to formulate the
equation for v in much stronger sense than merely (70). This formulation, however, will be used when
passing to the limit with respect to after passing to the limit with respect to .

We now want show that v v tends to 0 when goes to zero in an appropriate norm. To this purpose
let us expand
 
d |v v |2
I=  dx + 2 ( )|(v v )|2 dx
dt 2


+ 2 ( ( ) ( ))(div(v v ))2 dx

 
d |v |2 |v |2
=  v v + dx
dt 2 2



+ 2 ( ) |v |2 + |v |2 2v v dx



+ 2 ( ( ) ( )) |div v |2 + |div v |2 2div v div v dx. (71)

 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 823

To handle the rst term, let us notice that letting n in (40), using the lower semi-continuity of the
convex functions and the strong convergence of wn established in (63) we obtain

  
d |v |2
 dx + 2 ( )|v |2 dx + 2 ( ( ) ( ))(div v )2 dx
dt 2

 

2 ( ( ) ( ))div w div v dx 2 ( )t w : v dx 0. (72)

Now, the last term in (71) can be computed using = v in (70), we have

  
d |v |2
 dx + 2 ( )|v | dx + 2
2
( ( ) ( ))(div v )2 dx
dt 2

 
2 ( )t [w ] : v dx 2 ( ( ) ( ))div w div v dx = 0.

The middle term in (71) equals

 
d
 v v dx = (t ( v ) v + v t ( v ) t  v v ) dx (73)
dt

and the two rst terms make sense and can be handled using = v in (67) and = v in (70). Note that v
and t  are due to (68) regular enough to justify the integrability of the last term in (73) and we can write


t  v v dx

 
= ( ([w ] 2 ( ))) v : v dx + ( ([w ] 2 ( ))) v : v dx.

Therefore, after summing all expressions together and after some manipulation, we can show that

 
I 2 ( ( ) ( ))div[w ] div v dx + 2 ( ( ) ( ))div w div v dx

 

+2 ( ( ) ( ))div[w ] div v dx 2 ( ( ) ( ))div w div v dx

 
( )t [w ] : v dx ( )t w : v dx

 
+ ( )t w : v dx + ( )t [w ] : v dx 0,

in particular
824 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836


I ()t ([w ] w ) : (v v ) dx


+2 ( ( ) ( ))div ([w ] w ) div (v v ) dx.

Note that the r.h.s. of this inequality tends to 0 when 0. Indeed, we can bound (v v )
in L2 (0, T ; L2 ()) uniformly with respect to and [w ] w strongly in Lp (0, T ; W 1,p ()) for p < 4.
Therefore, using (71), we conclude that v v converges to zero in L (0, T ; L2 ()) L2 (0, T ; H 1 ())
when 0.
The limit functions (, w) = ( , w ) fulll

the continuity equation

t  + div ( w ) ( ) = 0 (74)

a.e. in (0, T ) ,
the momentum equation


t ( w ) , (W ,W ) (( w 2( )) w ) : dx dt
0

 T
+ 2(1 ) ( )D(w ) : dx dt + 2 ( )A(w ) : dx dt
0 0

+ 2(1 ) ( ( ) ( )) div w div dx dt
0

4(1 ) ( )2 ( ) : dx dt
0

4(1 ) ( ( ) ( ))( ) div dx dt
0
 

p( )div dx dt + s w s + (1 + |w |2 )w : dx dt = 0 (75)
0 0

for W , [0, T ],
the auxiliary equation for ( )

t ( ), (L2 (0, ;W 1,2 ()),L2 (0, ;W 1,2 ()))

 
(( w 2( )) ( )) : dx dt + 2 ( ) : dx dt
0 0

+ 2 ( ( ) ( ))( ) div dx dt
0
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 825

 

( ( ) ( ))div w div dx dt ( )t w : dx dt = 0 (76)
0 0

for all L2 (0, ; W 1,2 ()) with [0, T ].

3.4. Recovering the global weak -entropy solution

Due to the identication, we may now dene a new quantity

u = w 2(). (77)

Compared to [11], we cannot pass to the limit in the equations satised by w and v. We have to combine
them to deal with an equation on u. More precisely, multiplying (33) by and substracting to (32) we
obtain the weak formulation on u

 
t ( u ), (W ,W ) ( u u ) : dx dt + 2 ( )D(u ) : dx dt
0 0
 
+ ( )div u div dx dt p( )div dx dt
0 0
 
+ s w s dx dt + (1 + |w |2 )w dx dt = 0 (78)
0 0

satised for all W with [0, T ].

Passing to the limit in (78) with respect to follows the lines introduced by [7] using the uniform
estimates given by the regularized -entropy. See also [41] for details regarding the barotropic case with
chemical species. Now, after passing to the limit in the energy estimate (41) (with all parameters 0,
n , 0 in the same way), we obtain
   
d |w |2 |2( )|2 d
 + (1 ) dx +  e( ) dx + 2(1 ) ( )|D(u )|2 dx
dt 2 2 dt

 
+ 2 ( )|A(w )|2 dx + 2(1 ) ( ( ) ( ))(div u )2 dx

  
 ( )p ( )
+ 2 | |2 dx + |s w |2 dx + (1 + |w |2 )|w |2 dx 0, (79)


where we essentially used the regularization [2s w div((1 + |w |2 )w ). We then show that the limit
when 0 gives a solution (, u) which satises the following energy inequality

  
T
|w|2 |2()|2
sup  + (1 ) (t) dx + e()(t) dx + 2(1 ) ()|D(u)|2 dx dt
t[0,T ] 2 2
0

T T
+ 2 ()|A(w)|2 dx dt + 2(1 ) ( () ())(div u)2 dx dt
0 0
826 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

T   
 ()p () |w0 |2 |2(0 )|2
+ 2 ||2 dx dt 0
+ (1 ) 0 0
+  e( ) dx. (80)
 2 2
0

For the limit passage in the corresponding equations we use the density estimates due to the singular
pressure, as it was introduced in the work by the rst two authors and generalized to chemically reacting
mixture case by the third author and collaborators. In particular, one needs to ensure that
1
 2 u Ls (0, T ; Lr ())

with r, s > 2. Here assumptions of n, m and (13), (14) play an important role. Indeed, assuming for the
moment that  2 u L (0, T ; L2 ()),  L (0, T ; Lp ()) and u Lq1 (0, T ; Lq2 ()), we can write exactly
1

as in the proof of Lemma 6.1 from [9] that

 2 |u| =  2  |u|2 |u|12

1 1

for some 0 12 . Therefore

1 1
1
 2 uLs (0,T ;Lr ()) L2 (0,T ;Lp ())  2 u2
L (0,T ;L2 ()) uLq1 0,T ;Lq2 ()
12

with

1 1 2 1 1 1 1 2
= , = ++ . (81)
s q1 r p 2 q2

It follows from the -entropy estimate (80) that the best p we can take is equal

p = 6m 3.

Let us now determine q1 , q2 , we rst check that the -entropy estimate (80) gives us uniform bound on

T
 ()p ()
||2 dx dt.

0

In particular, for  <  , we obtain


+1n
 2 L2 (0, T ; L2 ())

which controls the negative powers of  close to vacuum. We can use this estimate to determine q1 and
some q3 , we have
 n
uLq1 (0,T ;Lq3 ()) c 1 +  2 L2j (0,T ;L6j ())  uL2 (0,T ;L2 ())
n

2
 
+1n n
c 1 +   2 L2 (0,T ;L6 ())  uL2 (0,T ;L2 ()) , (82)
2

where

+ 1 n 2j n 1 1 1
j= , q1 = =2 1 , = + . (83)
n j+1 +1 q3 6j 2
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 827

Thus in (81) we can take

3q3
q2 = = 3q1
3 q3

provided that

+ 1
q1 > 1, i.e. n < .
2

Inserting q2 = 3q1 and p = 6m 3 to (81) and using (83) we obtain r, s > 2 provided that in addition we
assume that

2n(3m 2) 3
> 1, m> ,
4m 3 4

but there is no restriction on + > 1.

4. NavierStokes equations with drag terms

Let us recall the compressible NavierStokes system with quadratic turbulent drag force

t  + div (u) = 0,
(84)
t (u) + div(u u) div (2()D(u)) (()div u) + r1 |u|u + p() = 0,

with the usual pressure law p() = a with > 1. The augmented regularized system reads

t  + div([w] ) 2div ([ ()] ) = 0,

t (w) + div(([w] 2[ ()] ) w) ((() 2( () ())) div(w v))
2(1 )div(()D(w)) 2div(()A(w))
+ 2s w div((1 + |w|2 )w) + r1 |w 2()|(w 2()) + p()
= 2(1 )div(()v),
t (v) + div(([w] 2[ ()] ) v)
2div(()v) + 2 (( () ())div (w v))
= 2div(()t w). (85)

The additional drag term r1 |w 2()|(w 2()) will not really aect the -entropy. More precisely
the contribution to the -entropy estimate may be written as

|w ()|(w ()) w dx

 
= |w ()|3 dx + |w ()|(w ()) () dx.

Note that the second term of the above right hand side rewrites as:
828 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836


|w 2()|(w 2()) ()


= ()|w 2()|div(w 2()) dx


w 2()
() (w 2()) (w 2()) dx
|w 2()|

 
uk
= ()|u|div u dx () uj j uk dx,
|u|

where u is dened by u = w 2(). One may write the estimate

 
   () 1/3
()|u||D(u)| dx  ()D(u)L2 ()  ()uL2 ()  ()D(u)L2 ()   uL2 () .
1/3

Now we use the hypothesis made on viscosities. We can bound

 
1  r1 ()3
2r1 ()|u||D(u)| dx  ()D(u)2L2 () + |u|3 L1 () + c(r1 ) dx,
2 3 2

and using (19) and (20)

  
()3 ()3
dx C + 1{(t,x):(t,x)1} dx C + c(r1 ) e() dx
2 2

with c(r1 ) 0 when r1 tends to zero. Thus the -entropy is not perturbed and we get uniform estimate on
|u|3 with respect to in L1 (0, T ; L1 ()). It is enough to conclude as in [10] because the extra estimate we
get replaces the MelletVasseur quantity involved in [31].

5. Hypocoercivity revisited on linearized compressible NavierStokes

In this section, we want to show to readers who are not familiar with compressible NavierStokes equations
why our -entropy equality may be seen as a nonlinear version of the so-called hypocoercivity property which
is strongly used in the framework of strong solutions. The interested readers are referred to [40] and [1] for
more general discussions around hypocoercivity and to [14] for deep mathematical results on compressible
NavierStokes equations in critical spaces.

5.1. Linearized barotropic system

In this subsection we consider the barotropic compressible NavierStokes system. Linearization around
the state (0 , u0 ) = (1, 0) on the barotropic system gives

t  + div u = 0,
2 (86)
t u +  2div(D(u)) + div u = 0
d
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 829

with D(u) = (u + t u)/2. The chosen coecient is the borderline case that means the one such that
+ 2/d = 0. We focus on the periodic domain = Td . The standard energy (multiplying the rst equation
of (86) by  and the second equation by u and summing them up) reads
   2
1 d 2   
|u| + ||2 dx + D(u) div u I dx = 0. (87)
2 dt  d 

Let be a constant that will be determined later on, then v = u + 2 satises the equation

2
t v +  u 1 2 div u = 0,
d

and therefore (multiplying this equation by v, the rst mass equation by  and summing them up)
    
1 d  2
|v| + ||
2 2
dx + 2 || + 2
|u| dx + 1 2
2
|div u|2 dx
2 dt d

 
1
4 1
2
div u  dx = 0.
d

This equality may be rewritten (using that = div curl curl) as

  
1 d 
|v|2 + ||2 dx + 2 ||2 dx + 2 |A(u)|2 dx
2 dt

   
1 1
+ 2 1 |div u|2 dx 42 1 div u  dx = 0, (88)
d d

where A(u) = (u (u)t )/2. Testing the gradient of the mass equation by , we get
 
1 d
|| dx +
2
 div u dx = 0.
2 dt


Assume now that 0 < < (d 1)/d, then multiplying the last relation by 42 1 1
d and adding to
(88) we get
  
1 d  1 d
|v| + ||
2 2
dx + 2 1
2
||2 dx
2 dt d dt

   
1
+ 2 ||2 dx + 2 |A(u)|2 dx + 2 1 |div u|2 dx = 0. (89)
d

Remark that this equality is the analog, deleting the term 1/d, of the -entropy we found in the nonlinear
framework: Note that
   
|A(u)|2 dx + (1 ) |div u|2 dx = [A(u)|2 dx + (1 ) |D(u)|2 dx.

In particular, this -entropy (89) provides the exponential decay in time to (0, 0) of (, u) in the L2 ()
norm if the initial perturbation (0 , u0 ) is in L2 ().
830 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

Let us now look at exponential decay in H 1 norm. Taking the curl of the momentum equation in (86)
and testing against curl u, we get
 
1 d
| curl u|2 dx + | curl u|2 dx = 0. (90)
2 dt

Taking the div of the momentum equation and testing it against div u
  
1 d
|div u| dx + (2 + )
2
|div u| dx +
2
 div u dx = 0.
2 dt

Adding this equation to the mass equation tested against , we obtain

 
1d 
|div u|2 + ||2 dx + (2 + ) |div u|2 dx = 0. (91)
2d

Thus (87), (89), (90), (91) provide an exponential decay in time for (, u)H 1 () to 0 assuming initial

perturbation (0 , u0 ) in H 1 (). Note in particular, that coecient can be arbitrary small.

5.2. Linearized heat-conducting compressible NavierStokes equations

In this subsection, we consider the linearized compressible NavierStokes equation with heat conductivity
around (1, 0, 1) namely


t  + div u = 0,

2
t u +  + 2div(D(u)) + div u = 0, (92)


d
2
t + div u K = 0,
d

with the periodic boundary conditions. Such system may be found for instance in [40, p. 51]. For this system,
we easily check that there exists an energy, dened as
    2 
1 d d  div u  Kd
|u| + || + ||2
2 2 
dx + 2 D(u) I dx + ||2 dx = 0. (93)
2 dt 2 d  2

Dierentiating the mass equation and the temperature equation with respect to space, we nd the following
equations:


t  + div u = 0,

2

t u +  + 2div(D(u)) + div u = 0,


d
2
t + div u = 0, (94)

d


t  + div u = 0,



t + 2 div u = 0.
d

Thus we can write a new equation satised by articial velocity u + 2


2
t (u + 2) +  + u 1 2 div u = 0.
d
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 831

Testing it against u + 2, we obtain

    
1 d  2
|u + 2| +  2 2
dx + 2 || dx + 2
|u| dx + 1 2
2
|div u|2 dx
2 dt d

   
1
42 1 div u  dx + u dx + 2s  dx = 0. (95)
d

Thus as in the barotropic case, we have

   
1 d  1
|u + 2| +  2 2
dx + 2 || dx + 2
2
|A(u)| dx + 2(1 )
2
|div u|2 dx
2 dt d

   
1
42 1 div u  dx + u dx + 2  dx = 0. (96)
d

Recall that for , we have

 
1 d
|| dx +
2
div u  dx = 0. (97)
2 dt


1
Multiplying this relation by 4 1 and adding to the previous one, we get
2
d
   
1 d 1
|u + 2| +  + 4 1 ||
2 2 2
dx + 2 ||2
2
2 dt d

    
1
+ 2 |A(u)| dx + 2 1
2
|div u|2 dx + u dx + 2  dx = 0. (98)
d

Recall that
  
1 d d 2 Kd
|| dx + ||2 dx + div u dx = 0, (99)
2 dt 2 2

which, when added to (98), yields

  
1 d 1 d
|u + 2|2 + 2 + 42 1 ||2 + ||2 dx
2 dt d 2

 
1
+ 2 |A(u)|2 dx + 2(1 ) |div u|2 dx
d

  
dK
+ || dx + ( 2
) || dx +
2
|( + )|2 dx = 0. (100)
2

Choosing such that 0 < < dK/2 and 0 < < (d 1)/d, the -entropy balance (100) gives
the exponential decay of (, u, ) in the L2 norm, note in particular an interesting interplay between
conductivity and pressure.
832 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

Let us now focus on exponential decay in H 1 norms. Let us take the equation satised by div u, test it
against div u and integrate, we get
   
1 d
|div u|2 dx +  div u dx + div u dx + (2 + ) |div u|2 dx = 0. (101)
2 dt

Let us now take the equation satised by and test it against , we get
  
1 d 2
||2 dx div u dx + K ||2 dx = 0. (102)
2 dt d

Recall that
 
1 d
|| dx
2
div u  dx = 0. (103)
2 dt

Thus adding (101), (102) to (103) gives

   
1 d d dK
|div u|2 + ||2 + ||2 dx + ||2 dx + (2 + ) |div u|2 dx = 0. (104)
2 dt 2 2

Recalling that the curl u satises (90), combining it with (104) and the -entropy equality (100), we get

the exponential decay in time of  (, u, )H 1 () if the initial perturbation (0 , u0 , 0 ) is uniformly
bounded in H 1 ().

6. Heat-conducting NavierStokes equations with -energy

In this section, we present the equations of motions for the heat-conducting uid written in terms of
the two velocities u and u + 2() with corresponding densities (1 ) and . We do not aim at
proving the existence result for such system but on showing that the two-velocity hydrodynamics in the
spirit of the work by S.M. Shugrin [37] is consistent with the study performed for the low Mach number
system in the rst part of this diptych in [11]. More precisely, we will show that the formal low-Mach
number limit for the two-velocities system gives the augmented system used in [11] to construct the
approximate solution. An important observation is that the system presented below does not coincide with
the usual heat-conducting compressible NavierStokes equations. Indeed, the two-velocities description of
the dynamics of the uid leads to dierent energy equation with a generalized temperature, called the
-temperature. However, this is not a priori the usual temperature, unless the system reduces to the angle
velocity one (i.e. the density  is equal to 0). This property was also explained in the works [37] and [20],
where the authors discuss the capillary-temperature.
We assume that is a periodic box in R3 , i.e. = T3 , and we consider the following two-velocity system


t  + div(u) = 0,



t (u) + div(u u) div(2()D(u)) (()divu) + p(, e ) = 0,



t ((u + 2()) + div(u (u + 2())) div(2()A(u)) + p(, e ) = 0, (105)


t E + div(E u) + div(p [(1 )u + (u + 2()]) + div Q ,


 


div (1 )S1 (u) + S2 (u + 2()) = 0,

where we denoted D(u) = 1

2 (u + t u) and A(u) = 1
2 (u t u).
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 833

The viscosity coecients (), () satisfy the BreschDesjardins relation

() = 2 () 2(). (106)

The total -energy E is dened as follows:

 (1 ) 2 
E =  e + |u| + |u + 2()|2 . (107)
2 2

Remark 9. Note that E u is expressed as a sum of two energies

" # " #
|u|2 |u + 2()|2
(1 )u e + + u e +
2 2

similarly to energy from [37].

Remark 10. Integrating the total -energy equation with respect to space, we obtain

d
E dx = 0.
dt

Thus (107) and the identity

(1 ) 2 |u + 2()|2 |2()|2
|u| + |u + 2()|2 = + (1 ) (108)
2 2 2 2

yields the following conservation property

  
d 1 (1 )
 e + |u + ()|2 + |()|2 dx = 0.
dt 2 2

This quantity may be treated as a generalization of the -entropy, found for the barotropic case, to the
heat-conducting case.

The viscous tensors S1 , S2 are given by

S1 = 2()D(u) + ()div u I

and

S2 = 2()A(u).

The heat ux Q is given by standard Fouriers law, i.e.

Q = K ,

where K is the positive heat-conductivity coecient and denotes the generalized temperature
(the -temperature). Let us consider an ideal polytropic gas, namely

p(, e ) = r + pc (), e = Cv + ec (),

834 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

where r and Cv are two constant positive coecients, see for instance [9]. For convenience, we denote
= 1 + r/Cv . Moreover, the additional pressure and internal energy, pc and ec respectively, are associated
to the zero Kelvin isothermal, which roughly speaking means that

lim pc () = .
0+

Further, we require that ec is a C 2 (0, ) nonnegative function and that the following constraint is satised

dec
pc () = 2 ().
d

Below we present two dierent formulations of the internal energy equation which lead to useful bounds
on -temperature similarly as in [9] for the usual temperature.
The rst formulation reads
 
Cv t ( ) + div(u ) +  div w

= 2(1 )()|D(u)|2 + 2()|A(u)|2 + 2(1 )( () ())|div u|2 + div(K )

with the Gruneisen parameter and where the mixing temperature becomes the usual temperature if
= 0. Note that for 0 1, the -temperature remains non-negative in view of the maximum principle.
The second formulation is based on the notion of generalized -entropy s. It is the usual entropy in
which the standard temperature has been replaced by the -temperature, i.e.

s = Cv log r log ,

thus, when , are suciently regular we can derive the following equation

t (s ) + div(us ) div (K log )

()|D(u)|2 ()|A(u)|2 ( () ())|div u|2
= 2(1 ) + 2 + 2(1 )

| |2
2() + K . (109)
2

Note that, recalling the relation

 2
 1  1
|D(u)| = D(u) divu I + |divu|2 ,
2 
3 3

the terms on right-hand side, when integrated over space, give nonnegative contribution using the assumption
on 3() + 2() and () = 2( () ()). Indeed, it suces to check that for the penultimate term we
have
 
 ()
dx =  ()||2 dx 0.

Using all this information, it could be possible to prove global existence of -entropy solution for the
heat-conducting compressible NavierStokes system under analogous assumptions as in [9], replacing the
usual temperature by the -temperature. The existence of the approximate solution could be then proven
by using the augmented system written in terms of w = u + 2() and v = 2() as it was done
in [37], or in Section 3 addressing barotropic ows
 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836 835

t  + div([w] ) div ([ ()] ) = 0,

t (w) + div(([w] [ ()] ) w) ((() ( () ())) div(w v))
(2 )div(()D(w)) div(()A(w)) + 2s w div((1 + |w|2 )w) + p(, e )
= (2 )div(()v),
t (v) + div((w [ ()] ) v) div(()v) + (( () ())div (w v))
= div(()t w) (110)

with the -total energy supplemented by the correction corresponding to the regularization of the
momentum

t (E ) + div( [w] [ ()]  E ) + div(p w) + divQ
 
div S1 w + (2 ) S2 v + s w w div((1 + |w|2 )|w|2 = 0,

and the set of initial conditions. Above, the total -energy E is dened as

1 (2 ) 2
E = e + |w|2 + |v| .
2 2
Note, however, this construction would not lead to the usual heat-conducting compressible NavierStokes
system in the limit 0. Indeed, the dierence is again due to -temperature that is not the usual one.
But, performing a formal low Mach number limit for this system, we would get p = 1, div w = 0 (comparing
terms of the same order). In the equation on w, being now incompressible, the pressure gradient p would
be replaced by Lagrangian multiplier . As a result, we would get the augmented system dened in [11]
in the part devoted to construction of solution.

Acknowledgements

The authors want to thank the referee for his/her valuable comments which allows to improve the quality
of the paper. The rst author acknowledges support from the ANR-13-BS01-0003-01 project DYFICOLTI.
The authors want to thank S.L. Gavrilyuk for really interesting comments and for pointing to them
the two interesting papers linked to media with equations of state that depend on derivatives. The third
author acknowledges the Post-Doctoral support of Ecole Polytechnique, she was also supported by MS grant
IdPlus2011/000661 and by the fellowship START of the Foundation For Polish Science.

References

[1] K. Beauchard, E. Zuazua, Large time asymptotics for partially dissipative hyperbolic systems, Arch. Ration. Mech. Anal.
199 (1) (2011) 177227.
[2] S. Benzoni, R. Danchin, S. Descombes, On the well-posedness of the EulerKorteweg model in several space dimensions,
Indiana Univ. Math. J. 56 (2007) 14991579.
[3] F. Bthuel, R. Danchin, P. Gravejat, J.-C. Saut, D. Smets, Les quations dEuler, des ondes et de Kortewegde Vries
comme limites asymptotique de lquation de GrossPitaevskii, in: Sminaire EDP (20082009), Expos no 1.
[4] D. Bresch, F. Couderc, P. Noble, J.-P. Vila, New extended formulations of EulerKorteweg equations based on a
generalization of the quantum Bohm identity, arXiv:1503.08678, 2015.
[5] D. Bresch, B. Desjardins, Quelques modles diusifs capillaires de type Korteweg, C. R. Acad. Sci. Paris, Mc. 332 (11)
(2004) 881886.
[6] D. Bresch, B. Desjardins, C.K. Lin, On some compressible uid models: Korteweg, lubrication and shallow water systems,
Commun. Partial Dier. Equ. 28 (2003) 10091037.
[7] D. Bresch, B. Desjardins, Existence of global weak solutions for a 2D viscous shallow water equations and convergence to
the quasi-geostrophic model, Commun. Math. Phys. 238 (12) (2003) 211223.
[8] D. Bresch, B. Desjardins, On the construction of approximate solutions for the 2D viscous shallow water model and for
compressible NavierStokes models, J. Math. Pures Appl. (9) 86 (2006) 362368.
836 D. Bresch et al. / J. Math. Pures Appl. 104 (2015) 801836

[9] D. Bresch, B. Desjardins, On the existence of global weak solutions to the NavierStokes equations for viscous compressible
and heat conducting uids, J. Math. Pures Appl. (9) 87 (1) (2007) 5790.
[10] D. Bresch, B. Desjardins, D. Grard-Varet, On compressible NavierStokes equations with density dependent viscosities
in bounded domains, J. Math. Pures Appl. (9) 87 (2) (2007) 227235.
[11] D. Bresch, V. Giovangigli, E. Zatorska, Two-velocity hydrodynamics in uid mechanics: Part I. Well posedness for zero
Mach number systems, J. Math. Pures Appl. 104 (4) (2015) 762800.
[12] D. Bresch, P. Noble, J.-P. Vila, P. Villedieu, Consistent numerical schemes for compressible ows, 2015, in preparation.
[13] J.M. Ball, G.-Q.G. Chen, Entropy and convexity for nonlinear partial dierential equations, theme issue 2005,
Philos. Trans. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 371 (2005) (2013).
[14] R. Danchin, Fourier analysis methods for compressible ows, Panor. Synth. Soc. Math. Fr., 2014, submitted for
publication.
[15] J. Dong, A note on barotropic compressible quantum NavierStokes equations, Nonlinear Anal., Real World Appl. 73
(2010) 854856.
[16] A. Drici, B. Haspot, Remarks on global controllability for the shallow-water system with two control forces,
arXiv:1202.0910, 2012.
[17] E. Feireisl, A. Novotn, Singular Limits in Thermodynamics of Viscous Fluids, Adv. Math. Fluid Mech., Birkhuser Verlag,
Basel, 2009.
[18] E. Feireisl, Y. Lu, J. Mlek, On PDE analysis of ows of quasi-incompressible uids, preprint MORE/2014/18, 2014.
[19] E. Feireisl, A. Vasseur, New perspectives in uid dynamics: mathematical analysis of a model proposed by Howard
Brenner, in: New Directions in Mathematical Fluid Mechanics, in: Adv. Math. Fluid Mech., Birkhuser Verlag, Basel,
2010, pp. 153179.
[20] S.L. Gavrilyuk, S.M. Shugrin, Media with equations of state that depend on derivatives, J. Appl. Mech. Tech. Phys. 37 (2)
(1996).
[21] M. Gisclon, I. Violet, About the barotropic compressible quantum NavierStokes equations, 2014, submitted for
publication, arXiv:1412.1332.
[22] B. Haspot, New formulation of the compressible NavierStokes equations and parabolicity of the density,
HAL Id: hal-01081580, 2014.
[23] B. Haspot, New entropy for Kortewegs system, existence of global weak solutions and ProdiSerrin theorem, 2014,
submitted for publication, arXiv:1102.5436.
[24] F. Jiang, A remark on weak solutions to the barotropic compressible quantum NavierStokes equations, Nonlinear Anal.,
Real World Appl. 12 (2011) 17331735.
[25] A. Jengel, Global weak solutions to compressible NavierStokes equations for quantum uids, SIAM J. Math. Anal. 42
(2010) 10251045.
[26] G. Kitavsev, P. Laurenot, B. Niethammer, Weak solutions to lubrication equations in the presence of strong slippage,
Methods Appl. Anal. 18 (2) (2011) 183202.
[27] O.A. Ladyzenskaja, V.A. Solonnikov, N.N. Uralceva, Linear and Quasilinear Equations of Parabolic Type, Transl. Math.
Monogr., vol. 23, American Mathematical Society, Providence, RI, 1967, translated from Russian by S. Smith.
[28] C.D. Levermore, W. Sun, K. Trivisa, A low Mach number limit of a dispersive NavierStokes system, SIAM J. Math. Anal.
44 (3) (2012) 17601807.
[29] X. Liao, A global existence result for a zero Mach number system, J. Math. Fluid Mech. 16 (1) (2014) 77103.
[30] M. asica, On steady solutions to vacuum less Newtonian models of compressible ow, Nonlinearity 27 (2014) 26632687.
[31] A. Mellet, A. Vasseur, On the barotropic compressible NavierStokes equations, Commun. Partial Dier. Equ. 32 (13)
(2007) 431452.
[32] P.B. Mucha, M. Pokorn, E. Zatorska, Approximate solutions to model of two-component reactive ow,
Discrete Contin. Dyn. Syst. 7 (5) (2014) 10791099.
[33] P.B. Mucha, M. Pokorn, E. Zatorska, Heat-conducting, compressible mixtures with multicomponent diusion:
construction of a weak solution, 2014, submitted for publication, arXiv:1401.5112.
[34] E. Nelson, Dynamical Theories of Brownian Motion, Princeton Univ. Press, 2001.
[35] L. Nirenberg, On elliptic partial dierential equations, Ann. Sc. Norm. Super. Pisa, Cl. Sci. (3) 13 (1959) 115162.
[36] P. Noble, J.-P. Vila, Stability theory for dierence approximations of some dispersive shallow water equations and
application to thin lm ows, 2014, submitted for publication, arXiv:1304.3805.
[37] S.M. Shugrin, Two-velocity hydrodynamics and thermodynamics, J. Appl. Mech. Tech. Phys. 39 (1994) 522537.
[38] A. Vasseur, C. Yu, Existence of global weak solutions for 3D degenerate compressible NavierStokes equations,
arXiv:1501.06803, 2015.
[39] J.L. Vzquez, The Porous Medium Equation. Mathematical Theory, Oxf. Math. Monogr., 2007.
[40] C. Villani, Hypocoercivity, Mem. Am. Math. Soc., vol. 202(950), 2009.
[41] E. Zatorska, Fundamental problems to equations of compressible chemically reacting ows, PhD thesis,
University of Warsaw, 2013.
[42] E. Zatorska, On the ow of chemically reacting gaseous mixture, J. Dier. Equ. 253 (12) (2012) 34713500.