The Navier-Stokes equations for the motion of

compressible, viscous uid ows with the no-slip

boundary condition
Mikhail Perepelitsa
Abstract
The Navier-Stokes equations for the motion of compressible, viscous uids
in the half-space R
3
+
with the no-slip boundary condition are studied. Given a
constant equilibrium state ( , 0), we construct a global in time, regular weak
solution, provided that the initial data
0
, u
0
are close to the equilibrium state
when measured by the norm
|
0
|
L
+|u
0
|
H
1
and discontinuities of
0
decay near the boundary of R
3
+
.
0.1 Introduction
We consider a model for the motion of a compressible, isothermal, viscous
ow based on the Navier-Stokes equations. With (t, x) and u(t, x) being the
density and the velocity of the uid, the model consist of the equations:

t
+ div (u) = 0, (1)

t
(u) + div (u u) ( +) div u u +P = f , (2)
3 + 2 0, > 0,
(t, x) R
+
, P() = A, A > 0,
1
and a set of initial and boundary conditions:
((0, x), u(0, x)) = (
0
(x), u
0
(x)) , x , (3)
u(t, x) = 0, (t, x) R
+
. (4)
There is an extensive literature concerning dierent aspects of the problem
(1) (4). For the detailed discussion of the results we refer the reader to the
recent monographs [3, 13]. We shortly mention some of them. It is known that
if the initial data of the problem are smooth then the problem is well-posed.
Moreover, a unique, global solution exists if the initial data are close to a
static equilibrium state, measured in strong norms, for example in H
3
_
R
3
+
_
,
see [11, 15]. On the other hand, there is a well-developed theory of weak
solutions of the problem (1) (4) and other related problems, see [8, 3]. A
typical result is contained in the following theorem.
Theorem (P.-L. Lions, [8]). Suppose that
9
5
and C
2+
, > 0. Sup-
pose that the initial data (
0
, m
0
) satisfy L

() , |m
0
|
2
/
0
L
1
() , where
we agree that m
0
= 0 on {
0
() = 0} . Then there is a global weak solution of
the problem (1)(4), (, u), such that (0, ) =
0
() and (0, )u(t, ) = m
0
Moreover, for any t > 0 the energy inequality holds.
_

_
1
2
(t, )|u(t, )|
2
+
A(t, )

1
_
+
_
t
0
_

|u|
2
+ ( +)| div u|
2

_
1
2
|u
0
|
2
+
A

0
1
_
.
Solutions constructed in the above theorem have somewhat limited reg-
ularity properties: L

(R
+
: L

()) and u L
2
_
R
+
: W
1,2
()
_
3
, and
thus, may incorporate some non-physical phenomena.
For the Cauchy problem, i.e. when the ow occupies the whole space
R
3
, global existence of weak solutions that remain near a static equilibrium
sate, ( , 0), was proved in [4], see also [5] for related results. In contrast
with the result of [11], solutions built in [4] are essentially weak; the density
is an element of L

. On the other hand, solutions possess many favorable

properties, such as, impossibility of spontaneous formation of vacuum, the
fact which is being implicitly assumed when equations (1) (2) are used to
model a motion of real uids.
2
Theorem (D. Ho, [4, 5]). Let = R
N
, N = 2, 3. Let > 0 and L > 0 be
given. There is a positive number c = c(N) and a pair of positive numbers
A, C depending on (, , , L, N, c), with the property that if
+ c (5)
and the initial data (
0
, u
0
) satisfy bounds
0
0
, a.e. R
3
+
,
_
R
N
|u
0
(y)|
2
+ (
0
(y) )
2
dy A
and
|u
0
|
L
2
N
(R
N
)
N L,
then, a global weak solution (, u) of the problem (1)(3) exists for which
||
L

(R
+
R
N
)
C ,
u L

({t : t > } R)
N
, > 0.
(We refer the reader to [5] for the complete statement of the Theorem.)
The analogous result was obtained for ows in domains with boundaries
under the Navier boundary condition, i.e. the condition that tangential veloc-
ity at the boundary is proportional to the tangential component of the stress,
see [5].
In this work we present a development of the existence theory of the near
equilibrium weak solutions initiated in [4] to the problems with the no-slip
boundary condition (4). The density component of the weak solution that we
construct is L

away form the boundaries and such that discontinuities in

(t, ) decay near the boundary. Specically, we measure (t, ) by the norm
(t, )

+|(t, )|
L

L
2, ]0, 1[,
where

is dened in (13). The localization of discontinuities in (t, ) inside

the domain corresponds to a physical situation when motion of a uid results
from disturbances that occur in the interior of the domain. At the level of
technical description of the proof, the introduction of the above functional
to measure the density is dictated by the fact that the L

norm along is
3
not suitable to control sound waves reected from the boundary. We prove a
local in time existence of weak solutions in this class with large initial data
and a global in time existence of weak solutions that are close to a static
equilibrium case initially. In the later case, for weak solutions to remain
near the equilibrium state globally in time we impose a certain structural
restriction on the model, i.e. on the relative size of and , given by (6),
which guarantees that sound waves reected from the boundary are, in fact,
weaker than incident waves. More detailed discussion of the result is given in
the next section.
0.2 Statement of the result
We will work with densities that have C

regularity at the boundary as it is

dened in the next denition.
Denition 1. Let L

_
R
3
+
_
be a closure of the space C
1
_
R
3
+
_
in the norm

+| |
L

(R
3
+
)
, where

is dened in (13).
We prove the following results.
Theorem 1 (Local existence of weak solutions). Let > 0. Let u
0
H
1
0
_
R
3
+
_
,

0
L

_
R
3
+
_
,
0
L
2
_
R
3
+
_
, ]0,
1
4
[ and
c <
0
< c
1
a.e. in R
3
+
,
for some c > 0. Then, there are T > 0 and a weak solution, (, u), of the
problem (1) (4), dened on [0, T] R
3
+
and there are constants c
i
, i = 1, 2
such that
c
1
< (t, x) < c
1
1
, t [0, T], x a.e. in R
3
+
,
sup
t[0,T]
_
(t, )

+|(t, ) |
L
2
(R
3
+
)
+|u(t, )|
H
1
0
(R
3
+
)
_
c
2
.
Theorem 2 (Global existence of weak solutions). For any > 0 and ]0,
1
4
[
there are c
0
= c
0
() > 0, c
i
= c
i
( , , , , A), i = 1, 2, such that if
c
0

+ 3
< 1, (6)
and a pair of measurable functions (
0
, u
0
) veries a smallness assumption:
|
0
|
L
2
(R
3
+
)
+|
0
|
L

(R
3
+
)
+
0

+|u
0
|
H
1
(R
3
+
)
c
1
, (7)
4
then, there exists a weak solution, (, u), of the problem (1) (4), dened
for all times t > 0. Moreover, for t R
+
, the following estimates hold with
independent of time c
2
> 0.
osc (t, ) +(t, )

< c
2
( osc (t, ) +(t, )

+|u
0
|
H
1
0
),
|u(t, )|
H
1
0
(R
3
+
)
+|(t, ) |
L
2
(R
3
+
)
c
2
(|u
0
|
H
1
0
(R
3
+
)
+|
0
|
L
2
(R
3
+
)
),
L
6
_
R
+
R
3
+
_
,

x
u L
2
(R
+
: L
6
_
R
3
+
_
),
(t)(u
t
+u u) L

(R
+
: L
2
_
R
3
+
_
) L
2
(R
+
: H
1
0
_
R
3
+
_
),
_

_
(8)
where (t) = min{1, t}.
Remark 1. Since the solution, constructed in theorem 2, is such that the
oscillations in density are small, there is no loss of generality in assuming
the pressure law P = A instead of the isentropic law, P = A

, 1.
Indeed, the derivation of a priori estimates in case > 1 is identical to case
= 1. Moreover, the strong convergence of the sequence of the approximate,
classical solutions for > 1 is established by the Lions-Feireisl theory, see [3].
Remark 2. We will prove theorem 2. Theorem 1 follows by a rather straight-
forward argument as all apriori estimates that we establish for theorem 2 can
be obtained locally in time for solutions with arbitrary large initial data.
The framework of the analysis was established in works [4, 8]. We shortly
describe the new issues appearing in the problem with the no-slip boundary
conditions. First, we notice that unlike the situation for the Cauchy problem,
L

norm alone is not well-suited to measure the density of a solution. Indeed,

the Navier-Stokes equations (2) can be written as a problem
( +)div u +u = a +( ),
u = 0, R
3
+
,
_
(9)
where a = D
t
u the acceleration. Assuming for a moment that a = 0, the
5
divergence of u can be represented as
( + 2) div u(t, x) =
+
+ 3
((t, x) )

2
+ 2
_
R
3
+

y

x
G(x, y)((t, y) )dy, (10)
where G(x, y) is the Greens function for the Laplaces equation in R
3
+
, see
section 0.9 for details. It can be seen from the above formula that div u is not
bounded in sup-norm if is a generic function in L
2
L

_
R
3
+
_
. The di-
vergence, div u, may blow-up at boundary points. Moreover, the singularities
caused by ( ) are unlikely to be balanced by the acceleration, a, because
the later vanishes at the boundary of the domain. On the other hand, div u
is the rate of the production of mass which, when unbounded, may cause the
blow-up of the density in L

norm. A better substitute is a norm

||
X
0
:=

+| |
2
+ osc . (11)
where

is dened by (13). It can be shown that

| div u|
X
0
c||
X
0
,
see Lemma 13. In order to investigate the question of long time stability we
have to show that the motion caused by the reection of sound waves from the
boundary can not destabilize a solution which is close to the static equilibrium
( , 0) initially. We consider the reection in the context of the linear elliptic
problem (9). We say that sound waves reected from the boundary of R
3
+
are weaker then incident waves when measured by | |
X
if
|( + 2) div u ( )|
X
c| |
X
, and 0 < c < 1. (12)
For X
0
introduced in (11) we were able to show that the above estimate holds
with
c =
c
0
()
+ 2
,
where the function c
0
() is of the order
1
. We do not know if this number
is smaller than 1 for the full range of {(, ) : > 0, 3 + 2 > 0} and this
is the reason for including a structural condition (6) in the theorem 2. On
the other hand, the constant c is less than 1 if the ratio

+
2
3

is small. In
6
Hydrodynamics the quantity +
2
3
is called a second viscosity coecient
or bulk viscosity. In certain situations, such that motion of a gas with large
relaxation times, the values of /( +
2
3
) are in fact small, see Section 81 of
[6]. For that reason our result may be thought of as a model for motion of
such type of gases. We use the estimate (12) with X = X
0
in the equation
(1) to show that the oscillations of the density in reected sound waves are
damped by pressure.
A main part of the paper is devoted to derivation of a priori estimates
for a generic classical solution of (1)(2). A weak solution is constructed as
a limit of global smooth solutions with appropriately smoothed initial data
using Lions-Feireisl theory, [8, 3].
0.3 Functional setting
By B
r
(x), r > 0, x R
3
, we denote a ball with radius r, centered at x R
3
+
.
We use symbol to denote the spacial gradient of a function and D
2
the set of
all spacial second derivatives. The norm | |
p
is the usual L
p
norm. We use the
standard notation W
k,p
_
R
3
+
_
, k N, 1 p < + for the space of weakly
dierentiable, up to the order k, functions, with derivatives from L
p
_
R
3
+
_
space. We use notation H
1
:= W
1,2
. In this paper we will abbreviate L
p
_
R
3
+
_
to L
p
and use the same notation for norms of scalar and vector functions.
Denote by
[u]

= sup
x,yR
3
+
, x=y
|u(x) u(y)|
|x y|

, ]0, 1[,
u

= sup
xR
3
+
,yR
3
+
|u(x) u(y)|
|x y|

, ]0, 1[, (13)

various Holder semi-norms. The following estimates are well-known, see
[2](Theorem 7.10, Theorem 7.17)
Lemma 1. Let u be a locally integrable function such that u L
2
_
R
3
+
_
and having zero trace on the boundary R
3
+
. Then, u L
6
_
R
3
+
_
and there is
c > 0, independent of u, such that
|u|
L
6 c|u|
L
2.
7
Lemma 2. Let u be a locally integrable function with u L
p
_
R
3
+
_
, p > 3.
Then, there is c = c(p) such that for a.e. x, y R
3
+
it holds
|u(x) u(y)| c|x y|

|u|
L
p,
where = 1 3p
1
.
Denition 2. A pair of functions
(, u) = ((t, x), u
1
(t, x), u
2
(t, x), u
3
(t, x))
is called a weak solution of (1)-(4) if
, u
i
, u
i
L
1
loc
_
R
+
R
3
+
_
, i = 1, 2, 3,
u
k
u
l
L
1
loc
_
R
+
R
3
+
_
, i, k, l = 1..3,
u L
2
_
R
+
R
3
+
_
,
u = 0, on R
3
+
,
_
and for all test functions ,
i
C

_
[t, T] : C

0
_
R
3
+
__
, i = 1, 2, 3, with
0 t < T < + it holds (summation over the repeated indexes is assumed)
_ _
R
+
R
3
+

t
+u
_
R
3
+
(, )(, )

T
t
= 0,
_ _
R
+
R
3
+
u
k

k
+u
k
u
j

_ _
R
+
R
3
+
( +) div udiv +
k
u
l

l
+ (P

P)
k

_
R
3
+
(, )u
k
(, )(, )

T
t
= 0.
To simplify the presentation we assume that the constant A = 1 in (2). It is
always possible to reduced to this case through the substitution (t, x, , u)
(a
2
t, ax, , au), a = A

1
2
, without changing the viscosity coecients.
8
0.4 The Lame equations
In this section we recall some elliptic estimates. The principal part of (2) is
an elliptic system of Lame equations (14). Consider the problem
( +)div u +u = f , R
3
+
,
u = 0, R
3
+
,
_
(14)
with the conditions > 0, + > 0. Here, f = (f
1
(x), f
2
(x), f
3
(x)). The
system is
_
H
1
0
_
3
elliptic, see Chap. 3, sec. 7 of [12], meaning that the
bilinear form
a(u, v) =
_
R
3
+
( +) div udiv v +u : v,
is coercive, i.e.
a(u, u) |u|
2
L
2
(R
3
+
)
.
This condition is sucient to imply the existence of the strong solution, The-
orem 2.1 in [12], Chap. 3.
Lemma 3 (Elliptic estimates I). Let f L
2
_
R
3
+
_
H
1
0
_
R
3
+
_
. Then, there
is a unique strong solution of (14), such that
|u|
L
2
(R
3
+
)
c|f |
H
1
0
(R
3
+
)
.
|D
2
u|
L
2
(R
3
+
)
c|f |
L
2
(R
3
+
)
.
Lemma 4 (Elliptic estimates II). If additionally f L
p
loc
_
R
3
+
_
, 1 < p <
then there is c = c(p) such that
|D
2
u|
p,B
1
(x)H
c
_
|f |
p,B
2
(x)H
+|f |
2
+|f |
H
1
0
_
.
The following Lemma is a well-known fact, the proof of which is based on
the Calderon-Zygmund estimate on singular integral operators.
Lemma 5 (Elliptic estimates III). Suppose that in the problem (14) f = P,
for some P L
p
, p ]1, [. Then, the problem has a unique weak solution
u L
3p
3p
, p < 3 or u L

loc
, p > 3, such that
|u|
L
p c|P|
L
p,
for some c = c(p).
9
0.5 1
st
energy estimate
In all subsequent estimate we assume
Hypothesis H
0
. For all (t, x),
(t, x) < M := 10 .
Lemma 6. Let
() =
_


s
2
(s ) ds, 0
and
E(t) =
_
R
3
+
(t, )|u(t, )|
2
/2 + ((t, )).
Then, for any smooth solution (, u) of the problem (1)(4) the following
equality holds.
E(t) +
_
t
0
_
R
3
+
( + 2)| div u(t, )|
2
+| curl u(t, )|
2
= E(0). (15)
The proof of this Lemma is well-known and can be found, for example, in
[4].
0.6 Bogovski operator in a half-space
The estimates of lemma 8 of this subsection can be found in [?]. We give here
an alternative, short proof of these estimates. Consider a problem
v = P, v

R
3
+
= 0.
If H = |x y|
1
a fundamental solution for the Laplaces equation, then
there is a representation formula for
div v = P
_
R
3
+

y

x
H(x, y

)P(y) dy,
where y

= (y
1
, y
2
, y
3
). Let
AP :=
_
R
3
+

y

x
H(x, y

)P(y) dy.
10
Lemma 7. Operator A has the following properties:
A : L
p
(R
3
+
) L
p
(R
3
+
)
is a bounded linear operator. Moreover,
3A2A
2
= 0.
Proof. The rst statement is a standard application of SIO theory. The proof
of the second statement. Firts of all we note that ( div v P) = 0. Then,
we can write
div v P =
_
R
3
+

ny
G(x, y)( div v P)(y) (16)
and
(x( div v P)) = 2( div P).
It follows then that
x( div vP) =
_
R
3
+

ny
G(x, y)y( div vP)(y) dy+2
_
R
3
+
G(x, y)
y
( div vP) dy
and
3( div v P) = x( div v P) +
_
R
3
+

ny
G(x, y)y( div v P)
+ 2
_
R
3
+

x
G(x, y)
y
( div v P).
Combining the last two formulas we obtain
3( div v P) =
_
R
3
+

ny
G(x, y) (y x)( div v P)(y))
+ 2
_
R
3
+

x
G(x, y)
y
( div v P)(y).
Using the fact that for y H,
x

ny
G(x, y) (y x) = 2
ny
G(x, y) and
exploiting (16) we conclude that
div v P = 2
_
R
3
+

x
G(x, y)
y
( div v P)(y)
which implies that
3( div v P) = 2A div v P.
11
Consider now the problem
div v = P, v

R
3
+
= 0.
Let v
i
be dened by
v
1
= P, v
1

R
3
+
= 0
and
v
2
= AP, v
2

R
3
+
= 0.
Each problem is uniquely solvable if for example P L
p
and we can write the
following representation formulas,
div v
1
= P AP
and
div v
2
= AP A
2
P =
5
2
AP
One can easily see then that
v = v
1
+
2
5
v
2
. (17)
We now prove the following lemma.
Lemma 8. For v introduced above,it holds:
|v|
p
c(p)|P|
p
, 1 < p < .
Moreover, if P = div g with g L
p
(H), g

R
3
+
= 0 then
|v|
p
c(p)|g|
p
, 1 < p < .
Proof. The rst part of the lemma follows from the representation formula
v
1
=
_
R
3
+
G(x, y)
y
P(y) dy
and the standard L
p
estimates on SIO. Similarly, for g as described in the
lemma one derives a formula
v
1
= P.V.
_
R
3
+

y
i

y
G(x, y)g
i
(y) dy +c
i
g
i
(x),
12
with c
i
=
_
|x|=1
x
i
xdS
x
. Thus, we have the estimate |v
1
|
p
c(p)|g|
p
,
1 < p < . On the other hand, formally writting A = I div ()
1
we
have a problem for v
2
:
v
2
= div g div ()
1
div g, v
2

R
3
+
= 0.
We deduce the estimate |v
2
|
p
c(p)|g|
p
from the estimate on v
1
.
0.7 A L
6
estimate on the density
Let v be a solution of the problem:
v

R
3
+
= 0, div v = ( )
5
.
We have, according to the estimates of the previous section,
|v|
p
c(p)| |
5
5p
.
Take the test function = v in (2) to obtain the following identity
_
t
0
_
R
3
+
u v+
_
t
0
_
R
3
+
(+) div v div u+u : v =
_
t
0
_
R
3
+
( )
6
.
(18)
We have
_
t
0
_
R
3
+
u : v c

_
t
0
_
R
3
+
|u|
2
+
_
t
0
_
R
3
+
|v|
2
c

_
t
0
_
R
3
+
|u|
2
+
_
t
0
| |
10
10
c

_
t
0
_
R
3
+
|u|
2
+
_
t
0
_
R
3
+
| |
6
. (19)
Also,
_
t
0
_
R
3
+
u u : v
_
t
0
|u|
2
6
|v|
3/2

_
t
0
|u|
2
2
| |
5
15/2
c
_
t
0
|u|
2
2
. (20)
13
On the other hand,
_
t
0
_
R
3
+
| u||v| c
_
t
0
| u|
2
|v|
6/5
c
_
t
0
| u|
2
| |
5
6
C

_
t
0
| u|
2
2
+
_
t
0
| |
6
6
. (21)
Combining above estimate and choosing the appropriate value of we obtain
the next lemma.
Lemma 9. There is c > 0 independent of t > 0 such that
_
t
0
_
H
| |
6
c
_
E(0) +E
2
(0) +
_
t
0
| u|
2
2
_
. (22)
0.8 Energy estimates of higher order
The following Lemmas are proved in exactly the same way as Lemma ?? from
[4]
Lemma 10. Under hypothesis H
0
, for t > 0 it holds:
sup
s[0,t]
_
R
3
+
|u(s, )|
2
+
_
t
0
_
R
3
+
| u(s, )|
2
c(|u
0
|
2
, |u
0
|
2
, |
0
|
2
)
+
_
t
0
_
R
3
+
|u|
3
.
Corollary 1. For any > 0 there is a C

such that
sup
s[0,t]
_
R
3
+
|u(s, )|
2
+
_
t
0
_
R
3
+
| u(s, )|
2
c(|u
0
|
2
, |u
0
|
2
, |
0
|
2
)
+
_
t
0
| u(s, )|
2
2
+|(s, ) |
6
6
+C

_
t
0
_
|u(s, )|
6
2
+|u(s, )|
2
2
_
. (23)
Proof. Indeed, by the Holder inequality and elliptic estimates of lemmas 3, 5
we have:
|u|
3
6
|u|
3/2
2
|u|
3/2
6
c|u|
3/2
2
(| u|
2
+| |
6
)
3/2
| u|
2
2
+| |
6
6
+C

_
|u|
6
2
+|u|
2
2
_
(24)
and the corollary follows from the previous lemma and lemma 9.
14
Combining the estimate of lemma 15 with the previous corollary and re-
stricting |u
0
|
2
+|u
0
|
2
+|
0
|
2
in a suitable way we obtain the next estimate.
sup
s[0,t]
_
R
3
+
|u(s, )|
2
+
_
t
0
| u(s, )|
2
2
+|(s, ) |
6
6
c(|u
0
|
2
, |u
0
|
2
, |
0
|
2
). (25)
In the above estimate c vanishes with |u
0
|
2
+|u
0
|
2
+|
0
|
2
.
Lemma 11. For (t) = min{1, t} it holds:
sup
s[0,t]
(s)
_
R
3
+
| u|
2
+
_
t
0
_
R
3
+
(s)| u(s, )|
2
c +
_
t
0
_
R
3
+
|u(s, )|
3
+
_
t
0
_
R
3
+
(s)|u(s, )|
4
,
with c = c(|u
0
|
2
, |u
0
|
2
, |
0
|
2
).
Corollary 2. For any > 0 there is C

such that
sup
s[0,t]
(s)
_
R
3
+
| u|
2
+
_
t
0
_
R
3
+
(s)| u(s, )|
2
c +
_
t
0
| u|
2
2
+C

_
t
0
_
|u(s, )|
6
2
+|u(s, )|
2
2
+|(s, ) |
6
6
_
+ sup
s[0,t]
(s)| u|
2
2
__
t
0
|u(s, )|
2
2
+| u(s, )|
2
2
_
. (26)
Proof. The corollary easily follows form the estimates on |u|
3
3
of the corollary
1 and a similar estimate on
|u|
4
4
|u|
2
|u|
3
6
c|u|
2
{| u|
2
+| |
6
}
3
.
Using again the smallness assumption on |u
0
|
2
+ |u
0
|
2
+ |
0
|
2
we
derive an apriori estimate:
sup
s[0,t]
(s)
_
R
3
+
| u|
2
+
_
t
0
_
R
3
+
(s)| u(s, )|
2
c, (27)
with c depending on the same parameters as the constant in (25).
15
0.9 Representation formulas for the solution of (14)
Now, we derive a representation formula for u. Let G be a Greens matrix for
the problem (14), i.e.
u(x) =
_
R
3
+
G(x, y)f (y) dy.
The explicit expression for the Greens matrix can be found, for example, in
[14]. Let
A =
+
2( + 2)
, B =
+ 3
+
and
ik
be the Kronecker symbol. Then,
G
k
i
(x, y) = A
__
B
ik
+ (x
i
y
i
)

y
k
__
1
4|x y|

1
4|x y

|
__
+x
1
_

ik
B
1
y
1

y
k
_
1
2

x
i
1
|x y

|
, i, k = 1, 2, 3. (28)
Let v, w be two vector elds dened for any t > 0 as the solutions of the
following problems.
Lv = u, x H, v

H
= 0
and
Lw = , x H, w

H
= 0,
where L = div +. It follows from the above considerations that
u(t, x) = v +w (29)
and
u =
_
R
3
+
G(x, y) u(t, y) dy, w =
_
R
3
+
G(x, y)((t, y) ) dy. (30)
Now, we compute a representation formula for div w.
For x = (x
1
, x
2
, x
3
) R
3
+
, let x

= (x
1
, x
2
, x
3
). Let
H(x, y) =
1
4|x y|
,
and denote by
G(x, y) = H(x, y) +H(x, y

) (31)
16
the Greens function for the Laplaces equation in R
3
+
. Consider an elliptic
problem (14) where we set
f = ( ).
Let
F = ( + 2) div w( ), (32)
be the notation for the viscous ux. Applying div to (14) we derive:
F = 0, (33)
and the following integral representation holds (the dependence of functions
on t is not written for notational convenience).
F(x) =
_
R
3
+

ny
G(x, )F(). (34)
Using (33), equations (14) can be written in the following form.

_
+
2( + 2)
Fx +u
_
=

+ 2
( )
and so,
w(x) +
+
2( + 2)
F(x)x =
+
2( + 2)
_
R
3
+

ny
G(x, y)F(y)y dS
y
+
_
R
3
+
G(x, y)
_

+ 2

y
((y) )
_
dy. (35)
We set

1
=
+
2( + 2)
,
2
=

+ 2
,
3
=
+ 3
2( + 2)
. (36)
We take div of the last equations and use integral representation (34) for F
to get the following equation (here and below the summation over repeated
indexes is assumed).
(2
1
+
3
)F(x) =
2
((x) )
+
1
_
R
3
+

ny

x
i
G(x, y)(y
i
x
i
)F(y) dS
y
+
2
_
R
3
+

x
G(x, y)
y
((y) ) dy.
One can easily verify that

ny

x
i
G(x, y)(y
i
x
i
) = 2
ny
G(x, y), y R
3
+
.
17
We use this identity in the last equation together with (34) to obtain the
following representation formula for F.

3
F(x) =
2
((x) )
2
_
R
3
+
div
y

x
G(x, y)((y) ) dy
=:
2
P
1
(x). (37)
Now we investigate the regularity of v.
Lemma 12. For any s > 0 it holds
|v(s, )|

ca
1
(s),
[v(s, )]

ca
1
(s),
where
a
1
(s) = |(s, ) u(s, )|
4
+|(s, ) u(s, )|
2
+|(s, ) |
6
(38)
with c depending on the same parameters as the constant in (27). Moreover,
for any 0 < s < t, > 0 it holds
_
t
s
a
1
()d c(, |u
0
|
2
, |u
0
|
2
, |
0
|
2
) + 10 +(t s), (39)
where c vanishes if is xed and all other parameters go to 0. Moreover, for
all t > 0, m < 0 :
_
t
0
e
ms
a
1
(s) c(m, |u
0
|
2
, |u
0
|
2
, |
0
|
2
), (40)
where c has the same properties as the one in (39).
Proof. The proof goes by the classical embedding lemmas 1, 2. Indeed, for
any x H we have
|v(x) (v)
B
1
(x)H
| c|D
2
v|
4,B
1
(x)H
and
|(v)
B
1
(x)H
|
L
c|v|
2,B
1
(x)H
.
On the other hand, elliptic estimates of lemma 4 imply ( note, that it fol-
lows from the energy estimates (6) and the equation (2) that | u(s, )|
H
1
0

|u(s, )|
2
+|(s, ) |
2
) that
|v|
2
c(| u|
2
+|u|
2
+| |
2
)
18
and
|D
2
v|
4,B
1
(x)H
c (| u|
4
+| u|
2
+|u|
2
+| |
6
) .
The rst statement of the lemma follows. To prove the second we notice that
| u|
4
| u|
1/4
2
| u|
3/4
6
and for s < t
_
t
s
| u|
4

_
t
0
| u|
1/4
2
| u|
3/4
6

3/8

3/8

_
t
s
_

1
| u|
2
2
+
1
s| u|
2
6
+
3/4
_
(41)
and
_
t
s
| u|
2
+| |
6

_
t
s

1
| u|
2
2
+
5
| |
6
6
+.
We conclude by applying estimates (25) and (27) and a trivial estimate
_
t
s

3/4
4 +(t s).
We will work out the regularity of w later, after developing some potential
estimates in the next subsection.
0.10 Some potential estimates
Let P
1
(x) be the function from (37). We will need the following lemma.
Lemma 13. For any ]0, 1[ and > 0 there are c > 0, c

> 0, independent
of (, , , t), such, that
P
1

c (|() |
2
+

) , (42)
and for any x R
3
+
,
|P
1
(x)|

+c

|() |
2
. (43)
Proof. We proof only the rst part of the Lemma. The proof for the second
part goes along the same line of arguments. Let x
1
, x
2
R
3
+
, |x
1
x
2
| <
and set B
2
= B(x
1
, 2), B
1
= B(x
1
, 4), B = B(x
1
, 2|x
1
x
2
|),
S
2
=
_
R
3
+
B
2
_
\ R
3
+
,
19
and
S =
_
R
3
+
B
_
\ R
3
+
.
We can write the following representation for
P
1
(x
1
) P
1
(x
2
) = ((x
1
) (x
2
))

_
R
3
+
\B
2
{ div
y

x
G(x
1
, y) div
y

x
G(x
2
, y)} ((y) )
+
_
S
2
{
x
G(x
1
, y)
x
G(x
2
, y)} n
y
((y) ) dS
y

_
B
2
\B
1
{ div
y

x
G(x
1
, y) div
y

x
G(x
2
, y)} ((y) (x
1
))

_
B
1
{ div
y

x
G(x
1
, y) div
y

x
G(x
2
, y)} ((y) (x
1
))
+
_
S
2
{
x
G(x
1
, y)
x
G(x
2
, y)} (n
y
)((y) (x
1
)) dS
y
. (44)
We set x

2
to be the projection of point x
2
onto R
3
+
. and consequently,
P
1
(x
1
) P
1
(x
2
) = ((x
1
) (x
2
))

_
R
3
+
\B
2
{ div
y

x
G(x
1
, y) div
y

x
G(x
2
, y)} ((y) )
+
_
S
2
{
x
G(x
1
, y)
x
G(x
2
, y)} n
y
((x
1
) ) dS
y

_
B
2
\B
1
{ div
y

x
G(x
1
, y) div
y

x
G(x
2
, y)} ((y) (x
1
))

_
B
1
\B
{ div
y

x
G(x
1
, y) div
y

x
G(x
2
, y)} ((y) (x
1
))

_
B
div
y

x
G(x
1
, y)((y) (x
1
))
+
_
B
div
y

x
G(x
2
, y)((y) (x

2
))
+ ((x

2
) (x
1
))
_
S
div
y

x
G(x
2
, y) n
y
dS
y
. (45)
Finally, we split G = H + H

as in (31) and use the fact that H is a fun-

damental solution of the Laplaces equation. With the notation K

(x, y) :=
20

y

x
H(x, y

) we obtain:
P
1
(x
1
) P
1
(x
2
) = ((x
1
) (x

2
))

_
R
3
+
\B
2
{K

(x
1
, y) K

(x
2
, y)} ((y) )
+
_
S
2
{
x
G(x
1
, y)
x
G(x
2
, y)} n
y
((x
1
) ) dS
y

_
B
2
\B
1
{K

(x
1
, y) K

(x
2
, y)} ((y) (x
1
))

_
B
1
\B
{K

(x
1
, y) K

(x
2
, y)} ((y) (x
1
))

_
B
K

(x
1
, y)((y) (x
1
))
+
_
B
K

(x
2
, y)((y) (x

2
))
+ ((x

2
) (x
1
))
_
S

x
G(x
2
, y) n
y
dS
y
. (46)
In the above representation formula we take x
1
R
3
+
and x
2
R
3
+
. The rst
term on the right is bounded by

|x
1
x
2
|

. The second is bounded by

|() |
2
|x
1
x
2
|. The third by osc ()|x
1
x
2
|. The fourth by
osc ()

1
|x
1
x
2
|

and the rest of the terms are bounded by

|x
1
x
2
|

,
because K

possesses property S

. The same bound holds for the last term

on the right, since
|
_
S

x
G(x
2
, y) n
y
dS
y
| c,
with c independent of x
1
, x
2
. Setting = 1 we obtain the required estimate.
0.11 Holder continuity of ow trajectories
In this section we consider the regularity of the ow generated by the velocity
u.
21
Let X(s, x; t) denote the trajectory of ow, i.e.
d
ds
X(s, x; t) = u(t, X(s, x; t)), X(t, x; t) = x, x R
3
+
. (47)
We choose two points x
1
R
3
+
and x
2
R
3
+
and consider trajectories
X(s, x
1
; t) and X(s, x
2
; t) which we abbreviate to X
s
1
and X
s
2
. We use for-
mula (30) in equation (1) to write it as
D
s
(X
s
1
X
s
2
) = v(s, X
s
1
) v(s, X
s
2
) +w(s, X
s
1
) w(s, X
s
2
). (48)
In the next lemma we estimate the right-hand side of the above equation.
Lemma 14.
|v(s, X
s
1
) v(s, X
s
2
)| a
1
(s)|X
s
1
X
s
2
|,
|w(s, X
s
1
) w(s, X
s
2
)| a
2
(s)|X
s
1
X
s
2
|,
where a
1
is dened in (38) and
a
2
= |(s, ) |
2
+|(s, ) |

+(s, )

.
Proof. The rst part of this lemma has been proved in lemma 12. The second
follows easily from the representation formulas for G (28) and the fact that
that the vector eld
k(x) :=
_
R
3
+

y
1
|x y|
(y) dy
satises the estimate
|k(x
1
) k(x
2
)| (|()|
2
+|()|

+()

) |x
1
x
2
|,
where x
1
R
3
+
, x
2
R
3
+
. The proof of this fact is done similarly to the proof
of lemma 8.1 of [9]. The details of the proof are left to the reader.
Now, from the system of ODEs (48) we derive the next lemma.
Lemma 15. For X
t
i
, i = 1, 2, dened in (47) it holds:
|X
s
1
X
s
2
| |x
1
x
2
|(1 + c(, |u
0
|
2
, |u
0
|
2
, |
0
|
2
))e
a(t)(ts)
, s ]0, t[,
a(t) = sup
[0,t]
(10 +c(|u
0
|
2
, |u
0
|
2
, |
0
|
2
) +|(s, ) |

+(s, )

) ,
(49)
where function c vanish when is xed and all other of its arguments approach
zero.
22
Proof. This is the Gronwalls inequality for (48) and the relation (49) follows
from the estimates on
i
(s) from previous lemmas.
0.12 Uniform estimates on density
In this section we assume
Hypothesis H
1
. For all (t, x),
m := 0.1 < (t, x).
The nal a priori estimate is contained in the
Lemma 16. Assuming the conditions of lemmas 15, 25, 27 there exist c
i
=
c
i
(, , , ), i = 1, 2, and c
0
= c
0
(), ]0, 1/4[, such that
osc (t, ) +(t, )

c
2
( osc
0
+c
0

) + (E(0) +|u
0
|
2
)
1
2
, (50)
provided that
E(0) +|u
0
|
2
+ osc
0
+
0

< c
1
and
c
0

+ 3
< 1,
Proof. In the estimates below we allow a generic constant c depend on both m
and M (lower and upper bound on ). Take x
1
R
3
+
, x
2
R
3
+
and consider
two ow trajectories X(s, x
1
, t), X(s, x
2
, t) with the initial data
X(t, x
1
, t) = x
1
, X(t, x
2
, t) = x
2
.
We abbreviate them as X
s
1
and X
s
2
. Let
i
(), F
i
, i = 1, 2, be the restriction
of (t, x) and F(t, x) to the rst and second trajectories. Set
log = log
1
log
2
, =
1

2
.
Generally, by f we will refer to the dierence of the values of the function
f(t, x) on these trajectories. One readily notices that
= (t)log
23
where Int[
1
,
2
]. We set
(t) =
_
t
0

+ 2
dt.
It holds that
(s) (t)
m
+ 2
(s t), s < t.
Consider equation (1). It can be written in the following form,
D
t
log +

+ 2
log = div v +

2

3
P
1
,
where v and P
1
were introduced in (30) and (37) respectively. This equation
is integrated to obtain an inequality
|(t)| Me
(t)
|(0)|
+
M
+ 2

_
t
0
e
(s)(t)
_
div v +

2

3
P
1
_
ds

. (51)
First, we obtain an estimate on osc (t). By the estimates of lemma 12 and
the estimate (43) we obtain that
osc (t, ) c osc
0
() +c

(E(0) +|u
0
|
2
)
1
2
+c(E(0) +|u
0
|
2
)
+
_
t
0
e

m
+2
(ts)
(s, )

ds. (52)
Now, we derive the estimate on

. For that in the equation (51) we restrict

x
1
R
3
+
, |x
1
x
2
| < 1 and divide the equation by |x
1
x
2
|

. We obtain
M
1
||
|x
1
x
2
|

e
(t)
|
0
|
|X
0
1
X
0
2
|

|X
0
1
X
0
2
|

0
|x
1
x
2
|

+
1
+ 2

_
t
0
e
(t)+(s)
div v(s)
|x
1
x
2
|

ds
+
1
+ 2
_
t
0
e
(t)+(s)

3
P
1
(s)
|X
s
1
X
s
2
|

|X
s
1
X
s
2
|

|x
1
x
2
|

ds

. (53)
Some simplications are in order. The ratio
|X
s
1
X
s
2
|

|x
1
x
2
|

was estimated in lemma

15 by
(1 + c(, E(0), |u
0
|
2
))e
a(t)(ts) ds
,
where a(s) are given by 49. We make the following
24
Hypothesis H
2
. There is
0
> 0 such that for any t > 0 the inequality holds:
sup
s[0,t]
a(s)
m
2( + 2)
.
Thus we have estimated
e
(t)+(s)
|X
s
1
X
s
2
|

|x
1
x
2
|

(1 + c(
0
, E(0), |u
0
|
2
))e

m
2(+2)
(ts)
, s < t.
It follows then, that the rst term on the right-hand side in (53) is bounded
by

+ osc
0
.
The second, using lemma 12, by c(E(0), |u
0
|
2
) that vanish with its argu-
ments. The third, by lemma (13) and formulas for
i
from (36)),
(1 + c(
0
, E(0), |u
0
|
2
))
( + 2)

3
_
t
0
e

m
2(+2)
(ts)
( osc (s, ) +(s, )

) ds

c
0
(1 + c(
0
, E(0), |u
0
|
2
))
m

+ 3
sup
s[0,t]
( osc (s, ) +(s, )

) . (54)
In the above estimate c
0
depends only on . An inequality then can be derived
from (53):
(t, )

c
0

+c osc
0
+c(E(0), |u
0
|
2
)
+
c
0
(1 + c(
0
, E(0), |u
0
|
2
)M
m

+ 3
sup
s[0,t]
( osc (s, ) +(s, )

) . (55)
We consider the system of inequalities (52) and (55). It is straightforward to
verify that if
c
0
M
m

+ 3
< 1,
then we can choose E(0) + |u
0
|
2
+ osc
0
+
0

so small, depending on
, , , ,
0
that Hypotheses H
0
H
2
hold and (50) is veried.
0.13 Proof of the existence
Consider now the sequence of data of the problem

n
0
, u
n
0
C

_
R
3
+
_
C

0
_
R
3
+
_
3
,
25
which approximates the given initial data in the space L
6
loc
(R
3
+
) L
6
_
R
3
+
_
3
.
Moreover, we require that M >
n
0
> m > 0, and smallness condition (7) as
required by analysis of the previous sections. Such a sequence, clearly exists.
We can take
n
0
(x) = (
0
(x) +n
1
)
n
1(x), u
n
0
= (u
0
(x))
n
1(x), where

is the standard mollier. Accordingly, let

n
, u
n
be the sequence of smooth
solutions of the problem with
n
0
, u
n
0
as the initial data. The existence of such
solutions follows from the local existence result [11] and a priori estimates
we obtained in the previous sections. In particular, we established that the
following norms are bounded with bounds independent of n.
{
n
} bounded in L

_
R
+
R
3
+
_
, (56)
{

n
u
n
} bounded in L

_
R
+
: L
2
(R
3
+
)
_
, (57)
{u
n
} bounded in L
2
_
R
+
R
3
+
_
, (58)
{D
2
u
n
} bounded in L
2
_
R
+
R
3
+
_
. (59)
By the weak stability result of P.-L. Lions, see Theorem 5.1 of [8], bounds
(56)(58) imply the existence of an accumulation point (, u) of the sequence
(
n
, u
n
) in the weak topology of L
6
loc
(R
3
+
) L
6
(R
3
+
) which is a weak solution
of (1)(4). Moreover, the bounds in the spaces from (56)(59) hold for this
(, u).
References
[1] B. Desjardins, Regularity of weak solutions of the compressible isen-
tropic Navier-Stokes equations, Commun. in PDE, 22(5&6),p.977-
1008(1997).
[2] D. Gilbarg and N.S. Trudinger, Elliptic Partial Dierential Equations
of Second Order, Springer(1998).
[3] E. Feireisl, Dynamics of Viscous Compressible Fluids, Oxford Lecture
Series in Mathematics and Its Applications, 26(2004).
[4] D. Ho, Discontinuous solutions of the Navier-Stokes equations for com-
pressible ow, Arch. Rat. Mech. Anal., 114(1991), p.15-46.
[5] D.Ho, Compressible ow in half-space with Navier boundary condi-
tions, J. Math. Fluid Mech. 7(2005) n.3, p.315-338.
26
[6] L.D. Landau and E.M. Lifshitz, Fluid Mechanics, A Course of Theoret-
ical Physics: Volume 6, 2
nd
edition, Elsevier, (2004).
[7] P.-L. Lions, Mathematical topics in uid dynamics, Vol. 1, Incompress-
ible models, Oxford Science Publication, Oxford(1998).
[8] P.-L. Lions, Mathematical topics in uid dynamics, Vol. 2, Compressible
models, Oxford Science Publication, Oxford(1998).
[9] A. Majda, A. Bertozzi, Vorticity and Incompressible Flow, Cambridge
texts in applied mathematics, Cambridge(2002).
[10] S. Mikhlin, N. Morozov, M. Paukshto, The integral equations of
the theory of elasticity, B.G. Teubner Verlagsgesellschaft, Stuttgart,
Leipzig(1995).
[11] A. Matsumura, T. Nishida. The initial value problem for the equations
of motion of compressible and heat conductive uids, Commun. Math.
Phys. 89(1983), p. 445-464.
[12] J. Nevcas, Les methodes directes en theorie des equations ellip-
tiques, Academia,

Editions de LAcademie Tchecoslovaque des Sciences,
Prague(1967).
[13] A. Novotn y and I. Straskraba, Introduction to the Mathematical Theory
of Compressible Flow Oxford University Press, Oxford(2004).
[14] V.D. Sheremet, Handbook of Greens Functions & Matrices, Wit Press,
Computational Mechanics(2003).
[15] V.A. Solonikov, The solvability of the initial-boundary value problem
for the equations of motion of a viscouse compressible uid, in Russian,
Zap. Nauch. Sem. LOMI 56(1976) p. 128-147.
[16] R. Temam, Navier-Stokes equations, Noth-Holland, Amsterdam(1997).
27