Uniqueness of 2-D compressible vortex sheets

Jean-Francois Coulombel

, and Paolo Secchi

April 21, 2008

CNRS, Universite Lille 1 and Team Project SIMPAF of INRIA Lille Nord Europe,
Laboratoire Paul Painleve, Batiment M2, Cite Scientique
59655 VILLENEUVE DASCQ CEDEX, France

Dipartimento di Matematica, Facolt`a di Ingegneria,

Via Valotti, 9, 25133 BRESCIA, Italy
Abstract
We consider compressible vortex sheets for the isentropic Euler equations of gas dynamics
in two space dimensions. Under a supersonic condition that precludes violent instabilities, in
previous papers [3, 4] we have studied the linearized stability and proved the local existence
of piecewise smooth solutions to the nonlinear problem. This is a free boundary nonlinear
hyperbolic problem with two main diculties: the free boundary is characteristic, and the
so-called Lopatinskii condition holds only in a weak sense, which yields losses of derivatives.
In the present paper we prove that suciently smooth solutions are unique.
AMS subject classication: 76N10, 35Q35, 35L50, 76E17.
Keywords: compressible Euler equations, vortex sheets, contact discontinuities, weak Lopatin-
skii condition, loss of derivatives.
1 Introduction and main results
We consider Euler equations of isentropic gas dynamics in the whole plane R
2
. Denoting by u
the velocity of the uid and the density, the equations read:
_

t
+ ( u) = 0 ,

t
( u) + ( u u) +p = 0 ,
(1)
where p = p() is the pressure law. In all this paper, p is assumed to be a strictly increasing
function of , dened on ]0, +[. We also assume that p is a C

function of . The speed of

sound c() in the uid is then dened by the relation:
c() :=
_
p

() .
In this paper, we are interested in solutions to (1) that are smooth on either side of a curve
(t) := {y
2
= (t, y
1
) , t [0, T
0
] , y
1
R}, and such that, at each time t [0, T
0
], the tangential
velocity (with respect to (t)) is the only quantity that experiments a jump across this curve.
The density, and the normal velocity should be continuous across (t). For such solutions, the
jump conditions across (t) read (see [3, 4]):

t
= u
+
= u

+
=

,
(2)
1
where denote the states on the two sides of the curve and := (
y
1
, 1) is a (space)
normal vector to (t). As detailed in [3], for the isentropic Euler equations (1), these solutions
are exactly the contact discontinuities in the sense of Lax [6]. Due to the jump of tangential
velocity, the vorticity is concentrated along (t). For this reason tangential discontinuities are
also called vortex sheets.
The interface (t), or equivalently the function , is part of the unknowns of the problem.
We thus deal with a free boundary problem.
Problem (1), (2) has simple solutions like the stationary rectilinear vortex sheets:
(, u) =
_
U
+
= (, v, 0) , if y
2
> 0,
U

= (, v, 0) , if y
2
< 0,
(3)
where , v R, > 0. Up to Galilean transformations, every rectilinear vortex sheet has this
form and, without loss of generality, one may also assume v > 0.
The uniform Kreiss-Lopatinski stability condition is never satised by contact discontinuities
in two or three space dimensions, see e.g. [7, 5] or [8, page 222]. In three space dimensions,
every contact discontinuity is even violently unstable (this violent instability is the analogue
of the Kelvin-Helmholtz instability for incompressible uids), while in two space dimensions a
large jump of the tangential velocity makes the contact discontinuity weakly stable. A precise
study of this weak stability has been performed in [3], where it is shown that for such weakly
stable vortex sheets, the linearized equations satisfy an a priori L
2
energy estimate with a loss
of one derivative. In [4] the existence of contact discontinuity solutions to the Euler equations is
proved by extending the linear well-posedness result to Sobolev spaces and then by introducing
a suitable Nash-Moser iteration scheme.
In the present paper we prove the uniqueness of smooth enough vortex sheets.
Given two solutions V
1
= (
1
, u
1
),
1
, and V
2
= (
2
, u
2
),
2
, let us denote the two sides
determined by each unknown front
i
by

i
= {(t, y) [0, T
0
] R
2
: (y
2

i
(t, y
1
)) > 0}.
Each solution has states V

i
on

i
, i = 1, 2.
The main result of the paper is the following Theorem:
Theorem 1. Let T
0
> 0. Assume that the stationary solution dened by (3) satises the
supersoniccondition:
v >

2 c() . (4)
Let us consider two solutions V
1
,
1
and V
2
,
2
of (1), (2) that have the form V

i
= U

+

V

i
with

V

i
H
7
(

i
), and
i
H
8
(]0, T
0
[R). There exists a constant K > 0 such that, if:

V

2

W
2,
(

2
)
+
W
3,
(]0,T
0
[R)
K , (5)
and if V
1
= V
2
and
1
=
2
at the initial time t = 0, then V
1
= V
2
,
1
=
2
for all t [0, T
0
].
Let us observe that our method to prove uniqueness also applies to weakly stable nonchar-
acteristic waves such as shock waves or phase transitions (whose existence follows from the
analysis in [4]). We shall focus here on contact discontinuities since the major technical prob-
lems arise in this situation. As usual, the analysis of noncharacteristic discontinuities presents
some simplications so we shall not deal with this case.
In order to solve problem (1), (2), it is convenient to straighten the unknown front in order
to work in a xed domain, see [3, 4]. To do so, the unknowns (, u), that are smooth functions
on either side of {y
2
= (t, y
1
)}, are replaced by the functions:
(

, u

)(t, x
1
, x
2
) := (, u)(t, x
1
, (t, x
1
, x
2
)) , (6)
2
where is any continuous function, smooth away from {x
2
= 0}, and satisfying the following
conditions:

x
2
(t, x
1
, x
2
) > 0 , (t, x
1
, 0) = (t, x
1
) .
With these requirements for , all functions

, u

are smooth on the xed domain {x

2
> 0}.
We also dene the functions:

(t, x
1
, x
2
) := (t, x
1
, x
2
) ,
that are both smooth on the half-space {x
2
> 0}.
Let us denote by v

and u

the two components of the velocity eld, that is, u

= (v

, u

).
Let us also set U

= (

, v

, u

). After the change of variables (6) the nonlinear equations (1)

read:

t
U
+
+ A
1
(U
+
)
x
1
U
+
+
1

x
2

+
_
A
2
(U
+
)
t

x
1

+
A
1
(U
+
)
_

x
2
U
+
= 0 ,

t
U

+ A
1
(U

)
x
1
U

+
1

x
2

_
A
2
(U

)
t

x
1

A
1
(U

)
_

x
2
U

= 0 ,
(7)
in the domain Q
T
0
:= {t [0, T
0
] , x
1
R, x
2
> 0}, where we have set:
A
1
(U) :=
_
_
_
_
v

0
p

0
0 0 v

_
_
_
_
, A
2
(U) :=
_
_
_
_
u

0 u

0
p

0 u

_
_
_
_
.
The boundary conditions (2) now read:

+
|
x
2
=0
=

|
x
2
=0
= ,

t
= v
+
|
x
2
=0

x
1
+ u
+
|
x
2
=0
= v

|
x
2
=0

x
1
+ u

|
x
2
=0
,

+
|
x
2
=0
=

|
x
2
=0
,
(8)
at
T
0
:= {t [0, T
0
] , x
1
R}. With an obvious denition for the dierential operator L,
system (7) also reads:
L(U
+
,
+
)U
+
= 0 , L(U

)U

= 0 . (9)
When no confusion is possible, we also write this system under the form L(U, )U = 0, where U
stands for the vector (U
+
, U

) and for (
+
,

). In the same way, the boundary conditions

(8) can be rewritten in the compact form:

+
|
x
2
=0
=

|
x
2
=0
= ,
B(U
+
|
x
2
=0
, U

|
x
2
=0
, ) = 0 . (10)
In [4] we have shown how to construct a solution U, to (9), (10), where has the form

= x
2
+

and satises:

x
2

+
(t, x
1
, x
2
) ,
x
2

(t, x
1
, x
2
) , (11)
for a suitable constant > 0, and the eikonal equations:

+
+ v
+

x
1

+
u
+

=
t

+ v

x
1

= 0 , (12)
in the whole domain Q
T
0
.
3
The eikonal equations (12), that are clearly imposed on the boundary {x
2
= 0} by (8) ensure
that the matrices A
2
(U

)
t

x
1

A
1
(U

) have a constant rank in the whole domain

{x
2
0}, and not only on the boundary. This constant rank property is crucial in [3, 4] to
perform a Kreiss symmetrizers construction and to derive a priori estimates.
Theorem 1 is a consequence of the following auxiliary result:
Theorem 2. Let T
0
> 0. Assume that the stationary solution dened by (3) satises (4).
Let us consider two solutions of (9), (10), (11), (12), U
1
,
1
,
1
, and U
2
,
2
,
2
of the form
U

i
= U

+

U

i
with

U

i
H
5
(Q
T
0
),
i
= x
2
+

i
with

i
H
5
(Q
T
0
), and
i
H
4
(
T
0
).
There exists a constant K > 0 such that, if:

2

W
2,
(Q
T
0
)
+

W
3,
(Q
T
0
)
K ,
and if U
1
= U
2
,
1
=
2
, and
1
=
2
at the initial time t = 0, then U
1
= U
2
,
1
=
2
,
1
=
2
for all t [0, T
0
].
Theorem 2 is a uniqueness result in the straightened variables (t, x
1
, x
2
), for a particular
choice of the lifting function (this lifting function is supposed to satisfy the eikonal equa-
tions (12)). Let us note that the physical problem is in the original variables (t, y
1
, y
2
) so the
physically relevant result is Theorem 1. However energy estimates and therefore uniqueness of
solutions are more easily derived in a xed domain so we shall use Theorem 2 as an intermediate
step in the proof of Theorem 1.
The rest of this paper is devoted to the proof of Theorem 2 then to the proof of Theorem
1. The proof of Theorem 2 follows the arguments of [2] with several modications that arise
from our functional framework (that is not the same as in [2]). However, the proof of Theorem
1 diers from the corresponding analysis in [2] because our choice of lifting function is not as
explicit as in [2]. We shall therefore need to develop new arguments that do not appear in [2]
to handle the construction of an appropriate lifting function. This is due to the characteristic
nondissipative nature of the boundary conditions in our problem.
2 Proof of Theorem 2
For convenience, in this section we drop the index and only keep the + and exponents when
a confusion is possible. Given two solutions U
1
,
1
,
1
and U
2
,
2
,
2
with the regularity stated
in Theorem 2, we denote:
U := U
1
U
2
, :=
1

2
, :=
1

2
.
For the sake of shortness, we also introduce the notation:

A(U, ) :=
1

x
2

(A
2
(U)
t

x
1
A
1
(U)) .
The calculation of the dierence L(U
1
,
1
)U
1
L(U
2
,
2
)U
2
gives (see e.g. [2]):
L(U
2
,
2
)U + [dA
1
(U
2
)U]
x
1
U
2
+ [d

A(U
2
,
2
)(U, )]
x
2
U
2
+[A
1
(U
1
) A
1
(U
2
)]
x
1
U + [A
1
(U
1
) A
1
(U
2
) dA
1
(U
2
)U]
x
1
U
2
+[

A(U
1
,
1
)

A(U
2
,
2
)]
x
2
U + [

A(U
1
,
1
)

A(U
2
,
2
) d

A(U
2
,
2
)(U, )]
x
2
U
2
= 0 .
(13)
4
Such an equation may be simplied by introducing the good unknown as in [1]:

U := U

x
2
U
2

x
2

2
. (14)
The substitution of (14) in (13) yields (see again [1]) the system of equations
1
:
L(U
2
,
2
)

U + C(U
2
,
2
)

U = f , (15)
where we have set:
C(U
2
,
2
)

U := [dA
1
(U
2
)

U]
x
1
U
2
+
1

x
2

2
[dA
2
(U
2
)

U
x
1

2
dA
1
(U
2
)

U]
x
2
U
2
,
and where the right-hand side in (15) is the quadratic term f =

4
i=1
f
i
with:
f
1
:= [dA
1
(U
2
)U]
x
1
U , (16a)
f
2
:= [A
1
(U
1
) A
1
(U
2
) dA
1
(U
2
)U]
x
1
U
1
, (16b)
f
3
:= [d

A(U
2
,
2
)(U, )]
x
2
U , (16c)
f
4
:= [

A(U
1
,
1
)

A(U
2
,
2
) d

A(U
2
,
2
)(U, )]
x
2
U
1
. (16d)
Next we consider the boundary conditions. Let us introduce the matrices:
b(t, x
1
) :=
_
_
0 (v
+
2
v

2
)
|x
2
=0
1 v
+
2 |x
2
=0
0 0
_
_
,
M(t, x
1
) :=
_
_
0
x
1

2
1 0
x
1

2
1
0
x
1

2
1 0 0 0
1 0 0 1 0 0
_
_
.
We take the dierence of the boundary conditions (8) for the two solutions and we obtain:

+
|
x
2
=0
=

|
x
2
=0
= ,
b + M U
|
x
2
=0
= g ,
where we have set = (
t
,
x
1
) and:
g :=
_
_
(v
+
v

)
|x
2
=0

x
1

v
+
|x
2
=0

x
1

0
_
_
. (17)
In terms of the good unknown dened by (14), the boundary conditions read:

+
|
x
2
=0
=

|
x
2
=0
= ,
b + M
_

x
2
U
+
2
/
x
2

+
2

x
2
U

2
/
x
2

2
_
|
x
2
=0
. .
b

+ M

U
|
x
2
=0
= g .
(18)
We wish to apply the L
2
a priori estimate for (15), (18) that we have derived in [3]. Given
any 0 < T T
0
, we assume that the solution (U
2
,
2
) satises:

2

W
2,
(Q
T
)
+

W
3,
(Q
T
)
K , (19)
where Q
T
:= {t [0, T] , x
1
R, x
2
> 0} and K is a positive constant. Then the following
result holds:
1
The computations with the good unknown in [1] require C
2
smoothness of the solutions, which is obtained
here from the Sobolev imbedding Theorem.
5
Lemma 1. Assume that the particular solution (3) satises the supersoniccondition (4),
and that the solution U

2
,

2
satises (19). There exist positive constants K
0
> 0, C
0
> 0
independent of T, such that if K K
0
, then the following estimate holds:

U
L
2
(Q
T
)
+
H
1
(
T
)
C
0
(f
H
1
(Q
T
)
+g
H
1
(
T
)
) . (20)
Proof. From U
1
= U
2
,
1
=
2
, and
1
=
2
at the initial time t = 0, we infer f = g = 0 at
t = 0. We then extend both f and g by 0 for t < 0 and we obtain f H
1
(] , T] R
2
+
),
g H
1
(] , T] R) from the classical Theorems on products of Sobolev functions. We can
also extend U, , , and therefore

U by zero for t < 0. The initial conditions for U
1,2
,
1,2
show
that U, ,

U belong to H
1
(] , T] R
2
+
) and belongs to H
1
(] , T] R). Moreover,
extending the coecients (U
2
,
2
) for negative times in an arbitrary way, the equations (15),
(18) are satised for all t ] , T
0
], and not only for t [0, T
0
].
If K is suciently small, then (U
2
,
2
) satisfy all the assumptions of Theorem 3 in [4]. We
apply Theorem 3 in [4] and infer that the unique solution (

U, ) of (15), (18) satises (20).
We notice that from U, H
5
(Q
T
0
) and equation (7) we may deduce
j
t
U =
j
t
= 0
at time t = 0, for j = 0, . . . , 4. Therefore we may extend U, by 0 for t < 0 and we obtain
U, ,

U H
5
(], T]R
2
+
). This property ensures that all constants in the Sobolev imbeddings
and interpolation inequalities used in the following calculations will not depend on T.
Now, we estimate the right-hand side f of (15):
Lemma 2. Let f be dened by (16). Under the assumptions of Theorem 2 there exists a positive
constant C independent of T such that:
f
H
1
(Q
T
)
C (U
H
9/4
(Q
T
)
+
H
9/4
(Q
T
)
)2 . (21)
Proof. We estimate the four terms given in (16). The most critical cases concern f
3
and f
4
and
we show them in detail. The estimates of f
1
and f
2
are proved in a similar way. Denoting a
generic space or time derivative by , from (16c) one has:
f
3

H
1
(Q
T
)
C (U U
L
2
(Q
T
)
+U U
L
2
(Q
T
)
+U
2
U
L
2
(Q
T
)
+U
L
2
(Q
T
)
+
2
U
L
2
(Q
T
)
+
2
U
L
2
(Q
T
)
) .
By the Holder inequality:
w
1
w
2

L
2
(Q
T
)
w
1

L
12/5
(Q
T
)
w
2

L
12
(Q
T
)
,
and the Sobolev imbeddings H
1/4
(Q
T
) L
12/5
(Q
T
), H
5/4
(Q
T
) L
12
(Q
T
), we infer:
f
3

H
1
(Q
T
)
C U
H
9/4
(Q
T
)
(U
H
5/4
(Q
T
)
+
H
9/4
(Q
T
)
) . (22)
Using a similar Holder inequality and Sobolev imbeddings, from (16d) one has:
f
4

H
1
(Q
T
)
C (U U
L
2
(Q
T
)
+U U
L
2
(Q
T
)
+U
L
2
(Q
T
)
+U
2

L
2
(Q
T
)
+
2

L
2
(Q
T
)
)
C (U
H
5/4
(Q
T
)
+
H
9/4
(Q
T
)
)2 .
(23)
From (16a), (16b) we also have (the details are omitted):
f
1

H
1
(Q
T
)
+f
2

H
1
(Q
T
)
C (U
H
5/4
(Q
T
)
+U
H
9/4
(Q
T
)
) U
H
5/4
(Q
T
)
. (24)
From (22), (23), (24), and the obvious estimate U
H
5/4
(Q
T
)
U
H
9/4
(Q
T
)
, we get (21).
6
Now we estimate the right-hand side g of (18):
Lemma 3. Let g be dened by (17). Under the assumptions of Theorem 2 there exists a positive
constant C independent of T such that:
g
H
1
(
T
)
C U
H
2
(Q
T
)

H
2
(
T
)
. (25)
Proof. From the denition of g given in (17) we have:
g
H
1
(
T
)
C (U
|
x
2
=0

L
2
(
T
)
+U
|
x
2
=0

2

L
2
(
T
)
+U
|
x
2
=0

L
2
(
T
)
) .
The thesis easily follows from the imbeddings H
2
(Q
T
) L

(Q
T
) and H
1/2
(
T
) L
4
(
T
).
The next step is to estimate in terms of U or

U. We rst compute the equation satised
by . The dierence of the two eikonal equations (12) for
1
and
2
gives (for each state +
and ):

t
+ v
2

x
1
= u v
x
1

1
, (26)
where u = u
1
u
2
and v = v
1
v
2
. By the introduction of the the good unknown (14), we
also have:

t
+ v
2

x
1
+
1

x
2

2
(
x
2
v
2

x
1

x
2
u
2
) = u v
x
1

1
. (27)
The result is the following:
Lemma 4. Under the assumptions of Theorem 2 there exists a positive constant C independent
of T such that:

H
9/4
(Q
T
)
C U
H
9/4
(Q
T
)
, (28a)

L
2
(Q
T
)
C

U
L
2
(Q
T
)
. (28b)
Proof. By taking derivatives of order two and three of (26) and integrating by parts over Q
T
we easily show:

H
2
(Q
T
)
C U
H
2
(Q
T
)
, and
H
3
(Q
T
)
C U
H
3
(Q
T
)
.
Therefore the linear map U is bounded in H
2
(Q
T
) and H
3
(Q
T
). By interpolation we infer
(28a). Next we consider (27). Multiplying by and integrating by parts over Q
T
immediately
yields (28b). Recall that vanishes for t < 0.
From the denition of the good unknown given in (14) and from (28b) we get:
U
L
2
(Q
T
)
C

U
L
2
(Q
T
)
.
Then, from (20), (21), (25) and (28) we may deduce:
U
L
2
(Q
T
)
+
H
1
(
T
)
C (U
2
H
9/4
(Q
T
)
+
2
H
2
(
T
)
) .
Using the interpolation inequalities:
U
H
9/4
(Q
T
)
C U
1/2
H
9/2
(Q
T
)
U
1/2
L
2
(Q
T
)
,

H
2
(
T
)
C
1/2
H
3
(
T
)

1/2
H
1
(
T
)
,
we eventually get the estimate:
U
L
2
(Q
T
)
+
H
1
(
T
)
C (U
H
9/2
(Q
T
)
+
H
3
(
T
)
) (U
L
2
(Q
T
)
+
H
1
(
T
)
) . (29)
7
We remark that the constant C in (29) does not depend on T.
By the elementary inequality w
L
2
(0,T)
T
t
w
L
2
(0,T)
if w
t=0
= 0, we deduce:
U
H
4
(Q
T
)
C T U
H
5
(Q
T
)
.
Another interpolation argument gives U
H
9/2
(Q
T
)
C T
1/2
U
H
5
(Q
T
)
, where C is still in-
dependent of T. Similarly we have
H
3
(
T
)
C T
H
4
(
T
)
, with C independent of T.
Therefore from (29) we infer:
U
L
2
(Q
T
)
+
H
1
(
T
)
C T
1/2
(U
H
5
(Q
T
)
+T
1/2
0

H
4
(
T
)
) (U
L
2
(Q
T
)
+
H
1
(
T
)
) , (30)
with C independent of T. Since U belongs to H
5
(Q
T
0
) and belongs to H
4
(
T
0
), (30) yields
U = 0, = 0 for t [0, T], provided that T is taken suciently small. From (28a), we also have
= 0 for t [0, T].
By repeating the same argument on the intervals [T, 2T], [2T, 3T], . . . , we prove that U =
0, = 0, = 0 on the whole time interval [0, T
0
]. This completes the proof of Theorem 2.
Let us note that the regularity required for the uniqueness Theorem 2 is less than the minimal
regularity of solutions constructed in [4].
3 Proof of Theorem 1
We consider a solution V = (, v, u), of the equations (1), (2) in the original variables (t, y
1
, y
2
).
Our goal is to construct a lifting function and to apply Theorem 2 in the straightened variables.
Let us rst develop some formal arguments to understand how should be constructed. The
function should satisfy (12), that is:

t
(t, x
1
, x
2
) + v(t, x
1
, (t, x
1
, x
2
))
x
1
(t, x
1
, x
2
) u(t, x
1
, (t, x
1
, x
2
)) = 0 ,
where the coecients v, u are known. This is a quasilinear equation with coecients that depend
on (t, x
1
). The construction and the regularity analysis of solutions for such equations are not
very practical. Instead of constructing , we are going to show that it is by far easier to construct
its inverse. More precisely, let us assume that is known and satises (11). Let (t, y
1
, y
2
)
be dened by the relations:
(t, y
1
, (t, y
1
, y
2
)) = y
2
, (t, x
1
, (t, x
1
, x
2
)) = x
2
.
We compute the equation satised by starting from (12). We use the relations:

t
(t, x
1
, (t, x
1
, x
2
)) +
y
2
(t, x
1
, (t, x
1
, x
2
))
t
(t, x
1
, x
2
) = 0 ,

y
1
(t, x
1
, (t, x
1
, x
2
)) +
y
2
(t, x
1
, (t, x
1
, x
2
))
x
1
(t, x
1
, x
2
) = 0 ,

y
2
(t, x
1
, (t, x
1
, x
2
))
x
2
(t, x
1
, x
2
) = 1 ,
and obtain:

t
(t, y
1
, y
2
) + v(t, y
1
, y
2
)
y
1
(t, y
1
, y
2
) + u(t, y
1
, y
2
)
y
2
(t, y
1
, y
2
) = 0 . (31)
The main point is that (31) is a linear transport equation for that should hold on the domain

+
, together with the continuity condition:
(t, y
1
, (t, y
1
)) = 0 . (32)
8
We shall see that the equations (31), (32) can be solved by applying the method of characteristics.
Once is constructed, we can obtain by inverting provided that
y
2
is uniformly positive.
Below we are therefore going to show how to construct with a suciently large Sobolev index
(in order to apply Theorem 2).
We thus consider a solution V, of the equations (1), (2) in the original variables, and assume
that this solution has the form V

= U

+

V

, with

V

H
7
(

), and H
8
(]0, T
0
[R).
Moreover, as in (5)

V

may be suciently small in W
2,
(

) and may be suciently small

in W
3,
(]0, T
0
[R). Recall that

denote the sets {(t, y) [0, T

0
] R
2
: (y
2
(t, y
1
)) > 0}.
We let
0
H
15/2
(R) denote the initial condition for . Then we consider a function
C

0
(R) such that equals 1 on the interval [
0

L
,
0

L
]. The initial condition for
is dened in the following way:

0
(y
1
, y
2
) = y
2

0
(y
1
) (y
2
) . (33)
It is clear that
0
(y
1
,
0
(y
1
)) = 0 for all y
1
. The Sobolev imbedding Theorem shows that
0
belongs to C
5
(R
2
) and the derivatives of
0
up to order 5 are bounded. We can also choose
such that the following inequalities hold:

y
2

0
(y
1
, y
2
)
1
2
, |
y
1

0
(y
1
, y
2
)|

L
.
Moreover, if
0
is suciently small in W
3,
(R), then
0
y
2
is small in W
3,
(R
2
). The choice
of , and therefore the construction of
0
, only depends on
0
.
The proof of Theorem 1 relies on two lemmas that we give now:
Lemma 5. Let p, k denote some positive integers. Let O
1
, O
2
denote two open sets of R
p
and
let f : O
1
O
2
be a C
k
-dieomorphism such that the jacobian J
f
of f satises:
inf
xO
1
|J
f
(x)| > 0 ,
and such that all derivatives d
1
f, . . . , d
k
f belong to L

(O
1
). Then the inverse f
1
of f satises:
inf
xO
2
|J
f
1(x)| > 0 ,
and the derivatives d
1
f
1
, . . . , d
k
f
1
belong to L

(O
2
).
Lemma 6. Under the same assumptions and notations as in Lemma 5, one has w H
k
(O
2
) if
and only if w f H
k
(O
1
).
The proofs of Lemmata 5 and 6 follow from classical calculus arguments and are therefore
omitted.
Our rst goal is to construct a solution of (31), (32) with the initial condition
0
. Let
us introduce the integral curves of the velocity eld u. From the Sobolev imbedding Theorem,
we know that u belongs to C
5
(
+
) and has bounded derivatives up to order 5. Let y
2
>
0
(y
1
)
and let X(t, y
1
, y
2
) denote the solution to the ordinary dierential equation:
_
_
_
dX
dt
(t, y
1
, y
2
) = u(t, X(t, y
1
, y
2
)) , t [0, T
0
] ,
X(0, y
1
, y
2
) = (y
1
, y
2
) .
The vector eld u is tangent to the boundary of
+
, so the integral curve X(t, y
1
, y
2
) is well-
dened for all t [0, T
0
] and stays in
+
. Moreover, standard arguments of ordinary dierential
equations theory show that the ow:
O
+
:= {(t, y
1
, y
2
) [0, T
0
] R
2
: y
2
>
0
(y
1
)}
+
(t, y
1
, y
2
) (t, X(t, y
1
, y
2
)) , (34)
9
is a C
5
-dieomorphism. Dierentiating the dierential equation with respect to the initial con-
ditions, we obtain uniform bounds of the form:
|d
y
X(t, y
1
, y
2
) I| C t , |d
2
y
X(t, y
1
, y
2
)| + +|d
5
y
X(t, y
1
, y
2
)| C .
Such bounds imply that up to restricting the nal time T
0
, the C
5
-dieomorphism (34) satises
the assumptions of Lemma 5. Its inverse has the form:

+
O
+
(t, y
1
, y
2
) (t, Y(t, y
1
, y
2
)) , (35)
and satises the conclusions of Lemma 5. The function is now dened on
+
as follows:
(t, y
1
, y
2
)
+
, (t, y
1
, y
2
) =
0
(Y(t, y
1
, y
2
)) . (36)
It is clear that belongs to C
5
and satises the transport equation (31) together with the
condition (32) on the boundary of
+
thanks to the properties of
0
. The derivatives of up
to order 5 are bounded and, up to restricting T
0
once again, we have the bound
y
2
1/4.
The end of the proof of Theorem 1 relies on the following result that establishes integrability
properties for :
Lemma 7. Let be dened by (36). Then the mapping
_
(t, y
1
, y
2
) (t, y
1
, y
2
) y
2
_
belongs
to H
5
(
+
).
Proof. We rst note that on
+
, the velocity u reads u = (v, 0)+ u with u H
7
(
+
) H
5
(
+
).
Applying lemma 6, we have:
_
(t, y
1
, y
2
) u(t, X(t, y
1
, y
2
))
_
H
5
(O
+
) .
Recall that the set O
+
is dened in (34). Then we use the ordinary dierential equation satised
by X and deduce that:
_
(t, y
1
, y
2
) X(t, y
1
, y
2
) (y
1
+ v t, y
2
)
_
H
5
(O
+
) .
Using the notation X = (X
1
, X
2
) and Y = (Y
1
, Y
2
) for the coordinates of X and Y, we can
write X
2
(t, y
1
, y
2
) = y
2
+

X
2
(t, y
1
, y
2
) with

X
2
H
5
(O
+
). Then we use the relation:
(t, y
1
, y
2
)
+
, y
2
= X
2
(t, Y(t, y
1
, y
2
)) = Y
2
(t, y
1
, y
2
) +

X
2
(t, Y(t, y
1
, y
2
)) ,
and apply Lemma 6 again to obtain:
_
(t, y
1
, y
2
) Y
2
(t, y
1
, y
2
) y
2
_
H
5
(
+
) .
Eventually, we have:
(t, y
1
, y
2
) =
0
(Y(t, y
1
, y
2
)) = Y
2
(t, y
1
, y
2
)
0
(Y
1
(t, y
1
, y
2
)) (Y
2
(t, y
1
, y
2
)) ,
and the claim of Lemma 7 follows.
Using all the properties shown above, we claim that the mapping:

+
Q
T
0
= [0, T
0
] R
2
+
(t, y
1
, y
2
) (t, y
1
, (t, y
1
, y
2
)) ,
10
is a C
5
-dieomorphism that satises the assumptions of Lemma 5. Its inverse has the form:
Q
T
0

+
(t, x
1
, x
2
) (t, x
1
, (t, x
1
, x
2
)) ,
and satises the conclusions of Lemma 5. We dene:
U
+
: (t, x
1
, x
2
) V
+
(t, x
1
, (t, x
1
, x
2
)) ,
so applying Lemma 6 we get U U
+
H
5
(Q
T
0
). Applying the same kind of argument as in
the proof of Lemma 7, we can also prove that x
2
H
5
(Q
T
0
). The initial condition for
only depends on the initial condition for . The calculations developed at the beginning of this
section prove that satises the eikonal equation (12).
Up to restricting T
0
again if necessary, the dieomorphism (34), and consequently also its
inverse (35), may be taken close enough to the identity map. If
0
y
2
is suciently small in
W
3,
(R
2
), then y
2
is small in W
3,
(
+
) and x
2
is small in W
3,
(Q
T
0
). Moreover, if

V

= V
+
U
+
is taken suciently small in W
2,
(

), then U U
+
is small in W
2,
(Q
T
0
).
The construction of on

, and consequently the construction of U

proceed in a
similar way. We have thus constructed a solution of the system (7), (8), (11), (12) with the
regularity required in Theorem 2 (on a possibly shorter time interval). Theorem 1 follows
by applying the construction above to two solutions V
1
,
1
and V
2
,
2
with the same initial
conditions. The corresponding solutions in the straightened variables will share the same initial
conditions and will therefore coincide on a small time interval [0, T]. This shows that the maximal
time up to which the solutions coincide is necessarily equal to T
0
. The proof of Theorem 1 is
now complete.
Acknowledgments The rst author would like to warmly thank the Mathematics Depart-
ment of Brescia University for its kind hospitality during the visiting period when this work was
completed.
References
[1] S. Alinhac. Existence dondes de rarefaction pour des syst`emes quasi-lineaires hyperboliques
multidimensionnels. Comm. Partial Dierential Equations, 14(2):173230, 1989.
[2] S. Alinhac. Unicite dondes de rarefaction pour des syst`emes quasi-lineaires hyperboliques
multidimensionnels. Indiana Univ. Math. J., 38(2):345363, 1989.
[3] J. F. Coulombel, P. Secchi. The stability of compressible vortex sheets in two space dimen-
sions. Indiana Univ. Math. J., 53(4):9411012, 2004.
[4] J. F. Coulombel, P. Secchi. Nonlinear compressible vortex sheets in two space dimensions.
Ann. Sci.

Ecole Norm. Sup. (4), 41(1):85139, 2008.
[5] J. A. Fejer, J. W. Miles. On the stability of a plane vortex sheet with respect to three-
dimensional disturbances. J. Fluid Mech., 15:335336, 1963.
[6] P. D. Lax. Hyperbolic systems of conservation laws. II. Comm. Pure Appl. Math., 10:537
566, 1957.
[7] J. W. Miles. On the disturbed motion of a plane vortex sheet. J. Fluid Mech., 4:538552,
1958.
[8] D. Serre. Systems of conservation laws. 2. Cambridge University Press, 2000.
11