Académique Documents
Professionnel Documents
Culture Documents
Introduction
These Lecture Notes are devoted to some aspects of the theory of the Navier-Stokes-Fourier system. We
shall discuss 1) existence of weak solutions, 2) existence of suitable weak solutions and relative entropies,
3) weak strong uniqueness property in the class of weak solutions. For physical reasons, we shall limit
ourselves to the three dimensional physical space, and for the sake of simplicity, to the ows in bounded
domains with no-slip boundary conditions.
1.1
1.1.1
There are several ways to dene weak solutions for the complete Navier-Stokes-Fourier system. Here, we
shall mention three of them: the convenience of each denition depends on the mathematical assumptions
that one imposes on the constitutive laws for pressure (internal energy) on one hand, and on the transport
coecients on the other hand. First one and the third one are continuations of the theories based on
the so called eective viscous ux identity started by P.L. Lions [32] and the second one, due to Bresch,
Desjardins [2] can be considered as a continuation of theories based on new `a priory estimates in the line
started by Kazhikov [28].
The rst approach due to Feireisl [14] based on a weak formulation of the continuity, momentum and
internal energy equations is convenient for the pressure laws
p(, ) = pc () + p (),
1
where
pc () , p () , /3.
Here, denotes the temperature, the density and is the adiabatic coecient of the uid. The heat
conductivity in this approach has to be temperature dependent (with a convenient polynomial growth),
and the viscosity coecients have to be constant.
The second approach due to Bresch, Desjardins [2], [3] (see also Mellet, Vasseur [36]) is convenient
in the case when the shear viscosity and the bulk viscosity depend on the density and satisfy the
dierential identity
2
( ) () = 2 () 2(),
3
and the pressure satises
p(, ) = + pc (),
where pc () is singular at 0 and behaves as at . The main ingredient in the prof in
this situation is the fact that the particular relation between viscosities stated above makes possible to
establish a new mathematical entropy identity, that provides estimates for the gradient of density. This
estimate implies compactness of the sequence of approximating densities.
The third approach is due to [17]. It is based on the weak formulation of continuity and momentum
equations and on the formulation of the conservation of energy in terms of the specic entropy that
involves explicitly the second law of thermodynamics with entropy production rate being a non negative
measure. This approach is applicable for the pressure laws p(, ) exhibiting the coercivity of type
and 4 for large densities and temperatures. The viscosity coecients are temperature dependent and
have to behave as 1 + (1 > 2/5), and the heat conductivity has to behave as 1 + 3 . This setting
includes at least one physically reasonable case of a monoatomic gas in the situation when the radiation
is not neglected and is given by the Stephan-Boltzman law.
The above formulation is suciently weak to allow existence of variational solutions for large data.
On the other hand it is suciently robust to yield most of low Mach and low Reynolds number limits of
mathematical uid mechanics, and - more surprisingly - it obeys the weak-strong uniqueness principle.
In this Lecture Notes, we shall concentrate on the third approach. We shall discuss within this concept
three issues: 1) existence of weak solutions (or, more precisely, in order to concentrate on the essence
of the problem, the compactness of the family of weak solutions); 2) Relative entropy inequality for the
complete system; 3) Weak strong uniqueness property in the class of weak solutions. This text does not
contain any new results: it is a compilation of our recent works [17, Chapter3] and [18].
1.1.2
The concept of weak solutions in uid dynamics was introduced in 1934 by Leray [30] in the context
of incompressible newtonian uids. It has been extended more than 60 years later to the Newtonian
compressible uids in barotropic regime (meaning that p = p() ) by Lions [32].
The Lions theory relies on two crucial observations:
1) A discovery of a certain weak continuity property of the quantity
4
p() + ( + )divu
3
called eective viscous ux. This part is essential for the existence proof; it employs certain cancelation properties that are available due to the structure of the equations, that are mathematically
expressed through a commutator involving density, momentum and the Riesz operator.
2
2) Theory of renormalized solutions to the transport equation that P.L. Lions introduced together with
DiPerna in [7]. In the context of compressible Navier-Stokes equations, the DiPerna-Lions transport
theory applies to the continuity equation. The theory asserts among others that the limiting
density is a renormalized solution to the continuity equation provided it is squared integrable. This
hypothesis is satised only provided 9/5. The condition on the squared integrability of the
density is the principal obstacle to the improvement of the Lions result.
Notice that some indications on the particular importance of the eective viscous ux have been known
at about the same time to several authors and used in dierent problems dealing with small data (see
Ho [26], Padula [38]) and that the suggestion to use the continuity equations to evaluate the oscillations
in the sequence of approximating densities has been formulated and performed in the one dimensional
case by D. Serre [44].
All physical reasonable adiabatic coecients belong to the interval (1, 5/3], the value = 5/3 being
reserved for the monoatomic gas. This is the reason why it is interesting and important to relax the
condition on the adiabatic coecient in the Lions theory. This has been done Feireisl et al. in [19]. The
new additional aspects of this extension are based on the previous observations by Feireisl in [13] and are
the following:
1) As suggested in [13], the authors have used the oscillations defect measure to evaluate the oscillations
in the sequence of approximating densities, and proved that it is bounded provided > 3/2.
2) The boundedness of the oscillations defect measure is a criterion that replaces the condition of the
squared integrability of the density in the DiPerna-Lions transport theory. Consequently if any
term of the sequence of approximating densities satises the renormalized continuity equation, and
if the oscillations defect measure of this sequence is bounded, then the weak limit of the sequence
is again a renormalized solution of the continuity equation.
1.1.3
The existence theory for the complete Navier-Stokes-Fourier system (with temperature dependent viscosities) employs both Lions and Feireisls techniques. In addition, it presents the following diculties:
1) In order to reduce the investigation to a situation similar to the barotropic case, one has to prove
rst the strong convergence of the temperature sequence. This point involves the treatment of the
entropy production rate as a Radon measure and a convenient use of the compensated compactness,
namely of the Div-curl lemma in combination with the theory of parametrized Young measures.
2) Even after the strong convergence of temperature is known, the weak continuity of the eective viscous
ux is not an obvious issue. It requires to use another cancelation property that mathematically
expresses through another commutator including shear viscosity, symmetric velocity gradient and
the Riesz operator.
3) Once the weak continuity property of the eective viscous ux is known, the proof follows the lines
of Lions and Feireisls approaches: a) one proves rst the boundedness of the oscillation defect
measure for the sequence of densities; b) the boundedness of oscillations defect measure implies
that the limiting density is a renormalized solution to the continuity equation; c) the renormalized
continuity equation is used to show that the oscillations in the density sequence do not increase in
time. This means the strong convergence of density.
3
1.1.4
Weak solutions are not known to be uniquely determined (cf. e.g. exposition of Feerman [12] dealing
with three dimensional incompressible Navier-Stokes equations) and may exhibit rather pathological
properties, see e. g. Ho and Serre [27]. So far, the best property that one may expect in the direction
of an uniqueness result, is the weak-strong uniqueness, meaning that any weak solution coincides with
the strong solution emanating from the same initial data, as long as the latter exists. The weak-strong
uniqueness principle is known for the incompressible Navier-Stokes equations since the works of Prodi[42],
Serrin [45], 1959, 1962. About 50 years later, the weak-strong uniqueness problem has been revisited
by Desjardins [6] and Germain [22]) for the compressible Navier-Stokes equations. They obtained some
partial and conditional results. Finally, the unconditional weak strong uniqueness principle has been
proved in [16] (see also related paper [20]).
Only very recently the weak strong uniqueness property has been proved in [18] for weak solutions of
the complete Navier-Stokes-Fourier system in the entropy formulation introduced in [17].
In all cases cited above, the weak strong uniqueness principle has been achieved by the method of
relative entropies. Relative entropy is a functional whose role is to measure the distance between a weak
solution of the investigated equations and any arbitrary smooth function exhibiting some characteristic
properties of solutions of the investigated equations, as e.g. the sign or the boundary conditions. This
functional must satisfy a convenient dierential inequality (called relative entropy inequality) that is then
used to evaluate the evolution of the relative entropy along the time line. There is no general algorithm to
construct this functional and this fact makes the construction of a convenient relative entropy functional
for any given specic problem quite dicult. The method of relative entropies has been used in various
context by dierent authors, see Dafermeos [5] and later on P.L. Lions [32], Saint-Raymond [43], Grenier
[23], Masmoudi [33], Ukai [48], Wang and Jiang [50], among others. The notion of dissipative solutions
introduced in Lions [31] for the incompressible Euler equations is very much related to the concept of
relative entropies.
1.2
Navier-Stokes-Fourier system
Navier-Stokes-Fourier system describes the motion of a compressible, viscous and heat conducting uid.
For simplicity, we suppose that the uid lls a xed bounded domain and we shall investigate its
evolution through an arbitrary large time interval (0, T ). The motion will be described by means of
three basic state variables: the mass density = (t, x), the velocity eld u = u(t, x), and the absolute
temperature = (t, x), where t (0, T ) is time variable and x R3 is the space variable in the
Eulerian coordinate system. We shall investigate the time evolution of these quantities. It is described
by the following system of partial dierential equations.
t + divx (u) = 0,
(1.1)
(1.2)
(1.3)
where p = p(, ) is the pressure, s = s(, ) is the (specic) entropy, is the entropy production rate
S is the viscous stress tensor and q is the heat ux. The rst equation expresses the mathematical
formulation of the balance of mass, the second one, the conservation of momentum, and the third one,
the conservation of energy in terms of the specic entropy.
4
(1.4)
where I is the identity tensor, while q is the heat ux satisfying Fouriers law
q = ()x .
(1.5)
(1.6)
The system of equations (1.1 - 1.3), with the constitutive relations (1.4), (1.5) is called Navier-StokesFourier system. Equations (1.1 - 1.6) are supplemented with initial conditions
(0, ) = 0 , u(0, ) = 0 u0 , s(, )(0, ) = 0 s(0 , 0 ),
0 0, 0 > 0,
(1.7)
and no-slip boundary conditions for velocity and zero heat transfer conditions through the boundary
u| = 0,
q n| = 0,
(1.8)
S(, x u) : x u
.
(1.9)
With this postulate, equation (1.3) becomes inequality. In order to compensate the loss of information,
we may postulate, that the total energy of the system in the volume is conserved, namely
)
(
d
1
|u|2 + e(, ) dx = 0.
(1.10)
dt 2
We introduce the quantity H = e s (that is reminiscent of the Helmholtz free energy), where
e = e(, ) denotes the specic energy of the gas. Adding equations (1.3) and (1.10), we obtain
)
(
1
d
|u|2 + H (, ) H (, )( ) H (, ) dx
(1.11)
dt 2
(
)
q(, x ) x
S(, x u) : x u
dxdt 0,
On the other hand, if (, , u) > 0, > 0 is a trio of smooth functions satisfying (1.11.8) one may
derive, at least formally, the so called relative entropy identity,
)
(
1
2
|u U| + E(, |r, ) (, ) dx
(1.12)
2
S(, x u)
q(, x )
+
: x u dxdt
: x dxdt
2
0
)
(
1
=
0 |u0 U(0, )|2 + E(0 , 0 |r(0, ), (0, )) dx
2
q(, x )
+
x dxdt
S(, x u) : x Udxdt
(
)
+
t U + u x U (U u) dx dt
0
x p(r, )
dx dt,
r
0
(
)
+
p(r, ) p(, ) divx U dxdt
(U u)
+
0
(
)(
)
s(r, ) s(, ) t + u x dx dt
1
)
) (
t p(r, ) + U x p(r, ) dx dt,
r
1.3
Constitutive relations
We assume that the thermodynamic functions p, e, and s are interrelated through Gibbs equation
ds(, ) = de(, )
p(, )
d.
2
(1.13)
(1.14)
(1.15)
Thus, the thermodynamic stability in terms of the function H , can be reformulated as follows:
7 H (, ) is strictly convex,
(1.16)
(1.17)
while
The above relations reect stability of the equilibrium solutions to the Navier-Stokes-Fourier system (see
Bechtel, Rooney, and Forest [1]) and play a crucial role in the analysis of the Navier-Stokes-Fourier
system.
Taking into account the eects of radiation expressed through the Stefan-Boltzman law, we shall
assume that the pressure p = p(, ) can be written in the form, in agreement with the above considerations
(
) a
p(, ) = /(1) P
(1.18)
+ 4 , a > 0, > 3/2,
3
1/(1)
where
P C 1 [0, ), P (0) = 0, P (Z) > 0 for all Z 0.
(1.19)
In agreement with Gibbs relation (1.13), the (specic) internal energy can be taken as
e(, ) =
)
1 /(1) (
4
P
.
+
a
1
1/(1)
(1.20)
Furthermore, by virtue of the second inequality in thermodynamic stability hypothesis (1.14), we have
0<
P (Z) P (Z)Z
< c for all Z > 0.
Z
(1.21)
Relation (1.21) implies that the function Z 7 P (Z)/Z is decreasing, and we suppose that
lim
P (Z)
= P > 0.
Z
(1.22)
1/(1)
)
+
4a 3
,
3
(1.23)
1 P (Z) P (Z)Z
< 0, lim S(Z) = 0.
Z
1
Z2
(1.24)
From the point of view of statististical mechanics, the above hypotheses are physically reasonable at
least in two cases: if = 5/3 they modelize the monoatomic gas, if = 4/3 they modelize the so called
7
relativistic gas. The reader may consult Eliezer, Ghatak, and Hora [9] and [17, Chapter 3] for the physical
background and further discussion concerning the structural hypotheses (1.18 - 1.22).
For the sake of simplicity and clarity of presentation, we take the transport coecients in the form
, C 1 [0, ) are globally Lipschitz and 0 (1 + ) (), 0 (),
C 1 [0, ),
0 > 0,
0 (1 + 3 ) () 1 (1 + 3 ), 0 < 0 1 ,
(1.25)
(1.26)
2
2.1
Here and heareafter, we shall suppose that the initial data satisfy the following assumptions
0 < 0 L (),
(
0 u0 L2/(+1) (; R3 ),
0 s(0 , 0 ) L1 (),
0 < < 0 ,
)
1
0 |u0 |2 + H (0 , 0 ) H (, )(0 ) H (, ) dx < .
2
(2.1)
(2.2)
(2.3)
In many parts of this note (as e.g. in the existence theory), the condition 0 < 0 could be relaxed to
0 0 and 0 < < 0 to 0 < 0 . Investigation of this issue is however not a subject of this exposition.
Denition 1
We say that a trio (, , u) is a weak solution to the Navier-Stokes-Fourier system (1.1 - 1.8) if:
(i) the density and the absolute temperature satisfy (t, x) 0, (t, x) > 0 for a.a. (t, x) (0, T ) ,
L (), u L (0, T ; L2/(+1) (; R3 )), u2 L (0, T ; L1 ()), L4 ())L2 (0, T ; W 1,2 ()),
log L2 (0, T ; W 1,2 (; R3 )) and u L2 (0, T ; W01,2 (; R3 ));
(ii) Cweak ([0, T ]; L ()) and equation (1.1) is replaced by a family of integral identities
(
)
dx =
t + u x dx dt
(2.4)
(
)
u dx =
u t + u u : x + p(, )divx S(, x u) : x dx dt (2.5)
(iv) the entropy balance (1.3), (1.9) is replaced by a family of integral inequalities
)
(
q(, x ) x
s(, ) dx +
S(, x u) : x u
dx dt
0
)
(
q(, x ) x
s(, )t + s(, )u x +
dx dt
(2.6)
(2.7)
gdx
means g(x, )dx g0 (x)dx.
Remark
Any weak solution satises the so called dissipation inequality:
)
(
1
2
|u| + H (, ) H (, )( ) H (, ) (, )dx
2
(
)
q(, x ) x
+
S(, x u) : x u
dxdt
0
)
(
1
0 |u0 |2 + H (0 , 0 ) H (, )(0 ) H (, ) dx
2
(2.8)
for a.a. (0, T ). Indeed, the dissipation inequality (2.8) is obtained from the sum of identity (2.7) and
the entropy balance (2.6) multiplied by .
Denition 2
We say that the triplet (, , u) is a renormalized weak solution to the Navier-Stokes-Fourier system (1.1
- 1.8) if it is a very weak solution, if b() Cweak ([0, T ], L1 ()) and if the couple (, u) satises the
continuity equation in the renormalized sense,
(
(
)
)
b() dx =
b() t + u x dxdt
b () b() divudxdt
(2.9)
0
Denition 3
We say that the triplet (, , u) is a suitable weak solution to the Navier-Stokes-Fourier system (1.1 1.8) if it is a weak solution and if it satises the relative entropy inequality
)
(
1
|u U|2 + E(, |r, ) (, ) dx
(2.10)
2
S(, x u)
q(, x )
+
: x u dxdt
: x dxdt
2
0
)
(
1
0 |u0 U(0, )|2 + E(0 , 0 |r(0, ), (0, )) dx
q(, x )
+
S(, x u) : x Udxdt
x dxdt
(
)
+
t U + u x U (U u) dx dt
0
+
0
(
)(
)
s(r, ) s(, ) t + u x dx dt
x p(r, )
dx dt,
r
)
p(r, ) p(, ) divx U dxdt
(U u)
)
) (
t p(r, ) + U x p(r, ) dx dt
r
(2.11)
(2.12)
where
H (, ) = e(, ) s(, ).
2.2
Main results
In this section, we shall formulate four theorems. First theorem asserts existence of weak solutions.
Theorem 2.1 Let R3 be a bounded Lipschitz
e, s satisfy hypotheses (1.18 - 1.24), and that the
Finally assume that the initial data (1.7) verify
system (1.11.8) admits at least one renormalized
10
The proof of this result can be found in [17, Chapter 3] for domains with regularity C 2, , (0, 1).
The necessary modications to accommodate the Lipschitz domains are explained in Poul [40]. In this
Lecture Notes, we shall prove solely the weak compactness property of the (hypothetical) set of weak
solutions. This property is usually considered to be the main ingredient in the existence proof. Note
however, that the compressible models are very much approximation sensitive. Therefore, the way
from the weak compactness to the real existence can still be very long and should not be underestimated.
Theorem 2.2 Let R3 be a bounded Lipschitz domain. Suppose that the thermodynamic functions p,
e, s satisfy hypotheses (1.18 - 1.24), and that the transport coecients , , and obey (1.25), (1.26).
Finally assume that the initial data (1.7) verify (2.12.3). Let (n , n , un ) be a sequence of weak solutions
to the complete Navier-Stokes-Fourier system (1.11.8). Then there exists a subsequence (denoted again
(n , n , un )) such that
n in L (0, T ; L (),
n in L2 (0, T ; W 1,2 ()),
un u in L2 (0, T ; W01,2 (; R3 )),
and the trio (, , u) is again a weak solution of the complete Navier-Stokes-Fourier system (1.11.8).
This compactness result will be proved in Section 3.
The second result to be discussed in this Lecture Notes claims that any weak solution satises the
relative entropy inequality.
Theorem 2.3 Let all assumptions of Theorem 2.1 be satised and let the trio (, , u) be a weak solution
to the complete Navier-Stokes-Fourier system (1.11.8).
Then (, , u) is a suitable weak solution. In particular, it satises the relative entropy inequality
(2.10) with test functions (r, , U) belonging to the class (2.11).
Theorem 2.3 has been proved in [18]. We shall reproduce this proof in Section 4.
The third result to be dealt with in this Lecture Notes is the weak strong uniqueness principle. It
asserts that any weak solution coincides with the strong solution emanating from the same initial data
as long as the latter exists.
Theorem 2.4 Let assumptions of Theorem 2.1 be satised, where, moreover, the function P introduced
in (1.19) is twice continuously dierentiable on (0, ). Let (, , u) be a weak solution of the Navier u
) is a classical (strong) solution to the NavierStokes-Fourier system (1.11.8). Assume that (
, ,
Stokes-Fourier system (1.11.8) in (0, T ) that satises dissipation inequality (2.8) and that belongs
to the class
0 < < ,
0 < < ,
L (0, T ; L (, R3 )),
u
(2.13)
t u
m L (0, T ; L ()),
, x , m
t , t ,
x , x u
emanating from the same initial data. Then
u
) in [0, T ] .
(, , u) = (
, ,
11
m = 1, 2
The weak strong uniqueness property in the class of weak solutions has been proved in [18]. This
proof will be reproduced in Section 5.
Remark(Relaxing regularity of test functions in Theorem 2.3.)
Using a density argument we can extend the class of test functions r, , U appearing in the relative
entropy inequality (2.10), (2.11).
Sucient conditions for the left hand side of the relative entropy inequality to be well dened are,
for example,
0 < r < , 0 < < ,
(2.14)
U L (0, T ; L2 (, R3 )).
(2.15)
A short inspection of the right hand side (2.10) implies that the integrals are well-dened if, for
example
t U L (0, T ; L6 (; R3 )), x U L (0, T ; L (, R33 )),
(2.16)
t L (0, T ; L4 ()),
x L (0, T ; L (; R3 )),
t r L (0, T ; L ()),
3
x r L (0, T ; L ()).
6
(2.17)
(2.18)
Finally,
U| = 0.
(2.19)
Consequently, Theorem 2.3, in particular, the relative entropy inequality (2.10), are valid even if we replace the hypotheses on smoothness and integrability of the test functions (r, , U) by weaker hypotheses,
namely (2.14-2.19). In particular, r, , U may be another (strong) solution emanating from the same
initial data 0 , 0 u0 .
Remark (Existence of classical solutions)
Existence of classical solutions in class (2.13) with suciently smooth initial data on a short time interval
[0, Tmax ) that depends on the size of the initial data is well known on a large class of domains. The
rst results in this direction are due to Valli, Zajaczkowski [49], Matsumura and Nishida [34], [35].
3.1
12
(3.1)
] [
]
H (, ) H (, ) + H (, ) H (, )( ) H (, ) .
Z>0
(3.2)
that can be derived from (1.211.22). Employing (1.15), (1.18), (1.20) and (3.2), we obtain
H (, ) H (, ) =
H (, z)dz =
(z ) e(, z)dz 4a
z
z 2 (z )dx
(3.3)
and
H (, ) H (
r, )(
r)H (
r, ) =
[
)
2 H(w, )dw dz =
)
1
p(w, )dw dz (3.4)
w
)] [
]
log(/) + 1 ( ) .
With observations (3.33.4) at hand we deduce from (2.8) the following estimates
esssup(0,T )
n u2n dx c,
(3.5)
esssup(0,T )
n dx c,
(3.6)
4n dx c
(3.7)
esssup(0,T )
(3.8)
(
)
wL2 (;R3 ) c(M ) x wL2 (;R33 ) +
w2 dx .
(3.10)
V
13
(
)
n
lim sup
n S 1/(1) dx = 0
1/(1)
n
{n n
}
n
whenever n 0 is bounded in L () and 0 n 0 in Lp ().
Then for any M0 > 0, 0 > 0, and S0 > 0 there exist = (M0 , 0 , S0 ) > 0, = (M0 , 0 , S0 ) > 0
such that for any non-negative functions ,
dx M0 ,
( + p )dx 0
and
(
S
1/(1)
dx S0
{ } .
we have
(3.11)
T(x w) = x w + Tx w 23 divw,
(3.12)
1
|T(x un )|2 dxdt
n
c;
whence employing rst Lemma 3.1 and then the standard Poincare inequality, we get
un L2 (0,T ;W 1,2 (;R3 )) c.
(3.13)
[1, 3/2]
(3.14)
(3.15)
Consequently, recalling (1.23), we verify that assumptions of Lemma 3.3 are satised with some 3 < p < 4.
(We recall that S0 is positive due to (2.6) with test function = 1.) Therefore, we deduce from (3.14)
and Lemmas 3.2, 3.3.
We get by the Sobolev imbedding and by interpolation from (3.7) and (3.16) that
n L3 (0,T ;L9 ()) c, n L17/3 ((0,T )) c
(3.17)
Recalling (3.7), (3.14), (3.13) and having in mind (1.4), (1.5), (1.21), (1.26) we deduce that
S(n , x un )L2 (0,T ;L4/3 (;R33 )) c.
(3.18)
(3.19)
(3.20)
(3.21)
and
as well as
Step 4 (Estimates via the Bogovskii operator)
Under assumptions (1.181.22)
|p(, )| c( + + 4 ).
(3.22)
Consequently, we can deduce from (3.63.7) only p(n , n L (0,T ;L1 ()) c. We however need for the
pressure better estimate than an estimate in L1 (). One way to get a better estimate is based on the
following lemma.
Lemma 3.4 Let Q be a bounded Lipschitz domain. There an operator BQ : Lp (Q) W01,p (Q; R3 ),
1 < p < satisfying,
(3.23)
T 1+
p
(n , n )dxdt c.
0
See [17, Section 2.2.5, Appendix] for more details and calculations.
Step 5 (Weak limits)
15
Bounds (3.63.7), (3.13), (3.163.23) imply existence of a subsequence (denoted again (n , n , un )) and
of a trio (, , u) such that
n in L (0, T ; L ()),
n in L (0, T ; L4 ()),
(3.24)
un u in L2 (0, T ; W 1,2 (; R3 )),
n in L2 (0, T ; W 1,2 ()).
Moreover, if we denote by g(, , u) a weak limit of the sequence g(n , n , un ) in L1 ((0, T ) )), we have
for the non linear quantities,
log n log in L2 (0, T ; W 1,2 ()),
p(n , n ) p(, ) in Lq ((0, T ) n ) with some q > 1,
S(n , x un ) S(, x u) in L4/3 ((0, T ) ; R33 ),
(3.25)
t 7
n dx, t 7
h(n ) dx, t 7
n un dx,
on [0, T ], where Cc (). This fact makes possible to use the Arzela-Ascoli compactness argument
which in combination with the density argument yields convergence of the sequences n , h(n ) and n un
in Cweak ([0, T ]; Lq ()) with some q > 6/5. Employing moreover the compact imbedding Lq () ,,
W 1,2 (), we get the convergence of these quantities in L2 (0, T ; W 1,2 ()). Summarizing the above, we
have
n in Cweak ([0, T ]; L ()) and in L2 (0, T ; W 1,2 ()),
h(n ) h() in Cweak ([0, T ]; Lq ()), q [1, ) and in L2 (0, T ; W 1,2 ()),
where h W 1, ((0, )),
(3.26)
T (
)
0 (0, ) dx =
t + u x dx dt
(3.27)
16
for any
T
0
and
0 u0 (0, ) dx
)
u t + u u : x + p(, )divx S(, x u) : x dx dt
Cc1 ([0, T )
; R3 ), | = 0;
(
)
Lk () t + u x dxdt =
Tk ()divudxdt
(
)
Tk () t + u x dxdt +
0 Lk (0 )(0, )dx
(Tk () Tk ())divudxdt
T (z) =
(3.29)
3.2
(3.28)
Tk (0 )(0, )dx,
(3.30)
Tk (w)
dw,
w2
z if z [0, 1],
concave on [0, ),
2 if z 3.
Step 1
The entropy inequality (2.6) can be rewritten as identity
(3.31)
=
0
)
(
q(n , x n ) x
n s(n , n )t + n s(n , n )un x +
dx dt
n
where n is a positive continuous linear functional on the (metric) space Cc1 ([0, T ) ) dened by the
above equation. We therefore easily verify by the standard Schwartz argument that
| < n , > | c(K, n)C(K) for any compact K [0, T ) ;
whence, in fact, n can be extended in a unique way to a positive linear functional on the (metric) space
Cc ([0, T ) ). By virtue of the Riesz representation theorem, n can be identied with a non negative
regular measure n on the family of Borel subsets of [0, T ) such that,
n (K) < , for any compact K [0, T )
17
dn , Cc ([0, T ) ).
< n , >=
[0,T )
2
t L 0
T L
and using (3.31) with test function L (t)1(x), we get, due to (3.20),
| < n , L ()1() > | c,
(3.32)
< n , >=
dn , C([0, T ] ),
[0,T ]
and we have
n [C([0,T ])] = n ([0, T ] ) =
dn < c,
[0,T ]
Wn = (Tk (n ), 0, 0, 0)
Since divVn = n and since the imbedding [C([0, T ] )] , W 1,q ((0, T ) ) is compact for any
q (1, 4/3), the assumptions of the lemma on (0, T ) are satised. Therefore,
4
4
Tk ()sM (, ) + aTk ()3 = Tk () sM (, ) + aTk () 3 ,
3
3
18
(3.33)
where sM (, ) = S(/ 1 ).
We shall rst prove that
Tk ()sM (, ) Tk () sM (, ),
where
Tk (z) = kT (z/k),
C[0, ) T =
z if z [0, 1],
T strictly increasing on [0, ) ,
limz T (z) = 2.
(3.34)
)
(
s sM (n , n ) Tk (n ) Tk () =
[ (
(
))
(
(
))](
)
(
(
))(
)
n sM n , Tk1 Tk (n ) sM n , Tk1 Tk ()
Tk (n )Tk () +n sM n , Tk1 Tk ()
Tk (n )Tk () .
To this end we write
(
(
)
n sM n , Tk1 Tk () (t, x)) = (t, x, n )
(3.35)
(3.36)
(y, w) =
(y, )dy (),
RM
(3.37)
for any h belonging to the class (2.9) and G W 1, ((0, )). This implies (3.35) by virtue of Lemma
3.6.
19
h()G()d (,) (, ) =
h()d ()
G()d ().
R2
Consequently,
(t, x, )G()(t, x) =
R2
(
)
P (wn ) P (w),
G(wn ) G(w),
P (wn )G(n ) P (w)G(w)
in L1 (Q).
(i) Then
P (w) G(w) P (w)G(w).
(ii) If G(z) = z and if
P (w) w = P (w)w,
then
P (w) = P (w).
In the last two formulas w denote weak limit of wn in L1 (Q).
Lemma 3.7 implies, in particular,
Tk ()3 Tk () 3
that in turn with (3.333.34)yields
Tk ()3 = Tk () 3
and nally, by monotone convergence, as k ,
4 = 3 .
(3.38)
n a.e. in (0, T ) .
(3.39)
20
Step 5
Coming back with (3.39) to the momentum equation (3.28), we obtain
0 u0 (0, ) dx
(3.40)
=
0
)
u t + u u : x + p(, )divx S(, x u) : x dx dt
for any
; R3 ), | = 0.
Moreover, estimate (2.8) yields boundedness of the sequences
)
(n )
(n ) (
2
(n )
T
x un + (x un ) divun ,
divun ,
x n
n
3
n
n
Cc1 ([0, T )
(3.41)
in L2 ((0, T ) )); whence by the lower weak continuity combined with (3.39) and (3.24) one gets
T
0
lim inf
n
(
)
q(, x ) x
S(, x u) : x u
dx dt
(
)
q(n , x n ) x n
S(n , x un ) : x un
dx dt,
n
(3.42)
0 s(0 , 0 )(0, ) dx +
for any
3.3
3.3.1
Cc1 ([0, T )
(
)
q(, x ) x
S(, x u) : x u
dx dt
(3.43)
)
(
q(, x )
s(, )t + s(, )u x +
x dx dt
), 0.
4
3 ()
(
)
1
p(, )Tk () p(, ) Tk () .
+ ()
21
(3.44)
Step 1
First we introduce operators A = x 1 and R = x x 1 ,
(1 )j (v) = F 1
[ i
j
F(v)()
||2
( 1 )ij (v) = F 1
[
]
i j
F(v)()
,
||2
v(x)exp(i x)dx.
R3
j=1
x A(v) = R(v),
divA(v) = v, v Lp (R3 ),
A(v)wdx =
R3
vA(w)dx,
R3
with
R(v)wdx =
R3
22
(
)
(t, x) p(n , n )Tk (n ) S(n , un ) : R[1 Tk (n )] dxdt
lim
(
)
(t, x) (p(, ) Tk () S(, u) : R[1 Tk ()] dxdt
=
0
lim
)
Tk (n )un R[n un ] n (un un ) : R[1 Tk (n )] dxdt
(3.45)
)
Tk ()u R[u] (u u) : R[1 Tk ()] dxdt,
3
where (R[z])j = k=1 Rjk [zk ]. In fact, other terms that should eventually appear at the right hand side
are zero by virtue of standard compactness arguments.
Step 3
Next, we shall need the following lemma in the spirit of Meyer [37], see [17, Theorem 10.27]
Lemma 3.9 (Commutators I) Let Un U in Lp (R3 ; R3 ), vn v in Lq (R3 ), where
1 1
1
+ = < 1.
p q
r
Then
vn R[Un ] R[vn ]Un vR[U] R[v]U
in Lr (R3 ; R3 ).
Lemma 3.9 is a consequence of the Dic-curl lemma 3.5. Indeed, realizing that
U R(V) V R(U) = X R(V) + Y R(U),
where
X = U R(U), Y = R(V) V.
Seeing that divX = divY = 0 and that curlR(U) = curlR(V) = 0, we obtain employing Lemma 3.5.
Vn R[Un ] Un R[Vn ] V R[U] U R[V]
provided Vn V in Lq (R3 ), Un U in Lp (R3 ; R3 ). Taking in the last formula subsequently Vn =
(vn , 0, 0), (0, vn , 0), and (0, 0, vn ) we get Lemma 3.9.
Step 4
Combining this lemma with the convergence established in the rst two lines of (3.26), we get
(
)
(
)
Tk (n )R[n un ] n un R[1 Tk (n )] (t) Tk ()R[u] u R[1 Tk ()] (t)
23
weakly in Lr (; R3 ) with some r > 6/5 for all t [0, T ]; whence by the compact imbedding Lr () ,,
W 1,2 () and the Lebsegue dominated convergence theorem used over (0, T ) we conclude
(
)
T
u
T
(
)R[1
u
]
R[1
T
(
)]
dx
n
k
n
n
n
n
n
k
n
0
(
)
u
T
()R[1
u]
R[1
T
()]
dx
k
T
0
=
0
)
(
(t, x) p(, )Tk () p(, ) Tk () dxdt
0 (
)
(t, x) S(, u) : R[1 Tk ()] S(, u) : R[1 Tk ()] dxdt.
T
(3.46)
Step 5
Hereafter we will need another another commutator lemma in the spirit of Meyer [37], see [17, Theorem
10.28]
Lemma 3.10 (Commutators II) Let w W 1,r (R3 ), z Lp (R3 ; R3 ), 1 < r < 3,
Then for all such s we have
1
r
1
p
1
3
<
1
s
< 1.
R[wz] wR[z]W ,s (R3 ;R3 ) CwW 1,r (R3 ) zLp (R3 ;R3 ) ,
where
1
s
1
6
We can write
T
+
0
)
( 2
(t, x) () + () Tk () divx u dxdt
3
{
[
(
)]
[
]}
Tk () R : () x u + (x u)T ()R : x u + (x u)T
dxdt
[
]
Tk ()()R : x u + (x u)T dxdt,
+
3
(t, x)
(4
3
)
(n ) + (n ) Tk (n ) divx un dxdt
Tk (n )(n , un ) dxdt
+ lim
(3.47)
and
(t, x)
0
T
0
(4
3
Tk ()(, u) dxdt,
24
)
() + () Tk () divx u dxdt
(3.48)
where
( [
(
)]
[
])
(n , un ) = R (t, x)(n ) un + (un )T (t, x)(n )R : un + (un )T .
Thanks to Lemma 3.10, the sequence (n , un ) is bounded in L1 (0, T ; W ,q (; R3 )) with some (0, 1),
q > 1; whence curlUn is compact in W 1,r ((0, T ) ; R33 ) (and of course divVn is compact in
W 1,r ((0, T ) ; R33 ), cf. (2.9)) with some r > 1, where
Vn [Tk (n ), Tk (n )un ],
Un [(n , un ), 0, 0, 0].
We may thus apply a convenient version of Div-curl lemma (see e.g. [17, Theorem 10.21]) to these
4-dimensional vector elds to get
(n , un ) Tk (n ) (, u) Tk (),
where, due to (3.39),
(, u) = (, u).
This result in combination with (3.46) and (3.47) yields the eective viscous ux identity (3.44).
3.3.2
(3.49)
T
(
)
)
Tk () Tk () dxdt
p(, ) Tk () p(, ) Tk () dxdt,
1+
0
1+
d lim sup
T
0
q
|Tk (n ) Tk ()| dx =
0
(
)
1
q
1
+
dxdt
|T
(
)
T
()|
k
n
k
(1 + )
25
[
c
T
0
]q/(+1)
1
|Tk (n ) Tk ()|+1 dxdt
,
1+
that stays at the right hand side of the last inequality can be calculated from the eective viscous ux
identity (3.44); using (3.13) we nd that
T
0
1
|Tk (n ) Tk ()|+1 dxdt c(n)Tk (n ) Tk ()L2 ((0,T ))
1+
(q2)/(2q2)
q/(2q2)
n0
(3.52)
where the expression at the left hand side is called oscillations defect measure, see [17, Chapter 3, Section
3.7.5].
On the other hand, relation (3.52) implies that the limit quantities , u satisfy the renormalized
equation of continuity (2.9), see [17, Lemma 3.8], that reads
Lemma 3.11 (Renormalized continuity equation) Let Q R3 be open set and let
n
in L1 ((0, T ) Q),
un u
in Lr ((0, T ) Q; R3 ),
un u
in Lr ((0, T ) Q; R33 ), r > 1.
Let
oscq [n ]((0, T ) Q) <
(3.53)
for 1q < 1 1r , where (n , un ) solve the renormalized continuity equation (2.9) (with replaced by Q).
Then the limit functions , u solve (2.9) with ( replaced by Q) as well.
We deduce using Lemma 3.11 that
(
)
Lk () t + u x dxdt =
Tk ()divudxdt
0 Lk (0 )(0, )dx,
(3.54)
26
for a.a. (0, T ), where the second term at the right hand side tends to 0 as k . Reasoning as in
(3.51) we verify that
p(, )Tk () p(, ) Tk () 0.
Consequently, there holds
)
log log ( ) dx 0.
Finally, since the function z z log z is strictly convex on (0, ), we necessarily have
log log = 0 a. e. in (0, T ) .
n a.e. in (0, T ) .
(3.55)
With (3.39) and (3.55) at hand, momentum equation (3.28) turns into (2.5) and entropy inequality
(3.43) turns into (2.6). Moreover, clearly
T
(
T
(
)
)
1
1 2
(t)
n u2n + n e(n , n ) dxdt
(t)
u + e(, ) dxdt,
0
2
0
2
cf. last line in (3.26), (3.23) and (3.39), (3.55). The latter limit yields (2.7). Theorem 2.3 is proved.
We deduce a relative entropy inequality satised by any weak solution to the Navier-Stokes-Fourier
system. To this end, consider a trio (r, , U) of smooth functions, r and bounded below away from
zero in [0, T ] , and U| = 0.
Taking = 12 |U|2 as a test function in (2.4), we get
)
1
1
|U|2 (, ) dx =
0 |U(0, )|2 dx +
U t U + u x U U dx dt.
(4.1)
2
0
2
Similarly, the choice = U in (2.5) gives rise to
u U(, ) dx
0 u0 U(0, ) dx
=
0
)
u t U + u u : x U + p(, )divx U S(, x u) : x U dx dt.
Combining relations (4.1), (4.2) with the total energy balance (2.7) we may infer that
)
)
(
(
1
1
|u U|2 + e(, ) (, ) dx =
0 |u0 U(0, )|2 + 0 e(0 , 0 ) dx
2
2
((
)
)
+
t U + u x U (U u) p(, )divx U + S(, x u) : x U dx dt.
0
(4.2)
(4.3)
Now, take = > 0 as a test function in the entropy inequality (2.6) to obtain
0 s(0 , 0 )(0, ) dx
s(, )(, ) dx
27
(4.4)
(
S(, x u) : x u
q(, x ) x
)
dx dt
)
(
q(, x )
s(, )t + s(, )u x +
x dx dt.
(
)
q(, x ) x
+
S(, x u) : x u
dx dt
0
)
(
1
0 |u0 U(0, )|2 + 0 e(0 , 0 ) (0, )0 s(0 , 0 ) dx
=
2
((
)
)
+
t U + u x U (U u) p(, )divx U + S(, x u) : x U dx dt
0
(4.5)
)
(
q(, x )
s(, )t + s(, )u x +
x dx dt.
H (r, )(, ) dx =
0 H(0,) (r(0, ), (0, ) dx
+
0
(
)
(
))
t H (r, ) + u x H (r, )
dx dt,
(
)
q(, x ) x
S(, x u) : x u
dx dt
(
)
+
H(0,) (0 , 0 ) (H(0,) )(r(0, ), (0, ))(0 r(0, )) H(0,) (r(0, ), (0, )) dx
((
+
0
)
)
t U + u x U (U u) p(, )divx U + S(, x u) : x U dx dt
)
(
q(, x )
x dx dt.
s(, )t + s(, )u x +
( (
)
(
))
t H (r, ) + u x H (r, )
dx dt
28
(
)
t r (H )(r, ) H (r, ) dx dt.
+
0
(
)
q(, x ) x
S(, x u) : x u
dx dt
(
)
+
H(0,) (0 , 0 ) (H(0,) )(r(0, ), (0, ))(0 r(0, )) H(0,) (r(0, ), (0, )) dx
((
)
)
t U + u x U (U u) p(, )divx U + S(, x u) : x U dx dt
)
( (
)
(
)
q(, x )
x dx dt.
s(, ) s(r, ) t + s(, ) s(r, ) u x +
(
)
+
r s(r, )t + r s(r, )u x dx dt
0
(
)
2
2
,
(H )(r, )t r + ,
(H )(r, )t dx dt
(
))
2
2
u ,
(H )(r, )x r + ,
(H )(r, )x
dx dt
(
)
t r (H )(r, ) H (r, ) dx dt.
+
0
In order to simplify (4.7), we recall several useful identities that follow directly from Gibbs relation
(1.13):
1
2
,
(H )(r, ) = p(r, ),
r
1
(4.8)
r s(r, ) = p(r, ),
r
(
)
2
,
(H )(r, ) = (( ) s) (r, ) = ( ) s(, ) (r, ) = 0,
and
r (H )(r, ) H (r, ) = p(r, ).
29
q(, x ) x
|u U|2 + E(, |r, ) (, ) dx+
S(, x u) : x u
dx dt (4.9)
2
0
)
(
1
2
(
)(
)
(u U) x U (U u) dx dt +
s(, ) s(r, ) U u x dx dt
0
( (
)
)
t U + U x U (U u) p(, )divx U + S(, x u) : x U dx dt
)
( (
)
(
)
q(, x )
s(, ) s(r, ) t + s(, ) s(r, ) U x +
x dx dt
((
)
)
+
1
t p(r, ) u x p(r, ) dx dt,
r
r
0
(
)
p(r, )
p(r, )
u
dx =
(U u)
U p(r, ) dx
r
r
r
(
)
)
p(r, ) (
+ 1
U p(r, ) + p(r, )divU dx
(U u)
r
r
Using the latter identity in the formula (4.9), we get the relative entropy inequality in the form (2.10).
Theorem 2.3 is proved.
Weak-strong uniqueness
In this section we prove Theorem 2.4 by applying the relative entropy inequality (2.10) to r = , = ,
u
, where (
) is a classical (smooth) solution of the Navier-Stokes-Fourier system such that
and U = u
, ,
) = 0 .
(0, ) = u0 , (0,
(0, ) = 0 , u
Accordingly, the integrals depending on the initial values on the right-hand side of (2.10) vanish, and we
apply a Gronwall type argument to deduce the desired result, namely,
and u u
.
, ,
Here, the hypothesis of thermodynamic stability formulated in (1.14) will play a crucial role.
30
5.1
Preliminaries, notation
Following [17, Chapters 4,5] we introduce essential and residual component of each quantity appearing
in (2.10). To begin, we choose positive constants , , , in such a way that
0<
1
min
(t, x) 2
max
(t, x) ,
2 (t,x)[0,T ]
(t,x)[0,T ]
0<
1
x) 2
x) .
min
(t,
max
(t,
2 (t,x)[0,T ]
(t,x)[0,T ]
In can be shown, as a consequence of the hypothesis of thermodynamic stability (1.14), or, more
specically, of (1.16), (1.17), that
2 if (, ) [, ] [, ]
| |2 + | |
E(, |
, ) c
(5.1)
where
hess (t, x) =
, hres = h hess .
0 otherwise
With this notation at hand, and thanks to (3.15), (3.22), we can write
)
(
ess |2 ,
c [1]res + [] + []4 + |[ ]ess |2 + |[ ]
E(, | , )
res
res
(5.2)
5.2
(5.3)
U=u
in (2.10) and using the fact that the initial values coincide, we obtain
Taking r = , = ,
(
)
)
(
q(, x ) x
1
2
|u u| + E(, |
, ) (, ) dx +
S(, x u) : x u
dx dt (5.4)
2
0
(
)(
)
|2 |x u
| dx dt +
u x dx dt
|u u
s(, ) s(
, )
u
0
( (
)
)
+u
x u
(
+ S(, x u) : x u
dx dt
t u
u u) p(, )divx u
31
)
( (
)
(
)
q(, x )
t + s(, ) s(
u
x +
s(, ) s(
, )
, )
x dx dt
)
)
((
u x p(
+
1
t p(
, )
, )
dx dt.
In order to handle the integrals on the right-hand side of (5.4), we proceed by several steps:
Step 1:
We have
| |x u
| dx dt 2x u
L (;R33 )
|u u
2
1
|2 dx.
|u u
2
(5.5)
Step 2:
(
)(
)
s(, ) s(
u x dx
, ) u
[ [
[ (
]
)]
L (;R3 ) 2
, )
x
|u
u
|
dx
+
s(,
)
s(
,
)
s(, ) s(
ess
, )
s(, ) s(
ess
res
|u
u
|
dx
,
2
dx
| dx u u
L2 (;R3 ) + c()
E(, |
, )
|u u
(5.6)
, )
s(, ) s(
res
| dx
|u u
[ (
)]
2
6/5
res L
()
, )
s(, ) s(
2
6/5
res L
(
()
)5/3
dx
E(, |
, )
(5.7)
s(, ) s(
u x dx
, ) u
(5.8)
[
]
L (;R3 ) u u
dx
2W 1,2 (;R3 ) + c()
x
E(, |
, )
0
dx
E(, |
, )
, through the
for any > 0. Here and hereafter, K(, ) is a generic constant depending on , , u
, through
, ,
norms induced by (2.13), and , , while K() is independent of but depends on , u
the norms (2.13).
32
Step 3:
Writing
(
)
(
)
x u
dx
+u
x u
(
) x p(
t u
u u) dx =
(
u u) divx S(,
, )
(
)
(
)
1
x u
dx,
x u
dx + (
) x p(
) x p(
u u) divx S(,
, )
( )(
u u) divx S(,
, )
=
we observe that the rst integral on the right-hand side can be handled in the same way as in Step 2,
namely,
[
(
)]
1
2
) x p(
2L2 (;R3 ) ,
( )(
u u) divx S(, x u
, )
dx K(, ) [ ]ess L2 () + u u
ess
[
(
)]
1
) x p(
( )(
u u) divx S(, x u
, )
dx
res
(
)
2
2
2L6 (;R3 ) ,
K(, ) []res L6/5 () + [1]res L6/5 () + u u
while, integrating by parts,
(
)
x u
dx
) x p(
(
u u) divx S(,
, )
(
=
)
x u
) : x (u u
) + p(
S(,
, )div
u u) dx
x (
Thus using again (5.1), (5.3) and continuous imbedding W 1,2 () , L6 () we arrive at
(
(
)
)
x u
t u
+
u
(
u
u)
dx
S(
,
)
:
(u
u
)
+
p(
,
)div
(
u
u)
dx
x
x
x
|2 u
+c |x |, |x |,
x |
(
)[
2
)
x u
) : x (u u
) + p(
S(,
, )div
u u) dx
x (
dx
E(, |
, )
(
)
t dx =
s(, ) s(
, )
[
]
s(, ) s(
, )
ess
[
]
s(, ) s(
, )
t dx +
33
res
t dx,
(5.9)
where
[
]
s(, ) s(
, )
t dx
(5.10)
res
(
)
dx,
t L ()
[s(, )]res dx + s(
, )L () []res dx K()
E(, |
, )
while
[
]
s(, ) s(
, )
[
]
( ) s(, ) s(
, )
ess
ess
t dx
[
]
s(, ) s(
, )
t dx +
| s(, )| +
sup
(,)[,][,]
ess
( ) s(, ) s(
,
)
ess
t dx,
t dx
(5.11)
)
L ()
| s(, )| t
sup
(,)[,][,]
(
)]
dx.
E(, |
, )
[ ]ess [ ]ess + [ ]ess dx K()
Finally, we write
[
]
s(, ) s(
, )
ess
[
]
) s(
s(
s(, ) s(
, )(
, )(
)
, )
t dx =
[
]
) + s(
s(
, )(
, )(
)
res
ess
t dx
[
]
) + s(
t dx,
s(
, )(
, )(
)
t dx +
where the rst two integrals on the right-hand side can be estimated exactly as in (5.10), (5.11). Thus
we conclude that
(
dx
s(, ) s(
, ) t dx K()
E(, |
, )
(5.12)
[
]
) + s(
t dx.
s(
, )(
, )(
)
Step 5:
Similarly to Step 4, we get
(
dx
x dx K()
s(, ) s(
, ) u
E(, |
, )
[
]
) + s(
u
x dx.
s(
, )(
, )(
)
Step 6:
34
(5.13)
Finally, we have
)
)
((
1
, ) dx
t p(
, ) u x p(
)
1(
+u
dx +
x p(
=
(
)
t p(
, )
, )
p(
, )div
x u dx
1
(u u
) dx,
, )
+ ( ) x p(
(5.14)
( ) 1 x p(
) dx
, ) (u u
[
]
(
)
2
W 1,2 (;R3 ) +
c |x |, |x | u u
E(, |
, ) dx
u x p(
1
t p(
, )
, )
dx
)
1(
+u
dx +
x p(
(
)
t p(
, )
, )
p(
, )div
x u dx+
dx.
2W 1,2 (;R3 ) + K(, )
u u
E(, |
, )
(5.15)
Step 7:
Summing up the estimates (5.5), (5.8), (5.9), (5.12 - 5.15), we can rewrite the relative entropy inequality (5.4) in the form
)
(
1
2
| + E(, |
|u u
, ) (, ) dx
(5.16)
2
)
(
x u
) : (x u x u
) S(, x u) : x u
dx dt
+
S(, x u) : x u S(,
)
(
q(, x ) x q(, x ) x
dx dt
+
)
]
[
(
1
(
)
p(, ) divx u
dx dt
+
p(
, )
(
)
+
0
]
1[
+u
dx dt
x p(
t p(
, )
, )
35
(
)[
]
) + s(
x dx dt
s(
, )(
, )(
)
t + u
]
1[
+u
dx
x p(
, )
, )
(
) t p(
(
)[
]
) + s(
x dx
s(
, )(
, )(
)
t + u
[
]
1
t + u
[t + u
x dx + (
x ] dx,
( ) s(
, )
) p(
, )
1
[t + u
x ] dx = (
dx.
(
) p(
, )
) p(
, )div
xu
[
]
t + u
x dx
( ) s(
, )
(5.17)
(5.18)
[
]
+u
dx ( ) p(
x s(
dx
( ) t s(
, )
, )
, )div
xu
[ (
)
(
)]
x )
x
x )
1
q(
,
q(
,
x u
) : x u
= ( )
S(,
divx
dx
dx
( ) p(
, )div
xu
Seeing that
(
)
p(
) p(
) p(, ) divx u
p(
,
)
,
)(
,
)(
dx
dx
E(, |
, )
L ()
cdivx u
x u
) : (x u x u
)
S(, x u) : x u S(,
+
36
(5.19)
) : x u
dxdt
S(, x u) : x u
S(, x u
(
q(, x ) x q(, x ) x
+
)
x )
x )
x q(,
q(
,
dxdt
+
x ( )
+( )
2
(
2W 1,2 (;R3 ) + K(, )
u u
)
]
1
|2 + E(, |
|u u
, )
dx dt
2
5.3
Step 1
Concerning the viscous term, we shall investigate separately the cases 0 < < and .
In the rst case, we have
(
)
1{0<<}
S(, u) : u S(, u) :
u + S(,
u) : (
u u) +
S(,
u) :
u
(
= 1{0<<}
(
)
) : x u
) +
S(, x u
S(, x u) : x u + S(, x u) S(,
u) : x (u u
[(
)
)
(
) +
S(, u) S(,
u : x (u u
S(, u) : u S(,
u) :
u
(
)
)
]
(
x u
) +
) S(,
) : x u
+ S(,
u) S(,
u) : x (u u
S(, x u
1{}
S(,
u) :
u
S(, u) : u S(, u) :
u + S(,
u) : (
u u) +
1{0<<}
[ (
)
( S(,
u) S(,
u) )
) +
)
S(, u) S(,
u) : x (u u
: x (u u
( u) S(, u) )
]
S(,
+( )
: x u
Putting together both two inequalities, taking into account, (1.4), (1.25), (5.2), (5.3), Young and
Holder inequalities, and employing Lemmas 3.1, 3.2, we nally deduce that
)
(
S(, u) : u S(, u) :
u + S(,
u) : (
u u) +
S(,
u) :
u dxdt (5.20)
= 1{}
37
)
dxdt,
)2 + E(, | , )
(u u
(
)
2 + q(,
log )
q(, log )
(log log )
= ()|(log
log ))|
,
log )
(log log ).
+( )q(
Whence,
)
(
q(,
)
q(, )
q(
,
q(, )
( ) +
: dxdt
+
(5.21)
| + E(, |
|u u
, ) (, ) dx
2
(
(
)
2
+
() x log x log
2
L ((0, ))
c
0
(5.22)
)
(
)
1
dxdt
)2 + E(, | , )
(u u
2
for a. a. (0, T ). We conclude by applying the Gronwall lemma to this inequality. Theorem 2.4 is
proved.
References
[1] S. E. Bechtel, F.J. Rooney, and M.G. Forest. Connection between stability, convexity of internal
energy, and the second law for compressible Newtonian fuids. J. Appl. Mech., 72:299300, 2005.
[2] D. Bresch and B. Desjardins. Stabilite de solutions faibles globales pour les equations de NavierStokes compressibles avec temperature. C.R. Acad. Sci. Paris, 343:219224, 2006.
[3] D. Bresch and B. Desjardins. On the existence of global weak solutions to the Navier-Stokes equations
for viscous compressible and heat conducting uids. J. Math. Pures Appl., 87:5790, 2007.
38
[4] J. Carrillo, A. J
ungel, P.A. Markowich, G. Toscani, and A. Unterreiter. Entropy dissipation methods
for degenerate parabolic problems and generalized Sobolev inequalities. Monatshefte Math., 133:1
82, 2001.
[5] C.M. Dafermos. The second law of thermodynamics and stability. Arch. Rational Mech. Anal.,
70:167179, 1979.
[6] B. Desjardins. Regularity of weak solutions of the compressible isentropic Navier-Stokes equations.
Commun. Partial Dierential Equations, 22:9771008, 1997.
[7] R.J. DiPerna and P.-L. Lions. Ordinary dierential equations, transport theory and Sobolev spaces.
Invent. Math., 98:511547, 1989.
[8] B. Ducomet and E. Feireisl. The equations of magnetohydrodynamics: On the interaction between
matter and radiation in the evolution of gaseous stars. Commun. Math. Phys., 266:595629, 2006.
[9] S. Eliezer, A. Ghatak, and H. Hora. An introduction to equations of states, theory and applications.
Cambridge University Press, Cambridge, 1986.
[10] J.L. Ericksen. Introduction to the thermodynamics of solids, revised ed. Applied Mathematical
Sciences, vol. 131, Springer-Verlag, New York, 1998.
[11] L. Escauriaza, G. Seregin, and V. Sverak L3, -solutions of Navier-Stokes equations and backward
uniqueness. Russian Mathematical Surveys, 58 (2):211250, 2003.
[12] C. L. Feerman. Existence and smoothness of the Navier-Stokes equation. In The millennium prize
problems, pages 5767. Clay Math. Inst., Cambridge, MA, 2006.
[13] E. Feireisl. On compactness of solutions to the compressible isentropic Navier-Stokes equations when
the density is not squared integrable. Comment. Math. univ. Carolinae42: 83-98, 2001
[14] E. Feireisl. Dynamics of viscous compressible uids. Oxford University Press, Oxford, 2004.
[15] E. Feireisl. Relative entropies in thermodynamics of complete uid systems. DCDS B, to appear.
[16] E. Feireisl, Bum Ja Jin, and A. Novotn
y. Relative entropies, suitable weak solutions, and weak-strong
uniqueness for the compressible Navier-Stokes system. J. Math. Fluid Mech., on line rst
[17] E. Feireisl and A. Novotn
y. Singular limits in thermodynamics of viscous uids. Birkhauser-Verlag,
Basel, 2009.
[18] E. Feireisl and A. Novotn
y. Weak-strong uniqueness property for the full Navier-Stokes-Fourier
system. Arch. Rat. Mech. Anal. 2012, on line rst
[19] E. Feireisl, A. Novotn
y, and H. Petzeltova. On the existence of globally dened weak solutions to
the Navier-Stokes equations of compressible isentropic uids. J. Math. Fluid Dynamics, 3:358392,
2001.
[20] E. Feireisl, A. Novotn
y, and Y. Sun. Suitable weak solutions to the NavierStokes equations of
compressible viscous uids. Indiana Univ. Math. J., 2011. To appear.
39
40
[38] A. Novotny, M. Padula. Lp approach to steady ows of viscous comprtessible uids in exterior
domains. Arch. Rat. Mech. Anal., 126: 243297, 1994
[39] A. Novotn
y and I. Straskraba. Introduction to the mathematical theory of compressible ow. Oxford
University Press, Oxford, 2004.
[40] L. Poul Existence of weak solutions to the Navier-Stokes-Fourier system on Lipschitz domains.
Discrete Contin. Dyn. Syst., Proceedings of the 6th AIMS International Conference, 834-843, 2007
[41] L. Poul On the dynamics of uids in meteorology. Central European Journal in Mathematics,
6(3):422-438, 2008
[42] G. Prodi Un teorema di unicit`a per le equazioni di Navier-Stokes. Ann. Mat. Pura Appl., 48:173182,
1959.
[43] L. Saint-Raymond. Hydrodynamic limits: some improvements of the relative entropy method. Annal.
I.H.Poincare - AN, 26:705744, 2009.
[44] D. Serre. Variation de grande amplitude pour la densite dun uid viscueux compressible, Physica
D, 48: 113128, 1991.
[45] J. Serrin. On the interior regularity of weak solutions of the Navier-Stokes equations. Arch. Rational
Mech. Anal., 9:187195, 1962.
[46] V.A. Solonnikov and V. E. Scadilov On a boundary value problem for a stationary system of NavierStokes equations. Proc. Steklov Inst. Math., 125, 186199, 1973.
[47] R. Temam. Navier-Stokes equations. North-Holland, Amsterdam, 1977.
[48] S. Ukai. The incompressible limit and the initial layer of the compressible Euler equation. J. Math.
Kyoto Univ., 26(2):323331, 1986.
[49] A. Valli, W. Zajaczkowski. Navier-Stokes equations for compressible uids: Global existence and
qualitative properties of the solutions in the general case Commun. Math. Phys., 103:259296, 1986.
[50] S. Wang and S. Jiang. The convergence of the Navier-Stokes-Poisson system to the incompressible
Euler equations. Comm. Partial Dierential Equations, 31(4-6):571591, 2006.
41