Jean-Francois Coulombel and Paolo Secchi - The Stability of Compressible Vortex Sheets in Two Space Dimensions

The Stability of Compressible Vortex Sheets
in Two Space Dimensions

JEAN-FRANC OIS COULOMBEL & PAOLO SECCHI
ABSTRACT. We study the linear stability of compressible vortex
sheets in two space dimensions. Under a supersonic condition
that precludes violent instabilities, we prove an energy estimate
for the linearized boundary value problem. Since the problem
is characteristic, the estimate we prove exhibits a loss of control
on the trace of the solution. Furthermore, the failure of the uni-
form Kreiss-Lopatinskii condition yields a loss of derivatives in
the energy estimate.
1. INTRODUCTION
A velocity discontinuity in an inviscid ow is called a vortex sheet. In three-space
dimensions, a vortex sheet has vorticity concentrated along a surface in the space.
In two-space dimensions, the vorticity is concentrated along a curve in the plane.
The present paper deals with compressible vortex sheets, i.e., vortex sheets in a
compressible ow.
If the solution is piecewise constant on either side of the interface of discon-
tinuity, one has planar vortex sheets in the three dimensional case and rectilinear
vortex sheets in the two dimensional case, respectively. The linear stability of pla-
nar and rectilinear compressible vortex sheets has been analyzed a long time ago,
see [12, 27]. In three space dimensions, planar vortex sheets are known to be vio-
lently unstable (see e.g. [30]). In the two dimensional case, subsonic vortex sheets
are also violently unstable, while supersonic vortex sheets are neutrally linearly sta-
ble, see e.g. [27, 30]. This result formally agrees with the theory of incompressible
vortex sheets. In fact, in the incompressible limit, the speed of sound tends to in-
nity, with the result that two-dimensional vortex sheets are always unstable. This
kind of instability is usually referred to as the Kelvin-Helmholtz instability. For
the incompressible theory of two-dimensional vortex sheets, we refer the reader to
2 JEAN-FRANC OIS COULOMBEL & PAOLO SECCHI
the books [7, 22]. Moreover, we refer to [14] for the study of the instability of
vortex sheets when heat conduction is taken into account.
However, the normal modes analysis performed to derive the linear stability
of supersonic vortex sheets is by far not sucient to guarantee the existence of
nonconstant vortex sheets (that is, contact discontinuities) solutions to the com-
pressible isentropic Euler equations. In this paper, we rst show that supersonic
constant vortex sheets are linearly stable, in the sense that the linearized system
(around these particular piecewise constant solutions) obeys an energy estimate.
Then we consider the linearized equations around a perturbation of a constant
vortex sheet, and we show that these linearized equations obey the same energy
estimate. This is a rst crucial step towards proving the existence of nonconstant
compressible vortex sheets.
Several points need to be highlighted. First of all, the existence of compress-
ible vortex sheets is a free boundary nonlinear hyperbolic problem. Moreover, the
free boundary is characteristic with respect to both left and right states since we
deal with contact discontinuities. This is one of the reasons why one can not apply
Majdas analysis on shock waves (see [20, 21]), that are noncharacteristic interfaces.
In some previous works devoted to weakly stable shock waves, see [10, 11], the rst
author has considered noncharacteristic hyperbolic Initial Boundary Value Prob-
lems that did not meet the uniform Kreiss-Lopatinskii condition. In the case of
vortex sheets, the analysis is closely related, with the additional diculty that the
boundary is characteristic (the present analysis thus relies more on the work of
Majda and Osher [23] rather than on the work of Kreiss [6, 17]). The connection
with [10, 11] is that in both cases, the analogue of the Kreiss-Lopatinskii condition
is fullled but not in a uniform way. Furthermore, in the case of vortex sheets as
in the case of shock waves, the linearized Rankine-Hugoniot conditions form an
elliptic system for the unknown front. This property is a key point in our work
since it allows to eliminate the unknown front and to consider a standard Bound-
ary Value Problem with a symbolic boundary condition (this ellipticity property
is also crucial in Majdas analysis on shock waves [20, 21]).
Regarding the energy estimates for the linearized problems, the failure of the
uniform Kreiss-Lopatinskii condition yields a loss of derivatives with respect to
the source terms. Furthermore, because the boundary is characteristic, we expect
to lose some control on the trace of the solution at the boundary. As a matter of
fact, we shall see that the only loss of control is on the tangential velocity (which
corresponds to the characteristic part of the solution). The good point is that
the ellipticity of the boundary conditions for the unknown front enables us to
gain one derivative for it, as in Majdas work on shock waves [20]. Going slightly
more into the details, we shall prove that the only frequencies for which we lose
some control on the solution correspond to bicharacteristic curves. Those curves
originate from those points at the boundary of the space domain where the so-
called Lopatinskii determinant vanishes. In the interior of the space domain, these
singularities propagate along two bicharacteristics associated with the incoming
modes.
Stability of Compressible Vortex Sheets 3
Let us now describe the content of the paper. In Section 2, we present the
nonlinear equations describing the evolution of compressible vortex sheets and
introduce some notations. Then, in Section 3, we shall consider the linearized
equations around a constant (stationary) vortex sheet. The main result for the
constant coecient linearized problem is given in Theorems 3.1 and 3.2. After
several reductions, we shall detail in Section 4 the normal modes analysis of the
linearized problem and construct a degenerate Kreiss symmetrizers in order to de-
rive our energy estimate. In Section 5, we rst present the variable coecients
linearized problem and introduce Alinhacs good unknown. Then we paralinearize
the equations, in order to use symbolic calculus and derive the energy estimate. A
precise estimate of the paralinearization errors is given. Eventually, we showhowto
control the dierent pieces of the solution, depending on their microlocalization.
The main result for the variable coecients linearized problem is given in Theo-
rem 5.1. In Section 6, we make some remarks about possible future achievements.
Appendix A is devoted to the proof of several technical lemmas and Appendix
B gathers the main results on paradierential calculus that are used throughout
Section 5.
2. THE NONLINEAR EQUATIONS
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:
(2.1)
_
t
+ (u) = 0,
t
(u) + (u u) +p = 0,
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
().
Let (, u)(t, x
1
, x
2
) be a smooth function on either side of a smooth hyper-
surface := x
2
= (t, x
1
). Then (, u) is a (weak) solution of (2.1) if and
only if (, u) is a classical solution of (2.1) on both sides of and the Rankine-
Hugoniot conditions hold at each point of :
t
[] [u ] = 0, (2.2a)
t
[u] [(u )u] [p] = 0, (2.2b)
where := (
x
1
, 1) is a (space) normal vector to . As usual, [q] = q
+
q
denotes the jump of a quantity q across the interface (see [29]).

Following Lax [18], we shall say that (, u) is a contact discontinuity if the
Rankine-Hugoniot conditions (2.2) are satised in the following way:
t
= u
+
= u
, p
+
= p
.
Because p is monotone, the previous equalities read
(2.3)
t
= u
+
= u
,
+
=
.
Since the density and the normal velocity are continuous across the interface ,
the only jump experimented by the solution is on the tangential velocity. (Here,
normal and tangential mean normal and tangential with respect to ). For this
reason, a contact discontinuity is a vortex sheet and we shall make no distinction
in the terminology we use.
Note that the rst two equalities above are nothing but eikonal equations: if
x
2
= (t, x
1
) is the equation of the interface , then satises
t
+
2
(
+
, u
+
,
x
1
) = 0 and
t
+
2
(
, u
,
x
1
) = 0,
on x
2
= 0, where
2
(, u, ) := u
_

1
_
, R,
is the second characteristic eld of the system (2.1). It is linearly degenerate since
the corresponding eigenvector, in the quasilinear form of (2.1), is given by
r
2
(, u, ) :=
_
_
_
0
1
_
_
_.
Recall that the space dimension equals 2.
The interface , or equivalently the function , is part of the unknowns of
the problem. We thus deal with a free boundary problem. As it is common in this
kind of situation, we rst straighten the unknown front in order to work in a xed
domain. More precisely, the unknowns (, u), that are smooth on either side of
x
2
= (t, x
1
), are replaced by the functions
(
+
, u
+
)(t, x
1
, x
2
) := (, u)(t, x
1
, (t, x
1
, x
2
)),
(
, u
)(t, x
1
, x
2
) := (, u)(t, x
1
, (t, x
1
, x
2
)),
where is a smooth function satisfying
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 do-

main x
2
> 0. For convenience, we drop the index and only keep the + and
exponents. We also dene the functions
:
(t, x
1
, x
2
) := (t, x
1
, :x
2
),
which are both smooth on the half-space x
2
> 0.
Let us denote by v and u the two components of the velocity, that is, u =
(v, u). Then the existence of compressible vortex sheets amounts to proving the
existence of smooth solutions to the following system:
(2.4a)
t
+
+v
+
x
1
+
+(u
+
+
v
+
x
1
+
)
x
2
x
2
+
+
+
x
1
v
+
+
+

x
2
u
+
x
2
+

+

x
1
x
2
+

x
2
v
+
= 0,
(2.4b)
t
v
+
+v
+
x
1
v
+
+(u
+
+
v
+
x
1
+
)
x
2
v
+
x
2
+
+
p
(
+
)
+

x
1
(
+
)
x
1
x
2
+

x
2
+
= 0,
(2.4c)
t
u
+
+v
+
x
1
u
+
+(u
+
+
v
+
x
1
+
)
x
2
u
+
x
2
+
+
p
(
+
)
x
2
x
2
+
= 0,
(2.4d)
t
+v
x
1
+(u
x
1
x
2
x
2
x
1
v

x
2
u
x
2

x
1
x
2

x
2
v
= 0,
(2.4e)
t
v
+v
x
1
v
+(u
x
1
x
2
v
x
2
+
p

x
1
x
1
x
2

x
2
= 0,
(2.4f)
t
u
+v
x
1
u
+(u
x
1
x
2
u
x
2
+
p
x
2
x
2
= 0,
in the xed domain x
2
> 0, together with the boundary conditions
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
.
For convenience, we rewrite the boundary conditions in the following way:
x
2
=0
=
x
2
=0
= , (2.5a)
(v
+
v
x
2
=0

x
1
(u
+
u
x
2
=0
= 0, (2.5b)
t
+v
+
x
2
=0

x
1
u
+
x
2
=0
= 0, (2.5c)
(
+
x
2
=0
= 0. (2.5d)
The functions
+
and
should also satisfy

(2.6)
x
2
+
(t, x
1
, x
2
) ,
x
2
(t, x
1
, x
2
) ,
for a suitable constant > 0.
In [20, 21], Majda makes the particular choice
:
(t, x
1
, x
2
) := :x
2
+(t, x
1
).
This choice is appropriate in the study of shock waves because these are nonchar-
acteristic discontinuities. In the study of contact discontinuities, it seems rather
natural to choose the change of variables
:
such that the eikonal equations
+
+
2
(
+
, u
+
,
x
1
+
) =
t
+
+v
+
x
1
+
u
+
= 0,
+
2
(
, u
,
x
1
) =
t
+v
x
1
= 0,
are satised in the whole closed half-space x
2
0. This choice, that is in-
spired from [13], has several advantages. First, it simplies much the expression
of the nonlinear equations (2.4). But it also implies that the so-called boundary
matrix has constant rank in the whole space domain x
2
0, and not only on
the boundary x
2
= 0. This will enable us to develop a Kreiss symmetrizers
technique, in the spirit of [23]. We shall go back to this feature later on.
The problem is thus the construction of (local in time) smooth solutions to
(2.4)(2.5)(2.6), once initial data have been prescribed. Of course, such initial
data will have to fulll a certain number of compatibility conditions. The rst step
in proving such an existence result is the study of the linearized problem around
a particular constant solution, and this is our rst main result, see Theorems 3.1
and 3.2. The second step is the study of the linearized problem around a (variable
coecients) perturbation of the constant solution. The extension to the variable
coecients linearized problem is addressed in the second part of the paper. Our
second main result states that the constant coecients energy estimate still holds
when one considers a variable coecients linearized problem, see Theorem 5.1.
To avoid overloading the paper, we introduce some compact notations for the
nonlinear equations (2.4). For all U := (, v, u)
T
, we dene
A
1
(U) :=
_
_
_
v 0
p
()/ v 0
0 0 v
_
_
_, A
2
(U) :=
_
_
_
u 0
0 u 0
p
()/ 0 u
_
_
_.
Then the nonlinear equations (2.4) read
(2.7a)
t
U
+
+A
1
(U
+
)
x
1
U
+
+
1
x
2
+
(A
2
(U
+
)
t
x
1
+
A
1
(U
+
))
x
2
U
+
= 0,
(2.7b)
t
U
+A
1
(U
)
x
1
U
+
1
x
2
(A
2
(U
)
t
x
1
A
1
(U
))
x
2
U
= 0.
With an obvious denition for the dierential operator L, the system (2.7) also
reads
(2.8) L(U
+
,
+
)U
+
= 0, L(U
)U
= 0.
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 (
+
,
). One should always

remember that the interior equations (2.7) are entirely decoupled. The coupling
between the right and left states is made by the boundary conditions (2.5).
There exist many simple solutions of (2.8)(2.5)(2.6), that correspond, in
the original variables, to rectilinear vortex sheets:
(, u) =
_
(, u
r
) if x
2
> t +nx
1
,
(, u
l
) if x
2
< t +nx
1
.
Here above, u
r
, u
l
are xed vectors in R
2
, and > 0, and n are xed real
numbers. These quantities are linked by the Rankine-Hugoniot conditions:
= v
r
n+u
r
= v
l
n+u
l
.
Changing observer if necessary, we may assume without loss of generality
= n = u
r
= u
l
= 0 and v
r
+v
l
= 0 (v
r
0).
In the new variables, this corresponds to the following regular solution of (2.8)
(2.5)(2.6):
(2.9) U
r

_
_
_
v
r
0
_
_
_, U
l

_
_
_
v
l
0
_
_
_,
r,l
(t, x
1
, x
2
) :x
2
,
with the relation v
r
+ v
l
= 0. We only consider the case v
r
0, and without
loss of generality, we assume v
r
> 0. In the next section, we study the linearized
equations around the particular solution dened by (2.9). Under a certain super-
sonic assumption, we shall show that the linearized equations satisfy an a priori
energy estimate.
3. THE CONSTANT COEFFICIENTS LINEARIZED SYSTEM
3.1. The linearized equations. Let us denote by

:
, u
:
,
:
some small per-
turbations of the exact solution given by (2.9). Up to second order, the perturba-
tions

U
:
= (
:
,

v
:
,

u
:
)
T
satisfy
t

U
+
+A
1
(U
r
)
x
1
U
+
+A
2
(U
r
)
x
2
U
+
= 0, (3.1a)
t

U
+A
1
(U
l
)
x
1
A
2
(U
l
)
x
2
= 0, (3.1b)
in the domain x
2
> 0, together with the linearized Rankine-Hugoniot relations
+
=
= , (3.2a)
(v
r
v
l
)
x
1
(
u
+
) = 0, (3.2b)
t
+v
r

x
1

u
+
= 0, (3.2c)
= 0, (3.2d)
on the boundary x
2
= 0. In short, equations (3.1)(3.2) read
(3.3)
_
L

U = 0, if x
2
> 0,
B(

U, ) = 0, if x
2
= 0,
with

U := (

U
+
,

U
), and obvious denitions for the operators L
and B:
L

U :=
t
_
U
+
_
+
_
A
1
(U
r
) 0
0 A
1
(U
l
)
_

x
1
_
U
+
_
+
_
A
2
(U
r
) 0
0 A
2
(U
l
)
_

x
2
_
U
+
_
,
B(

U, ) :=
_
_
_
_
(v
r
v
l
)
x
1
( u
+
u
t
+v
r

x
1
u
+

+

_
_
_
_
.
It is important to note that the interior equations do not involve the perturbation
, so L
is an operator that only acts on

U. This property also holds when one
studies the linearized shock wave equations around a planar shock, see [20, 25].
Proving an energy estimate for the linearized equations amounts to working
with source terms, both in the interior domain and on the boundary. From now
on, we thus consider the linear equations
(3.4)
_
L

U = f, if x
2
> 0,
B(

U, ) = g, if x
2
= 0,
and try to estimate

U and in terms of f and g (in appropriate functional spaces).
In order to simplify the subsequent calculations, we introduce some new unknown
functions, and dene the following quantities:
(3.5)
W
1
:=

v
+
, W
2
:=
1
2
(
+
/ +

u
+
/c) , W
3
:=
1
2
(
+
/ +

u
+
/c) ,
W
4
:=

v
, W
5
:=
1
2
(
/ +

u
/c) , W
6
:=
1
2
(
/ +

u
/c) .
We also dene the following vectors:
W := (W
1
, W
2
, W
3
, W
4
, W
5
, W
6
)
T
,
W
c
:= (W
1
, W
4
)
T
,
W
nc
:= (W
2
, W
3
, W
5
, W
6
)
T
.
The notations W
c
and W
nc
are introduced in order to separate the characteristic
part of the vector W and the noncharacteristic part of W. We shall go back
to this decomposition later on. It is obvious that estimating W is equivalent to
estimating

U.
Let us dene the following 6 6 symmetric matrices:
0
:=
_
_
_
_
_
_
_
_
_
1 0 0
0 2c
2
0 O
0 0 2c
2
1 0 0
O 0 2c
2
0
0 0 2c
2
_
_
_
_
_
_
_
_
_
, (3.6)
1
:=
_
_
_
_
_
_
_
_
_
v
r
c
2
c
2
c
2
2c
2
v
r
0 O
c
2
0 2c
2
v
r
v
l
c
2
c
2
O c
2
2c
2
v
l
0
c
2
0 2c
2
v
l
_
_
_
_
_
_
_
_
_
, (3.7)
2
:=
_
_
_
_
_
_
_
_
_
0 0 0
0 2c
3
0 O
0 0 2c
3
0 0 0
O 0 2c
3
0
0 0 2c
3
_
_
_
_
_
_
_
_
_
, (3.8)
where O stands for the 3 3 null matrix, as well as the following
(3.9) b :=
_
_
_
0 v
r
v
l
1 v
r
0 0
_
_
_ =
_
_
_
0 2v
r
1 v
r
0 0
_
_
_, M :=
_
_
_
c c c c
c c 0 0
1 1 1 1
_
_
_.
Then, using (3.5)(3.6)(3.9), the linear equations (3.4) equivalently read
(3.10)
_
1W = f, if x
2
> 0,
1(W
nc
, ) = g, if x
2
= 0,
with new f and g, and where we have let
1W :=
0
t
W +
1
x
1
W +
2
x
2
W,
1(W
nc
, ) := MW
nc
x
2
=0
+b
_
x
1
_
.
Note that the kernel of
2
is exactly the set of those W such that W
nc
= 0 (and
W
c
is arbitrary). The boundary x
2
= 0 is thus characteristic with multiplicity
2. As already noted in earlier works, see e.g. [19, 23], we expect to lose control of
the trace of W
c
, that is, on the trace of the tangential velocities (
v
+
,

v
). At the
opposite, we expect to control the trace of W
nc
on x
2
= 0, that is, we expect to
control the trace of (
+
,

,

u
+
,

u
).
3.2. The main result for the constant coecients case. Before stating our
energy estimate for the system(3.10), we need to introduce some Sobolev weighted
norms. First dene the half-space
:= (t, x
1
, x
2
) R
3
s.t. x
2
> 0 = R
2
R
+
.
The boundary is identied to R
2
.
For all real number s and all 1, dene the space
H
s
(R
2
) := u 1
(R
2
) s.t. exp(t)u H
s
(R
2
).
It is equipped with the norm
|u|
H
s
(R
2
)
:= |exp(t)u|
H
s
(R
2
)
.
Letting

u := exp(t)u, one has
|u|
H
s
(R
2
)
= |
u|
s,
, where
_
_
v
_
_
2
s,
:=
1
(2)
2
_
R
2
(
2
+
2
)
s
v()
2
d,
where

v is the Fourier transform of any function v dened on R
2
.
For all integers k, one can dene the space H
k
() in an entirely similar way.

The space L
2
(R
+
; H
s
(R
2
)) is equipped with the norm
2
L
2
(H
s
)
:=
_
+
0
_
_
v(, x
2
)
_
_
2
H
s
(R
2
)
dx
2
.
In the sequel, the variable in R
2
is (t, x
1
), while x
2
is the variable in R
+
.
Our rst main result is stated as follows.
Theorem 3.1. Assume that the particular solution dened by (2.9) satises
(3.11) v
r
v
l
> 2
_
2c.
Then there exists a positive constant C such that for all 1 and for all (W, )
H
2
() H
2
(R
2
), the following estimate holds:
2
L
2
()
+
_
_
W
nc
x
2
=0
_
_
2
L
2
(R
2
)
+
_
_
_
_
2
H
1
(R
2
)
(3.12)
C
_
1
1W
2
L
2
(H
1
)
+
1
2
_
_
1(W
nc
, )
_
_
2
H
1
(R
2
)
_
.
Introducing

W := exp(t)W and
`
:= exp(t), we easily nd that
(3.10) is equivalent to
(3.13)
_
_
1
W :=
0
W +1
W
= exp(t)f, if x
2
> 0,
1
W
nc
,
`
) := M
W
nc
x
2
=0
+b
_
_
`
+
t
`
x
1
`
_
_
= exp(t)g, if x
2
= 0.
Consequently, Theorem 3.1 admits the following equivalent formulation.
Theorem 3.2. Assume that (3.11) holds. Then there exists a positive constant C
such that for all 1 and for all (
W,
`
) H
2
()H
2
(R
2
), the following estimate
holds:
(3.14)
2
0
+
_
_
W
nc
x
2
=0
_
_
2
0
+
_
_
`
_
_
2
1,
C
_
1
2
1,
+
1
2
_
_
1
W
nc
,
`
)
_
_
2
1,
_
.
In (3.14), we have used the following notations for any v L
2
(R
+
; H
s
(R
2
)):
2
s,
:=
_
+
0
_
_
v(, x
2
)
_
_
2
s,
dx
2
.
For instance,
0,
is the usual norm on L
2
() and does not involve , so
we shall denote it by
0
. The norm
1,
is the weighted norm on
L
2
(R
+
, H
1
(R
2
)).
4. PROOF OF THEOREM 3.2
4.1. Some preliminary reductions. In this paragraph, we show that it is suf-
cient to prove Theorem 3.2 in the particular case 1
W 0. This rst reduction

simplies many subsequent calculations. The argument we use was introduced in
[23, page 636].
In order to simplify notations, we drop the tildas. Let W H
2
() and
H
2
(R
2
). Then we dene:
f := 1
W H
1
(), g := 1
(W
nc
, ) H
1
(R
2
).
Consider the following auxiliary problem:
(4.1)
_
_
_
1
W
1
= f, if x
2
> 0,
M
aux
W
nc
1
x
2
=0
= 0, if x
2
= 0,
where M
aux
is dened by
M
aux
:=
_
0 1 0 0
0 0 1 0
_
.
In the auxiliary problem (4.1), the boundary conditions are maximally dissipative
(see the formula (3.6) dening the matrix
2
). Since f L
2
(R
+
; H
1
(R
2
)), it
follows from [19] that there exists a function W
1
L
2
(R
+
; H
1
(R
2
)), such that the
trace of W
nc
1
on x
2
= 0 belongs to H
1
(R
2
), and that is a solution to (4.1). In
particular, the function W
1
satises the following estimates:
W
1
2
0

C
2
0
, (4.2a)
_
_
W
nc
1
x
2
=0
_
_
2
1,

C
2
1,
=
C
_
+
0
_
_
f(, x
2
)
_
_
2
1,
dx
2
. (4.2b)
Let us dene W
2
:= W W
1
. It satises
_
_
_
1
W
2
= 0, if x
2
> 0,
1
(W
nc
2
, ) = g MW
nc
1
x
2
=0
, if x
2
= 0.
Consequently, if we manage to prove that Theorem 3.2 holds true as long as the
interior source term is zero, we shall obtain
W
2
2
0
+
_
_
W
nc
2
x
2
=0
_
_
2
0
+
_
_
_
_
2
1,

C
2
_
_
g MW
nc
1
x
2
=0
_
_
2
1,
2
__
_
g
_
_
2
1,
+
_
_
W
nc
1
x
2
=0
_
_
2
1,
_
.
Then using (4.2) to estimate the H
1
norm of the trace of W
nc
1
as well as the L
2
norm of W
1
, we shall derive our main energy estimate (3.14). Without loss of
generality, we thus assume from now on that W and satisfy
(4.3)
0
W +
0
t
W +
1
x
1
W +
2
x
2
W = 0
in the interior domain , as well as the following boundary conditions
(4.4) MW
nc
x
2
=0
+b
_
_
+
t
x
1
_
_
= g, x
2
= 0.
With slight abuse of notations, we still denote the source termin the boundary
conditions by g, instead of g MW
nc
1
. This is of pure convenience.
Recall that all matrices
j
are symmetric, and that
0
is positive denite, see
(3.6). Taking the scalar product of (4.3) with W and integrating over yields the
following inequality:
2
0
C
_
_
W
nc
x
2
=0
_
_
2
0
.
Consequently, it is sucient to derive an estimate of the form
(4.5)
_
_
W
nc
x
2
=0
_
_
2
0
+
_
_
_
_
2
1,

C
2
_
_
g
_
_
2
1,
in order to obtain (3.14).
We shall derive (4.5) by means of a Kreiss symmetrizer, whose construction
is detailed in the next paragraphs. Once performed a Fourier transform in (t, x
1
),
the rst step consists in eliminating the front in the boundary conditions
(4.4). We emphasize that this operation is possible, even though the vortex sheet
is a characteristic boundary. Then we shall detail the normal modes analysis and
construct a symbolic symmetrizer.
4.2. Eliminating the front. As mentionned above, we focus on (4.3)(4.4),
and perform a Fourier transform in (t, x
1
). The dual variables are denoted by
(, ). We also dene := +i. This is the Laplace dual variable (indeed, the
previous manipulations amount to performing a Laplace transform with respect
to t). We obtain the following system of dierential equations:
(
0
+i
1
)
W +
2
d
W
dx
2
= 0, x
2
> 0, (4.6a)
b(, )
+M
W
nc
(0) =

g, (4.6b)
where b(, ) is simply dened by
(4.7) b(, ) := b
_
i
_
=
_
_
_
_
2iv
r
+iv
r
0
_
_
_
_
.
Recall that b and M are dened by (3.9). Observe that b(, ) is homogeneous
(of degree 1) with respect to (, ). In order to take such homogeneity properties
into account, we dene the hemisphere
:= (, ) C R s.t.
2
+v
2
r
2
= 1 and T 0.
Recall that is the Laplace dual variable of the time variable t, while is the
Fourier dual variable of x
1
, so / is a velocity. Our denition of takes this
property into account.
We denote by the set
:= (, , ) [0, +[ R
2
s.t. (, , ) (0, 0, 0) = ]0, +[ .
It is the set of frequencies we shall consider in the sequel. We always identify
(, ) R
2
and = +i C.
One crucial remark is that the symbol b(, ) is elliptic, that is, it does not
vanish on the closed hemisphere . More precisely, we have the following lemma.
Lemma 4.1. There exists a C
mapping Q dened on such that Q has values

in GL
3
(C), is homogeneous of degree 0, and satises
(, ) , Q(, )b(, ) =
_
_
_
0
0
(, )
_
_
_,
where is C
, homogeneous of degree 1, and has the additional property:

min
(,)
(, ) > 0.
Proof. We shall dene the mapping Q on and then extend Q by homo-
geneity. If we dene
(, ) , Q(, ) :=
_
_
_
_
0 0 1
+iv
r
2iv
r
0
2iv
r

iv
r
0
_
_
_
_
,
then we check that Qhas all the required properties. The corresponding is given
by
(, ) , (, ) := +iv
r
2
+4v
2
r
2
.
Note that the last row of Q(, ) is nothing but b(, )
, when (, ) .
Let us multiply the boundary conditions in (4.6) by the matrix Q(, ). We
obtain:
(4.8)
_
_
_
0
0
(, )
_
_
_

+
_
(, )
(, )
_
W
nc
(0) = Q(, )
g,
where has two rows while has one row and
_
(, )
(, )
_
:= Q(, )M.
The exact expression of is useless, but we shall use the expression of :
(4.9)
(, ) =
_
1 1 1 1
c( +iv
l
) c( +iv
l
) c( +iv
r
) c( +iv
r
)
_
,
(, ) .
Both and are homogeneous of degree 0 and C
on .
The last equation in (4.8) is
(, ) , (, )
+(, )
W
nc
(0) = b(, )

g,
since b
is the last row of Q. Using the ellipticity property of (see Lemma 4.1)
together with a uniform bound for and b
on , we obtain
(, ) , (
2
+v
2
r
2
)
2
C(
W
nc
(0)
2
+
g
2
).
Let us now integrate this last inequality with respect to (, ) R
2
. (Recall that
is the imaginary part of ). Using Plancherels Theorem, we obtain
_
_
_
_
2
1,
C
__
_
W
nc
x
2
=0
_
_
2
0
+
_
_
g
_
_
2
0
_
(4.10)
C
_
_
_
W
nc
x
2
=0
_
_
2
0
+
1
2
_
_
g
_
_
2
1,
_
.
In order to derive (4.5), it is therefore sucient to derive an estimate of the
trace of W
nc
. Consequently, we focus on the reduced problem
(
0
+i
1
)
W +
2
d
W
dx
2
= 0, x
2
> 0,
(, )
W
nc
(0) =

h,
and try to derive an estimate for

W
nc
(0). One has to remember that the source
term

h C
2
is easily estimated by

g, see (4.8).
In the next paragraph, we recall that under the assumption made in Theorems
3.1 and 3.2, the above boundary problemsatises the Kreiss-Lopatinskii condition
but violates the uniform Kreiss-Lopatinskii condition.
4.3. The normal modes analysis. Writing W = (W
1
, W
2
, W
3
, W
4
, W
5
, W
6
)
T
,
the two rst equations in (4.6) are:
( +iv
r
)
W
1
ic
2
W
2
+ic
2
W
3
= 0,
( +iv
l
)
W
4
ic
2
W
5
+ic
2
W
6
= 0.
They do not involve derivation with respect to the normal variable x
2
. For T >
0, we obtain an expression for

W
1
and

W
4
that we plug in the last four equations.
This operation yields a system of ordinary dierential equations of the following
form:
(4.11)
_
_
d
W
nc
dx
2
= (, )
W
nc
, if x
2
> 0,
(, )
W
nc
(0) =

h, if x
2
= 0.
The matrix (, ) in (4.11) is given by
(, ) :=
_
_
_
_
_
r
m
r
0 0
m
r

r
0 0
0 0
l
m
l
0 0 m
l

l
_
_
_
_
_
, (4.12)
r,l
:=
(1/c)( +iv
r,l
)
2
+(c/2)
2
+iv
r,l
,
m
r,l
:=
(c/2)
2
+iv
r,l
.
A well-known result, that is due to Hersh [15] in the noncharacteristic case
(see [23] for the extension to the characteristic case), asserts that the matrix (, )
has no purely imaginary eigenvalue as long as T > 0. As a consequence, the sta-
ble subspace of (, ) has constant dimension when T > 0. This dimension
equals the number of characteristics going out of the discontinuity. In our case,
those theoretical results can be checked directly by computing the eigenvalues and
the stable subspace of (, ). The following lemma gives an expression of the
stable subspace.
Lemma 4.2. Let C and R, with T > 0 and (, ) . The
eigenvalues of (, ) are the roots of the dispersion relations
2
=
2
r
m
2
r
=
1
c
2
( +iv
r
)
2
+
2
, (4.13a)
2
=
2
l
m
2
l
=
1
c
2
( +iv
l
)
2
+
2
. (4.13b)
In particular, (4.13a) (resp. (4.13b)) admits a unique root
r
(resp.
l
) of negative
real part. The other root of (4.13a) (resp. (4.13b)) is
r
(resp.
l
), and has
positive real part. The stable subspace 1
(, ) of (, ) has dimension 2, and is

spanned by the two following vectors:
E
r
(, ) :=
_
c
2
2
,
1
c
( +iv
r
)
2
+
c
2
2
( +iv
r
)
r
, 0, 0
_
T
, (4.14a)
E
l
(, ) :=
_
0, 0,
1
c
( +iv
l
)
2
+
c
2
2
( +iv
l
)
l
,
c
2
2
_
T
. (4.14b)
Both
r
and
l
admit a continuous extension to any point (, ) such that
T = 0 and (, ) . This allows to extend both vectors E
r
and E
l
in (4.14) to
the whole hemisphere . Those two vectors are linearly independent for any value of
(, ) .
The symbol (, ) is diagonalizable as long as both
r
and
l
do not vanish,
that is, when (:v
r
: c)i. Away from such points, admits a C
basis of
eigenvectors.
The proof follows from straightforward computations, and we shall omit it.
We point out that the stable subspace 1
(, ) is dened for all (, ) ,

while the matrix (, ) has some poles on the boundary of , see (4.12). The
poles are exactly those points (, ) verifying = iv
r,l
= iv
r
(recall
that we have v
r
= v
l
0).
Following Majda and Osher [23], we dene the Lopatinskii determinant as-
sociated with the boundary conditions in the following way:
(4.15) (, ) := det
_
(, )
_
E
r
(, ) E
l
(, )
__
,
with dened by (4.9) and (E
r
, E
l
) dened by (4.14). We emphasize that the
Lopatinskii determinant is dened on the whole hemisphere and is continu-
ous with respect to (, ). The rst step in proving an energy estimate for (4.11)
consists in determining whether vanishes on . The answer is given in the
following result.
Proposition 4.3. Assume that (3.11) holds. Then there exists a positive number
V
1
such that for any (, ) , one has (, ) = 0 if and only if
= 0 or = :iV
1
.
Each of these roots is simple. For instance, there exists a neighborhood ` of (0, 1/v
r
)
in and a C
function h dened on ` such that

(, ) `, (, ) = h(, ) and h(0, 1/v
r
) 0.
A similar result holds near (0, 1/v
r
) or near the points (:iV
1
, ) .
The number V
1
is given by V
2
1
:= c
2
+v
2
r
c
_
c
2
+4v
2
r
. In particular, one has:
0 < V
1
< v
r
c. When = 0 or = :iV
1
, both eigenmodes
r
and
l
are
purely imaginary.
We postpone the proof of Proposition 4.3 to Appendix A. We simply note
here that the three critical speeds V
1
, 0, V
1
are exactly the speeds of the kink
modes exhibited in the work by Artola and Majda [2]. As a matter of fact, Artola
and Majda used a geometric optics approach, while we have followed here a
normal modes analysis approach. However, the calculations are similar in both
cases (in [2], the number equals 1).
4.4. Constructing a symmetrizer: the interior points. We now turn to the
construction of our degenerate Kreiss symmetrizer. The construction is microlo-
cal and is achieved near any point (
0
,
0
) . The analysis is rather long since
one has to distinguish between ve dierent cases. In the end, we shall consider a
partition of unity to patch things together and derive our energy estimate.
In all the rest of the article, the letter denotes a generic positive constant
(typically, though not necessarily, a rather small one).
We rst consider the case (
0
,
0
) with T
0
> 0. Then the matrix
(, ) is diagonalizable for all (, ) close to (
0
,
0
). A smooth (that is, C
)
basis of eigenvectors is given by the following family:
E
r
(, ), E
l
(, ),
_
c
2
2
,
1
c
( +iv
r
)
2
+
c
2
2
+( +iv
r
)
r
, 0, 0
_
T
,
_
0, 0,
1
c
( +iv
l
)
2
+
c
2
2
+( +iv
l
)
l
,
c
2
2
_
T
.
Both vectors E
r
and E
l
are dened by (4.14). Therefore, there exists a C
mapping
T(, ), dened on a neighborhood ` of (
0
,
0
) in , with values in GL
4
(C),
and such that
(, ) `, T(, )(, )T(, )
1
=
_
_
_
_
_
r
0 0 0
0
l
0 0
0 0
r
0
0 0 0
l
_
_
_
_
_
.
The rst two columns of the matrix T(, )
1
are the vectors E
r
and E
l
. In the
neighborhood `, we dene the symmetrizer r in the usual way:
(, ) `, r(, ) :=
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 K 0
0 0 0 K
_
_
_
_
_
,
where K 1 is a positive real number, to be xed large enough. In what follows,
we use the standard notation
TM :=
M +M
2
for all square matrix M with complex entries (M
is the classical adjoint matrix).

The matrix r dened just above is hermitian, and we have
(4.16) (, ) `, T(r(, )T(, )(, )T(, )
1
) I I,
for some positive constant . This is because
r
and
l
have negative real part
when (, ) `, and 1 when (, ) .
We now simply need to x K 1 in order to recover an estimate for the trace
of W
nc
. We show that for K suciently large, the following inequality holds:
(4.17) (, ) `, r(, ) +C(
(, ))

(, ) I,
where C is a positive constant and

(, ) := (, )T(, )
1
.
Let Z = (Z
, Z
+
)
T
C
4
, with Z
, Z
+
C
2
. We write
(, )Z =

(, )
_
Z
0
_
+

(, )
_
0
Z
+
_
,
and recall that the rst two columns of T(, )
1
are E
r
and E
l
. Because the
Lopatinskii determinant does not vanish at (
0
,
0
), see (4.15) and Proposition
4.3, we obtain an estimate of the form
Z
2
C
0
(Z
+
2
+
(, )Z
2
),
for a suitable constant C
0
that is independent of (, ) `. With C
0
satisfying
this inequality, the denition of r yields
r(, )Z, Z)
C
4 +2C
0
(, )Z
2
= Z
2
+KZ
+
2
+2C
0
(, )Z
2
Z
2
+(K 2C
0
)Z
+
2
.
This gives (4.17) for K large enough (e.g., K = 2C
0
+1).
4.5. Constructing a symmetrizer: the boundary points (case 1). We now
turn to the construction of the symmetrizer near those points (, ) such that
T = 0. We rst prove a general result on the behavior of the eigenmodes
r,l
in
the neighborhood of such points.
Lemma 4.4. Let (
0
,
0
) so that T
0
= 0 and
0
(v
r
: c)i
0
. In
particular,
r
depends analytically on (, ) near (
0
,
0
). Then the two following
cases may occur:
(1) The eigenmode
r
has negative real part at (
0
,
0
), and, in a suitable neigh-
borhood ` of (
0
,
0
), one has
T
r
< 0.
(2) The eigenmode
r
is purely imaginary at (
0
,
0
). In this case, the derivative
r
calculated at (
0
,
0
) is a nonzero real number. In a suitable neighborhood
` of (
0
,
0
) in , we have
T
r
.
A completely similar result holds for
l
near all points (
0
,
0
) satisfying T
0
=
0 and
0
(v
l
:c)i
0
.
Proof. Because
0
(v
r
:c)i
0
, the eigenmode
r
is not zero at (
0
,
0
)
and it depends analytically on (, ) by the implicit functions theorem.
The rst case in Lemma 4.4 simply follows from the continuity of
r
with
respect to (, ). In the second case, we use (4.13a) to derive
=
1
c
2
( +iv
r
).
When =
0
and =
0
, one has
r
iR \ 0 and (
0
+iv
r
0
) iR. This
proves that the derivative
r
is real. We now remark that
0
iv
r
0
for, in
such a case,
r
has negative real part. Consequently, the derivative
r
is not
zero. The estimate on T
r
is obtained by performing a Taylor expansion of
r
at (
0
,
0
).
According to Proposition 4.3, there are exactly four types of points on the
boundary of :
(1) Those points (
0
,
0
) where (
0
,
0
) is diagonalizable and the Lopatinskii
condition is satised at (
0
,
0
).
(2) Those points (
0
,
0
) where (
0
,
0
) is diagonalizable and the Lopatinskii
condition breaks down at (
0
,
0
).
(3) Those points (
0
,
0
) where (
0
,
0
) is not diagonalizable, that is,
0
=
(:v
r
:c)i
0
. In this case, Proposition 4.3 asserts that the Lopatinskii condi-
tion is satised at (
0
,
0
).
(4) Those points (
0
,
0
) that are the poles of , that is,
0
= :iv
r
0
. At those
points, the Lopatinskii condition is satised.
As a matter of fact, an immediate consequence of Proposition 4.3 is that when-
ever the Lopatinskii condition fails at (
0
,
0
), then (
0
,
0
) is not a pole and the
symbol (, ) is (smoothly) diagonalizable in a neighborhood of (
0
,
0
). The
three rst categories of boundary points can thus be treated as in [6, 10, 17, 28],
provided the technical assumptions of [10] near instability points hold. We are
going to show that such technical assumptions hold true. The last category of
boundary points (the poles of the symbol ) requires special attention.
We now deal with the construction of our symmetrizer in case 1: (
0
,
0
)
is such that (
0
,
0
) is diagonalizable and the Lopatinskii condition is satised
at (
0
,
0
), that is, (
0
,
0
) 0.
Because (
0
,
0
) is diagonalizable, we have
0
(:v
r
: c)i
0
. Hence
there exists a neighborhood ` of (
0
,
0
) in and a smooth basis of eigenvectors
of dened on `. The smooth basis is the same as in the case of interior points
(see the preceding paragraph). We thus have
(, ) `, T(, )(, )T(, )
1
=
_
_
_
_
_
r
0 0 0
0
l
0 0
0 0
r
0
0 0 0
l
_
_
_
_
_
,
where, once again, the two rst columns of T(, )
1
are exactly E
r
and E
l
. We
choose r as in the case of interior points:
(, ) `, r(, ) :=
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 K 0
0 0 0 K
_
_
_
_
_
,
with K 1 to be xed large enough. Clearly, r is a hermitian matrix. Using
Lemma 4.4, we can already conclude that
(4.18) (, ) `, T(r(, )T(, )(, )T(, )
1
) I.
Because the Lopatinskii condition is satised at (
0
,
0
), it is possible to choose K
large enough so that the following estimate holds:
(4.19) (, ) `, r(, ) +C(
(, ))

(, ) I.
In (4.19), we have let, as usual,

(, ) := (, )T(, )
1
. The analysis is the
same as what we have done for interior points. The estimates (4.18)(4.19) are
similar (but not exactly identical) to (4.16)(4.17).
4.6. Constructing a symmetrizer: the boundary points (case 2). In this
paragraph, we consider a point (
0
,
0
) such that the Lopatinskii determinant
vanishes at (
0
,
0
). From Proposition 4.3, we know that the symbol is
smoothly diagonalizable on a neighborhood ` of (
0
,
0
) in . In other words,
there exists a smooth mapping T(, ) with values in GL
4
(C), and such that
(, ) `, T(, )(, )T(, )
1
=
_
_
_
_
_
r
0 0 0
0
l
0 0
0 0
r
0
0 0 0
l
_
_
_
_
_
.
The rst two columns of T(, )
1
are E
r
and E
l
. In this case, we dene our
symmetrizer r in the following (degenerate) way:
(, ) `, r(, ) :=
_
_
_
_
_
2
0 0 0
0
2
0 0
0 0 K 0
0 0 0 K
_
_
_
_
_
,
with K 1 to be xed large enough. The matrix r(, ) is hermitian and we have
(4.20) T(r(, )T(, )(, )T(, )
1
)
_
_
_
_
_
2
0 0 0
0
2
0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
,
(, ) `.
It only remains to x K appropriately in order to recover an estimate on the
boundary. The choice of K relies on the following lemma.
Lemma 4.5. Let (
0
,
0
) be a point where the Lopatinskii determinant
vanishes. Then there exists a neighborhood ` of (
0
,
0
) in and a constant
0
> 0
such that the following estimate holds for all (, ) `:
Z
C
2
,
(, )
_
E
r
(, ) E
l
(, )
_
Z
2

0
2
Z
2
.
Before proving Lemma 4.5, let us rst show that it enables us to obtain an
estimate between r and the boundary matrix . More precisely, we are going to
show the following estimate:
(4.21) (, ) `, r(, ) +C(
(, ))

(, )
2
I,
for an appropriate positive constant C. The denition of

(, ) is the same as in
the preceding cases.
Let Z = (Z
, Z
+
)
T
C
4
, where Z
and Z
+
belong to C
2
. Once again, we
write
(, )Z =

(, )
_
Z
0
_
+

(, )
_
0
Z
+
_
,
and we recall that the rst two columns of T(, )
1
are E
r
and E
l
. Using Lemma
4.5, we obtain
2
Z
2
C
0
(Z
+
2
+
(, )Z
2
).
We thus derive
r(, )Z, Z)
C
4 +
2C
0
(, )Z
2
=
2
Z
2
+KZ
+
2
+
2C
0
(, )Z
2

2
Z
2
+
_
K
2C
0
0
_
Z
+
2
.
Choosing K = 2C
0
/
0
+1 yields
r(, )Z, Z)
C
4 +
2C
0
(, )Z
2

2
Z
2
+Z
+
2

2
Z
2
.
Here we have much weakened our estimate for the last two components Z
+
. How-
ever, an inequality like (4.21) is simpler to use since it does not distinguish between
the dierent coordinates of the vector Z. One should remember that the real loss
of control is on the modes
r
and
l
, and not on the modes
r
and
l
.
The proof of Lemma 4.5 is postponed to Appendix A. It relies on the fact that
the roots of the Lopatinskii determinant are simple (see Proposition 4.3).
4.7. Constructing a symmetrizer: the boundary points (case 3). In this
paragraph, we consider a point (
0
,
0
) such that
0
= iv
r
0
:ic
0
. (The
case
0
= iv
l
0
:ic
0
is entirely similar, and we shall not detail it). Because v
l
=
v
r
and v
r
> c
2, we have
0
iv
l
0
:ic
0
, and therefore, the eigenmode
l
depends smoothly on (, ) in a neighborhood ` of (
0
,
0
). Oppositely,
r
is only continuous with respect to (, ) near (
0
,
0
), but
2
r
is C
near
(
0
,
0
). This is because (4.13a) has a double root when (, ) = (
0
,
0
).
When (, ) is close to (
0
,
0
), the following family is a C
basis of C
4
:
(m
r
, m
r
, 0, 0)
T
, (i, 0, 0, 0)
T
, E
l
(, ),
_
0, 0,
1
c
( +iv
l
)
2
+
c
2
2
+( +iv
l
)
l
,
c
2
2
_
T
.
Recall that m
r
is dened by (4.12). Let T(, )
1
be the (regular) matrix whose
columns are those four vectors. We compute
T(, )(, )T(, )
1
=
_
_
_
a
r
O O
O
l
0
O 0
l
_
_
_,
where a
r
is the 2 2 matrix dened as follows:
a
r
(, ) :=
_
_
_
_
_
c
2
r
+iv
r
i
2im
r
c
2
r
+iv
r
c
2
r
+iv
r
_
_
_
_
_
.
In particular, we have
a
r
(
0
,
0
) =
_
0 i
0 0
_
=: iN.
Moreover, we make the following observations:
For all (, ) close to (
0
,
0
) so that iR, a
r
has purely imaginary coe-
cients.
The lower left coecient
r
of a
r
satises
(
0
,
0
) R\ 0.
Here we have exhibited a basis in which Ralstons result [28] applies. Readers who
are familiar with the theory will have recognized the block structure condition,
that is a consequence here of the strict hyperbolicity of (2.1), see [24].
We are looking for a symmetrizer r under the form
r(, ) =
_
_
_
s(, ) O O
O 1 0
O 0 K
_
_
_,
where K 1 is a real number, to be xed large enough, and s is some 2 2
hermitian matrix, depending smoothly on (, ). More precisely, we are looking
for the matrix s under the following form
s(, ) =
_
0 e
1
e
1
e
2
_
. , .
E
+
_
f(, ) 0
0 0
_
. , .
F(,)
i
_
0 g
g 0
_
. , .
G
,
where e
1
, e
2
and g are real numbers, and f is a real valued C
mapping that
vanishes at (
0
,
0
), see [6, 17, 28]. We choose e
1
in the following way:
e
1
:=
_
(
0
,
0
)
_
1
R\ 0,
where
r
is dened just above. This choice may look surprising but it will be
justied later on.
Now we observe that our choice of s yields (
0
= 0):
r(
0
,
0
) =
_
_
_
_
_
0 e
1
0 0
e
1
e
2
0 0
0 0 1 0
0 0 0 K
_
_
_
_
_
.
Moreover, the rst and third columns of T(
0
,
0
)
1
are nothing but E
r
(
0
,
0
)
and E
l
(
0
,
0
). This is due to the equality
r
= m
r
when (, ) = (
0
,
0
).
Because Lopatinskiis condition is satied at (
0
,
0
), we can choose e
2
and K
large enough such that the following estimate holds:
r(
0
,
0
) +C(
(
0
,
0
))

(
0
,
0
) I.
As was done before, we have let

(, ) := (, )T(, )
1
. Up to shrinking
`, we have thus derived the following estimate
(4.22) (, ) `, r(, ) +C(
(, ))

(, )
1
2
I,
for a suitable constant C.
Now, we show how to choose the real valued function f and the real number
g. We rst write
a
r
(, ) = a
r
(
0
,
0
) +(a
r
(i, ) a
r
(
0
,
0
)) +(a
r
(, ) a
r
(i, )),
and then use Taylors formula. We obtain
a
r
(, ) = iN +(a
r
(i, ) a
r
(
0
,
0
)) +
a
r
(i, ) +
2
M(, ),
for a suitable continuous function M. Because a
r
has purely imaginary coecients
when is purely imaginary, we have
a
r
(i, ) a
r
(
0
,
0
) =
_
ib
1
(i, ) 0
ib
2
(i, ) ib
3
(i, )
_
=: iB
r
(i, ),
for some C
, real valued mappings b

1
, b
2
, b
3
. Those three mappings obviously
vanish at (
0
,
0
). We choose
f(, ) := e
1
(b
1
(i, ) b
3
(i, )) +e
2
b
2
(i, ),
so that f is a C
real valued mapping that vanishes at (

0
,
0
). Moreover, this
choice of f implies that the matrix
(E +F(, ))(N +B
r
(i, ))
is real and symmetric for all (, ). Consequently, we obtain
T(s(, )a
r
(, )) = T
_
GN +E
a
r
(i, )
_
+L(, ),
where L is continuous (it is even C
) and L(
0
,
0
) = 0. Our choice for e
1
yields
T
_
GN +E
a
r
(
0
,
0
)
_
=
_
0 0
0 g
_
+
_
1

_
,
where the are coecients that only depend on e
1
, e
2
(that have already been
xed) and (
0
,
0
). Choosing g large enough, and shrinking ` if necessary, we
end up with
T(s(, )a
r
(, ))
1
4
I,
and we thus obtain
(4.23) (, ) `, T(r(, )T(, )(, )T(, )
1
) I.
4.8. Constructing a symmetrizer: the boundary points (case 4). We now
consider the last case, which is (
0
,
0
) with
0
= iv
r
0
. (We shall not
detail the case
0
= iv
l
0
that is entirely similar). The symbol is not dened
at (
0
,
0
), while the stable subspace 1
of admits a continuous extension at

this point. The family (E
r
, E
l
) is a C
basis of 1
near (
0
,
0
), see Lemma 4.2,
and Lopatinskiis condition is satised near (
0
,
0
), see Proposition 4.3.
The eigenmode
r
has negative real part when (, ) is close to (
0
,
0
), see
(4.13a). Oppositely, the eigenmode
l
is purely imaginary when (, ) is close
to (
0
,
0
) and T = 0, see (4.13b). Furthermore, both
r
and
l
depend
analytically on (, ) in a neighborhood ` of (
0
,
0
).
The matrix is not smoothly diagonalizable on `. This is because the eigen-
vector associated with the eigenvalue
r
tends to be parallel to the eigenvector
associated with the eigenvalue
r
. Consequently, Majda and Oshers construc-
tion of a symmetrizer in this case involves a singularity in the symmetrizer, see
[23]. We prefer to avoid this singularity and construct a smooth (that is, C
)
symmetrizer in the whole neighborhood `. This is possible if we go back to the
original system:
(
0
+i
1
)
W +
2
d
W
dx
2
= 0, x
2
> 0, (4.24a)
(, )
W
nc
(0) =

h, (4.24b)
The following analysis is inspired from [4]. For (, ) in a neighborhood ` of
(
0
,
0
), the following matrix is regular (that is, invertible):
(4.25) J(, ) :=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
1 ic( +iv
r
c
r
) 0
0
c
2
2
0 O
0 ( +iv
r
)(
r

r
) 1
1 ic( +iv
l
c
l
) ic( +iv
l
+c
l
)
O 0 ( +iv
l
)(
l
l
) ( +iv
l
)(
l
+
l
)
0
c
2
2
c
2
2
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
.
To avoid overloading the equations, we shall simply denote by O the 3 3 zero
matrix. The determinant of J (, ) is given by
det J (, ) =
c
2
2

4
( +iv
l
)
l
,
and it is easy to check that this quantity does not vanish near (
0
,
0
). The matrix
J depends smoothly on (, ) in the whole neighborhood `. It has no pole.
Let us dene
(4.26a)
r
:= ( +iv
r
)(
r

r
) =
1
c
( +iv
r
)
2
+
c
2
2
( +iv
r
)
r
,
(4.26b)
:
l
:= ( +iv
l
)(
l
:
l
) =
1
c
( +iv
l
)
2
+
c
2
2
:( +iv
l
)
l
.
The main idea now is that (4.24a) has a simple expression if we decompose
the vector

W in the basis dened by J . In other words, the matrices (
0
+
i
1
)J (, ) and
2
J (, ) have a rather similar structure. Performing some
manipulations on the rows of these two matrices, we shall transform (4.24a) into
an almost diagonal system of dierential equations.
After some simplications, we obtain the following expressions:
(
0
+i
1
)J
=
_
_
_
_
_
_
_
_
_
_
_
_
+iv
r
0 ic
2
ic
2
c
4
r
0 O
ic
2
2c
3
r
2c
2
( +iv
r
)
3pt] +iv
l
0 0
3pt] O ic
2
2c
3
l
2c
3
+
l
ic
2
c
4
l
c
4
l
_
_
_
_
_
_
_
_
_
_
_
_
,
2
J =
_
_
_
_
_
_
_
_
_
_
_
_
0 0 0
0 c
4
2
0 O
0 2c
3
r
2c
3
0 0 0
O 0 2c
3
l
2c
3
+
l
0 c
4
2
c
4
2
_
_
_
_
_
_
_
_
_
_
_
_
,
where
r
and
:
l
are dened by (4.26a)(4.26b). These expressions may look
somehow complicated, but it is not so hard to simplify them multiplying on the
left by a suitable symmetrizer. Indeed, for all (, ) in a neighborhood ` of
(
0
,
0
), let us dene the following matrix:
(4.27) R
1
(, ) :=
_
_
_
_
_
_
_
_
_
_
_
_
1 0 0
0 1 0 O
2ic
4
( +iv
r
c
r
) 2c
3
r
c
4
2
1 0 0
O 2ic
4
( +iv
l
+c
l
) c
4
2
2c
3
+
l
2ic
4
( +iv
l
c
l
) c
4
2
2c
3
l
_
_
_
_
_
_
_
_
_
_
_
_
.
First we obtain
R
1
2
J = diag(0, c
4
2
, 2c
7
2
, 0, 4c
7
2
( +iv
l
)
l
, 4c
7
2
( +iv
l
)
l
),
where diag(a
1
, . . . , a
p
) stands for the diagonal matrix whose diagonal elements
are a
1
, . . . , a
p
. We also obtain the following expression:
(4.28) R
1
(
0
+i
1
)J =
_
_
_
_
_
_
_
_
_
_
_
_
+iv
r
0 ic
2
ic
2
c
4
r
0 O
0 0 2c
7
r
+iv
l
0 0
O 0 4c
7
2
l
( +iv
l
) 0
0 0 4c
7
2
l
( +iv
l
)
_
_
_
_
_
_
_
_
_
_
_
_
.
Recall that
l
, and (+iv
l
) do not vanish in the neighborhood ` of (
0
,
0
),
up to shrinking `. The following matrix is therefore a C
mapping of (, ) on
`:
(4.29) R
2
(, )
:= diag
_
1, 1,
1
2c
7
2
, 1,
1
4c
7
l
( +iv
l
)
,
1
4c
7
l
( +iv
l
)
_
.
We dene
S(, ) := R
2
(, )R
1
(, ),
where R
1
and R
2
are dened by (4.27)(4.29). It is now easy to derive the follow-
ing equalities for all (, ) in `:
S(, )
2
J (, ) = diag(0, c
4
2
, 1, 0, 1, 1), (4.30a)
S(, )(
0
+i
1
)J (, ) (4.30b)
=
_
_
_
_
_
_
_
_
_
+iv
r
0 ic
2
ic
2
c
4
r
0 O
0 0
r
+iv
l
0 0
O 0
l
0
0 0
l
_
_
_
_
_
_
_
_
_
.
In particular, it is important to observe that both matrices S and J are C
on the
whole neighborhood `. Up to shrinking `, we may assume that
r
has negative
real part for all (, ) `. This is possible because
r
= at (
0
,
0
).
Though a little complicated, the preceding calculations are based on the sim-
ple idea that the dierential equations (4.24a) have an easy expression if we replace
the standard coordinates by the coordinates on the stable subspace. The diculty
comes fromthe fact that we deal with the dierential systemsatised by

W and not
with the system satised by

W
nc
. Because the boundary is characteristic, (4.24a) is
not an ordinary dierential equation, and we thus need to diagonalize simultane-
ously (
0
+i
1
) and
2
. Even though the matrix S(
0
+i
1
)J is not
diagonal, we shall see that the reduced expressions (4.30a)(4.30b) are sucient
to derive energy estimates in such a neighborhood ` of the pole (
0
,
0
).
4.9. Derivation of the energy estimate. We now turn to the derivation of
the estimate (4.5). Recall that we are considering a function W H
1
() such
that
(
0
+i
1
)
W +
2
d
W
dx
2
= 0, x
2
> 0,
(, )
W
nc
(0) =

h,
where

h is obtained fromthe source term

g in (4.6) after eliminating the unknown
front:
(, ) ,

h =
_
0 0 1
+iv
r
2iv
r
0
_

g.
Thanks to (4.10), it is sucient to get an estimate of the trace of W
nc
on
x
2
= 0 in order to derive (4.5).
The previous analysis shows that for all (
0
,
0
) , there exists a neighbor-
hood ` of (
0
,
0
) in and mappings dened on this neighborhood that satisfy
certain properties. Because is a C
compact manifold, there exists a nite cov-

ering (`
1
, . . . , `
I
) of by such neighborhoods, and a smooth partition of unity
(
1
, . . . ,
I
) associated with this covering. Namely, the
i
s are nonnegative, C
,
and satisfy
Supp
i
`
i
and
I
_
i=1
2
i
1.
There are three dierent cases.
In the rst case, `
i
is a neighborhood of an interior point or a neighborhood
of a boundary point corresponding to cases 1 and 3 above (boundary points that
are not poles and for which the Lopatinskii condition is satised). On such a
neighborhood, there exist two C
mappings r
i
and T
i
such that
r
i
is hermitian,
T
i
has values in GL
4
(C),
the following estimates hold for all (, ) `
i
:
T(r
i
(, )T
i
(, )(, )T
i
(, )
1
)
i
I, (4.31a)
r
i
(, ) +C
i
((, )T
i
(, )
1
)
(, )T
i
(, )
1
I. (4.31b)
The inequalities (4.31) are direct consequences of (4.16)(4.17)(4.18)(4.19)
(4.22)(4.23).
We dene
U
i
(, , x
2
) :=
i
(, )T
i
(, )
W
nc
(, , x
2
).
Both mappings r
i
and T
i
are not dened on but only on the neighborhood
`
i
. However, only the values of these mappings on the support of
i
will be
involved in the subsequent calculations, so we choose for convenience to extend
these mappings to the whole hemishpere . Then we extend
i
, r
i
, T
i
to the
whole set of frequencies as homogeneous mappings of degree 0 with respect to
(, ).
Because `
i
does not contain any pole of the symbol , which is dened by
(4.12), one easily shows that U
i
satises
dU
i
dx
2
= T
i
(, )(, )T
i
(, )
1
U
i
, x
2
> 0,
(, )T
i
(, )
1
U
i
(0) =
i
h.
We take the scalar product of this later ordinary dierential equation with r
i
U
i
and integrate with respect to x
2
on R
+
. Then we take the real part of the obtained
equality and use (4.31). These operations yield the classical Kreiss estimate:
_
+
0
U
i
(, , x
2
)
2
dx
2
+
1
2
U
i
(, , 0)
2
C
i
i
(, )
2
h(, )
2
.
Now we use the denition of U
i
and a uniform bound for |T
i
(, )
1
| on the
support of
i
, to derive
i
(, )
2
_
+
0
W
nc
(, , x
2
)
2
dx
2
+
i
(, )
2
W
nc
(, , 0)
2
(4.32)
C
i
i
(, )
2
h
2
.
In the second case, `
i
is a neighborhood of a zero of the Lopatinskii deter-
minant. On such a neighborhood, there exist two C
mappings r
i
and T
i
such
that
r
i
is hermitian,
T
i
has values in GL
4
(C),
the following estimates hold for all (, ) `
i
:
T(r
i
(, )T
i
(, )(, )T
i
(, )
1
)
i
3
I, (4.33a)
r
i
(, ) +C
i
((, )T
i
(, )
1
)
(, )T
i
(, )
1

2
I. (4.33b)
These inequalities correspond to (4.20)(4.21).
As was done before, we rst extend r
i
and T
i
as C
mappings on the whole

hemisphere . Then we extend T
i
and
i
as homogeneous mappings of degree 0
with respect to (, ), and we extend r
i
as a homogeneous mapping of degree 2
with respect to (, ). Thus (4.33) reads
(4.34a) T(r
i
(, )T
i
(, )(, )T
i
(, )
1
)
i
3
I,
(4.34b) r
i
(, ) +C
i
(
2
+v
2
r
2
)((, )T
i
(, )
1
)
(, )T
i
(, )
1

2
I,
for all (, ) R
+
`
i
. Once again, we dene
U
i
(, , x
2
) :=
i
(, )T
i
(, )
W
nc
(, , x
2
).
Because `
i
does not contain any pole of , we still have
dU
i
dx
2
= T
i
(, )(, )T
i
(, )
1
U
i
, x
2
> 0,
(, )T
i
(, )
1
U
i
(0) =
i
h.
We perform the same calculations as above (in the rst case), but now we use
(4.34) instead of (4.31). We obtain
i
(, )
2
_
+
0
W
nc
(, , x
2
)
2
dx
2
+
i
(, )
2
W
nc
(, , 0)
2
(4.35)
C
i
i
(, )
2
h
2
(
2
+v
2
r
2
).
In the third and last case, `
i
is a neighborhood of a pole of the symbol . For
instance, `
i
is a neighborhood of a point (iv
r
0
,
0
) . In this case, we have
shown that there exists C
mappings J
i
and S
i
dened on `
i
such that
J
i
has values in GL
6
(C),
both relations (4.30a)(4.30b) hold on `
i
.
Recall that
r
has negative real part on `
i
.
Here, we extend
i
, J
i
and S
i
as homogeneous mappings of degree 0 with
respect to (, ). Moreover, we shall make as if J
i
and S
i
were dened on the
whole hemisphere (this is of pure convenience since only the values on the
support of
i
are involved in what follows). We dene
U
i
(, , x
2
) :=
i
(, )J
i
(, )
1
W(, , x
2
) C
6
.
The components of the vector U
i
are denoted as follows
U
i
= (U
i,1
, U
i,2
, . . . , U
i,6
)
T
,
and we also dene
U
nc
i
:= (U
i,2
, U
i,3
, U
i,5
, U
i,6
)
T
C
4
.
Using the denition (4.25) of J
i
(, ), it is clear that the vector U
nc
i
is given by a
relation of the form
U
nc
i
= T
i
(, )
1
W
nc
, with T
i
(, ) GL
4
(C).
We also note that the rst and third column vectors of the matrix T
i
(, ) are
exactly the vectors E
r
(, ) and E
l
(, ). With these notations and denitions,
we get
(
0
+i
1
)J
i
(, )U
i
+
2
J
i
(, )
dU
i
dx
2
= 0, x
2
> 0,
(, )T
i
(, )U
nc
i
(0) =
i
(, )
h.
Multiplying the equation in x
2
> 0 by S
i
(, ) and using (4.30a)(4.30b), we
obtain the following system:
( +iv
r
)U
i,1
+ic
2
U
i,3
= 0, (4.36a)
ic
2
U
i,1
+
c
4
2
+v
2
r
r
U
i,2
c
4
2
+v
2
r
2
dU
i,2
dx
2
= 0, (4.36b)
r
U
i,3
+
dU
i,3
dx
2
= 0, (4.36c)
( +iv
l
)U
i,4
= 0, (4.36d)
l
U
i,5
dU
i,5
dx
2
= 0, (4.36e)
l
U
i,6
+
dU
i,6
dx
2
= 0. (4.36f)
Recall that when (, ) belongs to the conical set R
+
`
i
, one has
T
r
(
2
+v
2
r
2
)
1/2
and T
l
,
for a suitable constant > 0. Because U
i,3
(x
2
) and U
i,6
(x
2
) belong to L
2
(R
+
),
for > 0, (4.36c) and (4.36f) imply U
i,3
0 and U
i,6
0. Using (4.36a) and
(4.36d), we also obtain U
i,1
0 and U
i,4
0. Eventually, (4.36b) and (4.36e)
reduce to
r
U
i,2
dU
i,2
dx
2
= 0 and
l
U
i,5
dU
i,5
dx
2
= 0.
Because the rst and third columns of T
i
(, ) are nothing but E
r
and E
l
dened
in Lemma 4.2, the boundary conditions for U
i,2
and U
i,5
read:
(4.37) (, )
_
E
r
(, ) E
l
(, )
_
_
_
U
i,2
(0)
U
i,5
(0)
_
_
=
i
(, )
h.
Using the properties of
r
and
l
on the conical set R
+
`
i
, we derive
(
2
+v
2
r
2
)
1/2
_
+
0
U
i,2
(, , x
2
)
2
dx
2
CU
i,2
(, , 0)
2
,
_
+
0
U
i,5
(, , x
2
)
2
dx
2
CU
i,5
(, , 0)
2
.
Because the uniform Lopatinskii condition is satised on `
i
, (4.37) yields the
following estimate:
_
U
i,2
(, , 0)
U
i,5
(, , 0)
_
2
C
i
(, )
2
h
2
.
Eventually, we obtain
_
+
0
U
nc
i
(, , x
2
)
2
dx
2
+U
nc
i
(, , 0)
2
C
i
(, )
2
h
2
.
We now use the denition of the vector U
nc
i
to derive
(4.38)
2
i
_
+
0
W
nc
(, , x
2
)
2
dx
2
+
2
i
W
nc
(, , 0)
2
C
2
i
h
2
.
We now add up (4.32)(4.35)(4.38), and use that the
i
s form a partition
of unity. We obtain
_
+
0
W
nc
(, , x
2
)
2
dx
2
+
W
nc
(, , 0)
2
h
2
(
2
+v
2
r
2
).
We have already recalled that

h is simply obtained from the source term

g in (4.6)
by multiplying by a uniformly bounded matrix. Thus integrating the previous
inequality with respect to (, ) R
2
and using Plancherels theorem yields the
desired estimate:
W
nc
2
0
+
_
_
W
nc
x
2
=0
_
_
2
0

C
2
_
_
g
_
_
2
1,
.
Combining with (4.10), we have nished to prove (4.5).
5. THE VARIABLE COEFFICIENTS LINEARIZED PROBLEM
5.1. The linearized equations and the main result. We introduce the lin-
earized equations around a state given by a perturbation of the constant solution
in (2.9). More precisely, let us consider the functions
(5.1) U
r
=
_
_
_
v
r
0
_
_
_+

U
r
, U
l
=
_
_
_
v
r
0
_
_
_+

U
l
,
r
,
l
,
where ,

v
r
are xed positive constants (in this section we introduce the small
change of notation v
r

v
r
for the piecewise constant solution) and where
U
r
(t, x
1
, x
2
) =
_
_
_
r
(t, x
1
, x
2
)
v
r
(t, x
1
, x
2
)
u
r
(t, x
1
, x
2
)
_
_
_,

U
r
(t, x
1
, x
2
) =
_
_
_
r
(t, x
1
, x
2
)
v
r
(t, x
1
, x
2
)
u
r
(t, x
1
, x
2
)
_
_
_,
U
l
(t, x
1
, x
2
) =
_
_
_
l
(t, x
1
, x
2
)
v
l
(t, x
1
, x
2
)
u
l
(t, x
1
, x
2
)
_
_
_,

U
l
(t, x
1
, x
2
) =
_
_
_
l
(t, x
1
, x
2
)
v
l
(t, x
1
, x
2
)
u
l
(t, x
1
, x
2
)
_
_
_.
The functions
r
,
l
are perturbations of the change of variables. The index r
(resp. l) denotes the state on the right (resp. on the left) of the interface (after the
change of variables). We assume that
U
r
, U
l
,
r
,
l
W
2,
(), (5.2a)
|(U
r
, U
l
)|
W
2,
()
+|(
r
,
l
)|
W
2,
()
K
0
, (5.2b)
where K
0
is a suitable positive constant, and that the perturbations

U
r
,

U
l
have
compact support. These quantities are linked by the Rankine-Hugoniot condi-
tions and the continuity condition for the functions
r,l
that, written in the form
of (2.5), become
r
(t, x
1
, 0) =
l
(t, x
1
, 0) = (t, x
1
), (5.3a)
(v
r
v
l
)
x
2
=0

x
1
(u
r
u
l
)
x
2
=0
= 0, (5.3b)
t
+v
r
x
2
=0

x
1
u
r
x
2
=0
= 0, (5.3c)
(
r

l
)
x
2
=0
= 0. (5.3d)
The functions
r
and
l
should also satisfy
r
+v
r

x
1
r
u
r
= 0, (5.4a)
l
+v
l
x
1
l
u
l
= 0, (5.4b)
together with
(5.5)
x
2
r

0
,
x
2
l

0
,
for a suitable constant
0
> 0, in the whole closed half-space x
2
0.
Let us consider the families U
:
s
= U
r,l
+sU
:
,
:
s
=
r,l
+s
:
, where s is a
small parameter. We compute the linearized equations
L
(U
r,l
,
r,l
)(U
:
,
:
) :=
d
ds
L(U
:
s
,
:
s
)U
:
ss=0
= f
:
.
We obtain
(5.6)
t
U
+
+A
1
(U
r
)
x
1
U
+
+
1
x
2
r
(A
2
(U
r
)
t
r

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

x
2
+
(
x
2
r
)
2
(A
2
(U
r
)
t
r

x
1
r
A
1
(U
r
))
x
2
U
r
+
1
x
2
r
_
dA
2
(U
r
)U
+
x
1
+
A
1
(U
r
)
x
1
r
dA
1
(U
r
)U
+
_
x
2
U
r
= f
+
,
in the domain x
2
> 0, and a similar equation with U
, U
l
,
l
, f
instead
of U
+
,
+
, U
r
,
r
, f
+
. Recall that, according to the denition in (2.7), (2.8), the
rst row in (5.6) may be simply denoted by L(U
r
,
r
)U
+
, namely we set:
L(U
r
,
r
)U
+
:=
t
U
+
+A
1
(U
r
)
x
1
U
+
+
1
x
2
r
(A
2
(U
r
)
t
r

x
1
r
A
1
(U
r
))
x
2
U
+
.
The equation (5.6) and the corresponding one for (U
) may be simplied by
the introduction of la bonne inconnue (the good unknown) as in [1]:
(5.7)

U
+
:= U
+
x
2
x
2
U
r
,

U
:= U
x
2
x
2
U
l
.
A direct calculation shows that

U
+
,

U
satisfy
L(U
r
,
r
)

U
+
+C(U
r
, U
r
,
r
)

U
+
+

+
x
2
x
2
[L(U
r
,
r
)U
r
] = f
+
,
L(U
l
,
l
)

U
+C(U
l
, U
l
,
l
)

U
x
2
x
2
[L(U
l
,
l
)U
l
] = f
,
where
C(U
r
, U
r
,
r
)

U
+
:= [dA
1
(U
r
)

U
+
]
x
1
U
r
+
1
x
2
r
[dA
2
(U
r
)

U
+
x
1
r
dA
1
(U
r
)

U
+
]
x
2
U
r
,
and with a similar expression for C(U
l
, U
l
,
l
)

U
. In view of the results in

[1, 13], we neglect the zeroth order term in
+
,
in the linearized equations and

thus consider the linear equations
L
r

U
+
:= L(U
r
,
r
)

U
+
+C(U
r
, U
r
,
r
)

U
+
= f
+
, (5.8a)
L
l

U
:= L(U
l
,
l
)

U
+C(U
l
, U
l
,
l
)

U
= f
. (5.8b)
We easily verify, using (5.2), that the coecients of the operators L(U
r
,
r
) and
L(U
l
,
l
) are in W
2,
(), that is
A
1
(U
r
) W
2,
(),
1
x
2
r
(A
2
(U
r
)
t
r

x
1
r
A
1
(U
r
)) W
2,
(),
A
1
(U
l
) W
2,
(),
1
x
2
l
(A
2
(U
l
)
t
x
1
l
A
1
(U
l
)) W
2,
().
Moreover, we have C(U
r,l
, U
r,l
,
r,l
) W
1,
(). It is clear that the linearized
equations (5.8) forma symmetrizable hyperbolic system. For instance, a Friedrichs
symmetrizer for the operator L
r
is given by
S
r
:=
_
_
_
_
(p
(
r
)/
r
) 0 0
0
r
0
0 0
r
_
_
_
_
.
Using (5.4), we compute
S
r
x
2
r
(A
2
(U
r
)
t
r

x
1
r
A
1
(U
r
))
=
1
x
2
r
_
_
_
_
0 p
(
r
)
x
1
r
p
(
r
)
p
(
r
)
x
1
r
0 0
p
(
r
) 0 0
_
_
_
_
,
and we thus expect to control the traces of

U
+,1
and (

U
+,3

x
1
r

U
+,2
) on the
boundary x
2
= 0. These considerations motivate the introduction of the fol-
lowing operator:
(5.9) P
U
:
x
2
=0
:=
_

U
:,1
U
:,3
x
1

U
:,2
_
x
2
=0
.
We now turn to the linearized boundary conditions. The linearization of (2.5)
gives
+
x
2
=0
=
x
2
=0
= ,
(v
r
v
l
)
x
1
+(v
+
v
)
x
1
(u
+
u
) = g
1
,
t
+v
r

x
1
+v
+
x
1
u
+
= g
2
,
= g
3
,
on the boundary x
2
= 0. Let us introduce the vector b
0
= (0, 1, 0)
T
and the
matrices
b(t, x
1
) :=
_
_
_
0 (v
r
v
l
)
x
2
=0
1 v
r x
2
=0
0 0
_
_
_,
M(t, x
1
) :=
_
_
_
0
x
1
1 0
x
1
1
0
x
1
1 0 0 0
1 0 0 1 0 0
_
_
_.
Let us also denote U = (U
+
, U
)
T
, = (
t
,
x
1
)
T
and g = (g
1
, g
2
, g
3
)
T
.
Then the boundary conditions equivalently read
+
x
2
=0
=
x
2
=0
= , b+MU
x
2
=0
= g.
In terms of the good unknown

U dened by (5.7), the linearized boundary condi-
tions read
+
x
2
=0
=
x
2
=0
= , (5.10a)
B
(

U, ) := b+M
_
_
x
2
U
r
/
x
2
x
2
U
l
/
x
2
l
_
_
. , .
b
+M

U
x
2
=0
= g. (5.10b)
We observe that the linearized boundary conditions only involve the traces of P
U
+
and P
, with P dened by (5.9). With this notation, we can state our main result
(the norms are the weighted norms dened in Section 3).
Theorem 5.1. Assume that the particular solution dened by (5.1) satises
(5.11)

v
r
>
_
2c(),
and that the perturbations

U
r,l
,
r,l
have compact support and are small enough
in W
2,
(). Then there exist some constants C
1
and
1
1, that only depend on
K
0
and
0
(dened in (5.2), (5.5)), such that for all
1
and for all (

U, )
H
2
() H
2
(R
2
) the following estimate holds:
(5.12)
2
L
2
()
+
_
_
P
x
2
=0
_
_
2
L
2
(R
2
)
+
_
_
_
_
2
H
1
(R
2
)
C
1
_
1
2
L
2
(H
1
)
+
1
2
_
_
B
(

U, )
_
_
2
H
1
(R
2
)
_
.
The linearized operators L
and B
are dened in (5.8) and (5.10).

The remaining part of this section is devoted to the proof of Theorem 5.1.
5.2. Some preliminary transformation. Let us consider again the linearized
equations (5.8). After multiplication by the Friedrichs symmetrizer dened above
and a straightforward integration by parts, we easily prove the following lemma.
Lemma 5.2. There exist two constants C > 0 and
0
1 such that for all

0
, the following estimate holds:

U
+
2
L
2
()

C
r

U
+
2
L
2
()
+C
_
_
P
U
+
x
2
=0
_
_
2
L
2
(R
2
)
,
where the operator P is dened in (5.9). The estimate for

U
is the same, namely:
2
L
2
()

C
l

U
2
L
2
()
+C
_
_
P
x
2
=0
_
_
2
L
2
(R
2
)
.
As was done in the constant coecients case, it remains to show an estimate
of the traces P
U
:
x
2
=0
and the front function in terms of the source terms in
the interior domain and on the boundary. In order to prove such an estimate,
it is convenient to transform further the interior equations (5.8) in order to deal
with a problem with a constant and diagonal boundary matrix (i.e., the matrix
coecient of
x
2
in the dierential operators L
r,l
). This is possible because the
boundary matrix has constant rank in the whole closed half-space. Namely, let us
consider the coecients of
x
2
U
:
in (5.8). The coecients are equal to
1
x
2
(A
2
(U)
t

x
1
A
1
(U)),
where we forget for the moment the indices r, l. Under the assumption (5.4), this
coecient reduces to the matrix
A
2
(U, ) =
1
x
2
_
_
_
_
_
0
x
1

_
p
()/
_
x
1
0 0
_
p
()/
_
0 0
_
_
_
_
_
which has eigenvalues
1
= 0,
2,3
= :
c()
x
1
)
x
2
.
Here we have introduced the notation
x
1
) =
_
1 +(
x
1
)
2
. In order to di-
agonalize the above matrix we compute the eigenvectors associated to the above
eigenvalues. We obtain respectively the vectors
_
0 1
x
1
_
T
,
_
x
1
)
c()

x
1
c()
_
T
,
_
x
1
)
c()

x
1

c()
_
T
.
Observe that these eigenvectors are not orthonormal (because A
2
is not symmet-
ric). Thus, we may dene the following (non orthogonal) matrix
T(U, ) :=
_
_
_
_
_
_
_
_
0
x
1
)
x
1
)
1
c()

x
1
c()

x
1
x
1
c()

c()
_
_
_
_
_
_
_
_
,
which permits to diagonalize the above matrix A
2
(U, ):
T
1
(U, )A
2
(U, )T(U, ) =
_
_
_
0 0 0
0
2
0
0 0
3
_
_
_.
In order to obtain a constant boundary matrix in the dierential operators, we also
introduce the matrix
A
0
(U, ) :=
_
_
_
1 0 0
0
1
2
0
0 0
1
3
_
_
_ =
_
_
_
_
_
_
_
_
_
1 0 0
0

x
2
c()
x
1
)
0
0 0

x
2
c()
x
1
)
_
_
_
_
_
_
_
_
_
.
It follows that A
0
T
1
A
2
T = I
2
:= diag(0, 1, 1). Let us dene the new unknown
functions W
+
:= T
1
(U
r
,
r
)

U
+
, W
:= T
1
(U
l
,
l
)

U
, and set
T
r,l
:= T(U
r,l
,
r,l
), A
r,l
0
:= A
0
(U
r,l
,
r,l
).
After multiplication on the left side of the equations in (5.8) by A
r,l
0
T
1
r,l
, we see
that W
:
solve the equations
A
r
0

t
W
+
+A
r
1

x
1
W
+
+I
2
x
2
W
+
+A
r
0
C
r
W
+
= F
+
, (5.13a)
A
l
0
t
W
+A
l
1
x
1
W
+I
2
x
2
W
+A
l
0
C
l
W
= F
, (5.13b)
where we have set (with slight abuse of notation)
A
r,l
1
:= A
r,l
0
T
1
A
1
T(U
r,l
,
r,l
),
C
r,l
:= [T
1
t
T +T
1
A
1
x
1
T +T
1
A
x
2
T +T
1
CT](U
r,l
, U
r,l
,
r,l
),
F
:
= A
r,l
0
T
1
r,l
f
:
.
The above equations (5.13) are equivalent to the linearized equations (5.8). Intro-
ducing

W
:
:= e
t
W
:
, the equations (5.13) become equivalent to
1
W
+
:= A
r
0
W
+
+A
r
0

t
W
+
(5.14a)
+ A
r
1

x
1
W
+
+I
2
x
2
W
+
+A
r
0
C
r
W
+
= e
t
F
+
,
1
l

W
:= A
l
0
+A
l
0
(5.14b)
+ A
l
1
x
1
+I
2
x
2
+A
l
0
C
l
= e
t
F
.
Recall that we have A
r,l
j
W
2,
(), and C
r,l
W
1,
(). Using the vector W =
(W
+
, W
)
T
as dened above, the boundary conditions (5.10) become equivalent
to
+
x
2
=0
=
x
2
=0
= , (5.15a)
b+
b+M
_
T
r
0
0 T
l
_
W
x
2
=0
= g. (5.15b)
Introducing

W
:
and
`
:
:= e
t
:
,
`
:= e
t
, the equations (5.15) are also
equivalent to
`
+
=
`
=
`
, (5.16a)
1
W,
`
) := b
0
`
+b
`
+
b
`
+M
_
T
r
0
0 T
l
_
x
2
=0
= e
t
g. (5.16b)
From (5.2) we have
b W
2,
(R
2
),

b W
1,
(R
2
),
M W
2,
(R
2
), T
r,l
x
2
=0
W
2,
(R
2
).
The next step is to look for an a priori estimate of the solution to the (weighted)
linearized problem (5.14), (5.16). In view of Lemma 5.2, we are looking for an
estimate of P
U
+
and P
. Of course, we shall derive this estimate using the new

function W. The reader should keep in mind the relations
P
U
+
x
2
=0
=
_
_
_
x
1
)(W
+
2
+W
+
3
)
x
2
=0
c
r
x
1
)
2
(W
+
2
W
+
3
)
x
2
=0
_
_
_,
P
x
2
=0
=
_
_
_
x
1
)(W
2
+W
3
)
x
2
=0
c
l
x
1
)
2
(W
2
W
3
)
x
2
=0
_
_
_,
from which we easily deduce the estimate
(5.17) |P
U
+
x
2
=0
|
L
2
(R
2
)
+|P
x
2
=0
|
L
2
(R
2
)
C(|(W
+
2
, W
+
3
)
x
2
=0
|
L
2
(R
2
)
+|(W
2
, W
3
)
x
2
=0
|
L
2
(R
2
)
).
We are thus led to estimating the trace of the vector (
W
+
2
,
W
+
3
,
2
,
3
), when
W is a solution to the (weighted) linearized equations (5.14), (5.16). From now

on, for the sake of simplicity, we drop the tildas and write W
:
,
:
, instead of
W
:
,
`
:
,
`
. Observe that
:
, are coupled to W
:
only through the boundary
conditions.
5.3. Paralinearization. We refer to Appendix B for the denition of parad-
ierential symbols and operators, where the reader will also nd the main results
on paralinearization and symbolic calculus. We recall that the Fourier dual vari-
ables of (t, x
1
) are (, ), and that we always denote = +i the Laplace dual
variable of t. Recall we have introduced the positive constants K
0
,
0
in (5.2),
(5.5). We now turn to the paralinearization of the linearized equations.
(1) The boundary conditions. Dene the following symbols:
b
0
:=
_
_
_
0
1
0
_
_
_,
b
1
(t, x
1
) :=
_
_
_
v
r
v
l
v
r
0
_
_
_(t, x
1
, 0),
b(t, x
1
, , , ) := b
0
+ib
1
(t, x
1
).
Because b
0
is constant, we have
b
0
+b
0
t
= T
b
0
.
The main paralinearization estimate (Theorem B.9) yields
|b
1
x
1
T
ib
1
|
1,
C|b
1
|
W
2,
(R
2
)
||
0

C(K
0
)
||
1,
.
We now easily obtain
(5.18) |b
0
+b
0
t
+b
1
x
1
T
b
|
1,

C(K
0
)
||
1,
.
We also have the following inequalities:
|
bT
b
|
1,
C|
b|
W
1,
(R
2
)
||
0

C(K
0
,
0
)
||
1,
, (5.19a)
|T
b
|
1,
C|
b|
L
(R
2
)
||
1,
C(K
0
,
0
)||
1,
, (5.19b)
where

b is dened by (5.10). Eventually, we dene the symbol
M(t, x
1
) := M(t, x
1
, 0)
_
T
r
0
0 T
l
_
(t, x
1
, 0),
with the matrices M, T
r
, T
l
dened above. Recall that the state around which the
equations are linearized satises
r
(t, x
1
, 0) =
l
(t, x
1
, 0) = (t, x
1
),
r
(t, x
1
, 0) =
l
(t, x
1
, 0).
A direct calculation yields
M=
_
_
_
_
_
_
_
0
c
r
x
1
)
2
c
r
x
1
)
2
0
c
l
x
1
)
2
c
l
x
1
)
2
0
c
r
x
1
)
2
c
r
x
1
)
2
0 0 0
0
x
1
)
x
1
) 0
x
1
)
x
1
)
_
_
_
_
_
_
_
.
Thus the matrix M only acts on the noncharacteristic part of the vector W =
(W
+
, W
), that is, W
nc
:= (W
+
2
, W
+
3
, W
2
, W
3
). Since M W
2,
(R
2
), we have
|MW
x
2
=0
T
M
W
x
2
=0
|
1,

C
|M|
W
2,
(R
2
)
|W
nc
x
2
=0
|
0
(5.20)
C(K
0
)
|W
nc
x
2
=0
|
0
.
Adding (5.18)(5.19)(5.20), we obtain the paralinearization estimate for the
boundary operator:
(5.21) |1
(W, )T
b
T
M
W
x
2
=0
|
1,
C(K
0
,
0
)
_
||
1,
+
1
|W
nc
x
2
=0
|
0
_
.
We recall that the boundary operator 1
is dened by (5.16). Observe that in the

paralinearized version of 1
, there is no more zeroth order term in .

(2) The interior equations. We rst estimate the paralinearization error for
xed x
2
, and then integrate with respect to x
2
. For instance, we have
A
r
0
W
+
T
A
r
0
W
+
2
1,
=
_
+
0
2
_
_
A
r
0
W
+
(, x
2
) T
A
r
0
W
+
(, x
2
)
_
_
2
1,
dx
2
C
_
+
0
_
_
A
r
0
(, x
2
)
_
_
2
W
2,
(R
2
)
_
_
W
+
(, x
2
)
_
_
2
0
dx
2
C
_
_
A
r
0
_
_
2
W
2,
()
W
+
2
0
C(K
0
)
W
+
2
0
.
In a completely similar way, we obtain the following estimates
A
r
0

t
W
+
T
iA
r
0
W
+
1,
C(K
0
)W
+
0
,
A
r
1

x
1
W
+
T
iA
r
1
W
+
1,
C(K
0
)W
+
0
,
A
r
0
C
r
W
+
T
A
r
0
C
r
W
+
1,
C(K
0
,
0
)W
+
0
.
Adding these inequalities, we end up with the paralinearization estimate for the
interior equations:
1
r
W
+
T
A
r
0
+iA
r
1
+A
r
0
C
r
W
+
I
2
x
2
W
+
1,
(5.22)
C(K
0
,
0
)W
+
0
,
where the linearized operator 1
r
is dened by (5.14). The estimate for the equa-
tion on W
is identical:
1
l
W
A
l
0
+iA
l
1
+A
l
0
C
l
W
I
2
x
2
W
1,
(5.23)
C(K
0
,
0
)W
0
.
(3) Eliminating the front. We proceed as in the constant coecients case, and
show how to eliminate the front in the (paralinearized) boundary conditions. If
the perturbation is small enough (in the L
norm), there exists a constant c > 0

(depending only on K
0
) such that
b(t, x
1
, , , )
2
c(
2
+
2
+
2
).
Applying G ardings inequality (Theorem B.7), we obtain
TT
b
, )
L
2
(R
2
)

c
2
_
_
_
_
2
1,
,
for all
0
(where
0
only depends on K
0
). Using the rules of symbolic cal-
culus (Theorem B.6), we have T
b
= (T
b
)
b
+R
, where R
is of order 1.
Consequently, we have an estimate of the form
||
1,
C(K
0
)|T
b
|
0
.
For all
0
, we thus obtain
||
1,
C(K
0
)(|T
b
+T
M
W
x
2
=0
|
0
+|W
nc
x
2
=0
|
0
) (5.24)
C(K
0
)
_
1
|T
b
+T
M
W
x
2
=0
|
1,
+|W
nc
x
2
=0
|
0
_
.
From (5.21) and (5.24) we deduce for
0
large enough (depending on K
0
)
the estimate
||
1,
C(K
0
)
_
1
|1
(W, )|
1,
+|W
nc
x
2
=0
|
0
_
,
which shows that it only remains to prove an estimate of W
nc
x
2
=0
in terms of the
source terms.
For all (, ) in the hemisphere , we dene the matrix
(t, x
1
, , , ) :=
_
0 0 1
+iv
r
(t, x
1
, 0) i(v
r
v
l
)(t, x
1
, 0) 0
_
,
and we extend as a homogeneous mapping of degree 0 with respect to (, ).
We have b 0, and
0
2
. Applying Theorem B.6, we thus obtain:
|T
b
|
1,
= |T
b
T
b
|
1,
C(K
0
)||
1,
,
|T
M
W
x
2
=0
T
M
W
x
2
=0
|
1,
C(K
0
)|W
nc
x
2
=0
|
0
.
Using the decomposition
T
M
W
x
2
=0
= (T
M
T
M
)W
x
2
=0
+T
(T
M
W
x
2
=0
+T
b
) T
b
,
we get the following estimate
|T
M
W
x
2
=0
|
1,
(5.25)
C(K
0
)(|W
nc
x
2
=0
|
0
+|T
b
+T
M
W
x
2
=0
|
1,
+||
1,
).
As was done in the constant coecients case, we dene the symbol of the
reduced boundary conditions:
(t, x
1
, , , ) R
4
R
+
, (t, x
1
, , , ) := (t, x
1
, , , )M(t, x
1
).
We now focus on the paralinearized system with reduced boundary conditions:
(5.26)
_
_
T
A
r
0
+iA
r
1
+A
r
0
C
r
W
+
+I
2
x
2
W
+
=

F
+
, x
2
> 0,
T
A
l
0
+iA
l
1
+A
l
0
C
l
W
+I
2
x
2
W
=

F
, x
2
> 0,
T
x
2
=0
=

G, x
2
= 0.
Our aim is to prove an energy estimate for the paralinearized equations (5.26).
Once this is done, we shall obtain an energy estimate for the linearized equations.
More precisely, we have the following proposition.
Proposition 5.3. Assume that there exists a constant C
0
, depending only on K
0
and
0
, such that the solution W to (5.26) satises
(5.27)
_
_
W
nc
x
2
=0
_
_
2
0
C
0
_
1
2
1,
+
1
2
_
_
G
_
_
2
1,
_
,
for all
0
(where
0
only depends on K
0
and
0
). Then the thesis of Theorem 5.1
holds.
Proof. The proof is straightforward. We rst write
T
A
r
0
+iA
r
1
+A
r
0
C
r
W
+
+I
2
x
2
W
+
= 1
r
W
+
+error,
T
A
l
0
+iA
l
1
+A
l
0
C
l
W
+I
2
x
2
W
= 1
l
W
+error,
and estimate the error terms with the help of (5.22)(5.23). We use (5.27) to
derive
_
_
W
nc
x
2
=0
_
_
2
0
C
0
_
1
2
1,
+
1
2
0
+
1
2
_
_
T
x
2
=0
_
_
2
1,
_
,
where, as usual, 1
W = (1
r
W
+
, 1
l
W
). Using (5.24) and (5.25), and choosing

large enough, we obtain the following inequality:
_
_
W
nc
x
2
=0
_
_
2
0
+
_
_
_
_
2
1,
C
0
_
1
2
1,
+
1
2
0
+
1
2
_
_
T
b
+T
M
W
x
2
=0
_
_
2
1,
_
.
Eventually, we use (5.21) to derive (up to choosing large enough):
_
_
W
nc
x
2
=0
_
_
2
0
+
_
_
_
_
2
1,
C
0
_
1
2
1,
+
1
2
0
+
1
2
_
_
1
(W, )
_
_
2
1,
_
.
Then one uses the denitions
e
t

U
+
= T
r
W
+
, e
t

U
= T
l
W
,
e
t
A
r
0
T
1
r
L
r

U
+
= 1
r
W
+
, e
t
A
l
0
T
1
l
L
l

U
= 1
l
W
,
as well as (5.17) and Lemma 5.2 to derive (5.12). The reader will easily check that
the constants C
0
, C
0
etc. involved in the energy estimates only depend on K
0
and
0
.
Thanks to Proposition 5.3, we only need to prove the estimate (5.27) for the
paralinearized system (5.26). This will be done in the next paragraphs.
Recall that the boundary matrix in (5.26) only acts on W
nc
= (W
+
2
, W
+
3
, W
2
,
W
3
) and not on the full vector W. Namely, the rst and fourth columns of van-
ish. Consequently, we feel free to write the boundary conditions under the form
T
W
nc
x
2
=0
=

G, that is, we consider as a matrix with only four columns and two
rows.
5.4. Microlocalization. To derive the desired energy estimate for (5.26), we
follow the general strategy of the constant coecients case. Namely, we rst con-
sider the two equations that do not involve any x
2
derivative:
T
+iv
r
W
+
1
+T
ic
2
r
/
r
x
1
r
)
W
+
2
+ T
ic
2
r
/
r
x
1
r
)
W
+
3
+order 0 terms = F
+
1
,
T
+iv
l
1
+T
ic
2
l
/
l
x
1
l
)
W
2
+ T
ic
2
l
/
l
x
1
l
)
W
3
+ order 0 terms = F
1
.
Formally, the idea is to invert the operators T
+iv
r,l
and to substitute the cor-

responding value of W
:
1
into the four remaining equations. We shall thus get a
system of the form
_
_
_
x
2
W
nc
= T
A
W
nc
+T
E
W
nc
+source term, x
2
> 0,
T
W
nc
x
2
=0
= source term, x
2
= 0,
where A is of degree 1 and E is of degree 0. (Both matrices A and E are block
diagonal since the equations for W
+
and W
are decoupled). An important issue

is to show that this operation can be achieved. Namely, the zeroth order terms in
the two scalar equations above involve W
+
1
and W
1
. When inverting the opera-
tors T
+iv
r,l
, one needs to take the zeroth order terms into account, in order to
avoid introducing W
:
1
in the nal equation for W
nc
. We shall show that such an
inversion is possible. But in this paragraph, we focus on the rst order term and
explicit the symbol A. Consider the following 2 2 matrix:
(5.28) A
r
:=
_
_
A
r
1
A
r
3
A
r
3
A
r
2
_
_
,
with
(5.29a) A
r
1
:=
c
r
x
2
r
2( +iv
r
)
x
1
r
)
3
( +iv
r
)
x
2
r
c
r
x
1
r
)
+
x
2
r

x
1
r
i
x
1
r
)
2
,
(5.29b) A
r
2
:=
c
r
x
2
r
2( +iv
r
)
x
1
r
)
3
+
( +iv
r
)
x
2
r
c
r
x
1
r
)
+

x
2
r

x
1
r
i
x
1
r
)
2
,
(5.29c) A
r
3
:=
c
r
x
2
r
2( +iv
r
)
x
1
r
)
3
.
The denition of A
l
is completely similar, changing the r index by l. The symbol
A mentionned above is nothing but the block diagonal matrix
(5.30) A :=
_
A
r
0
0 A
l
_
.
The set of poles of A is denoted by
p
, that is,
p
:= (t, x
1
, x
2
, , )

such that = i v
r,l
(t, x
1
, x
2
).
As was done in the constant coecients case, we denote by 1
(t, x
1
, x
2
, , ) the
stable subspace of A(t, x
1
, x
2
, , ). This stable subspace is well dened when
T > 0, and admits a continuous extension up to any (, ) such that iR
and (, ) (0, 0).
At each point (t, x
1
, 0) of the boundary , the subspace
Z 1
(t, x
1
, 0, , ) s.t. (t, x
1
, , )Z = 0
is nontrivial (that is, not reduced to 0) if and only if
= i
v
r
(t, x
1
, 0) +v
l
(t, x
1
, 0)
2
or = iV
1
(t, x
1
) or = iV
2
(t, x
1
),
for suitable functions V
1,2
W
2,
(R
2
). This is just because when one freezes
the coecients on the boundary and computes the associated Lopatinskii deter-
minant, the calculations yield the same result as in Section 4. Recall that the
state (U
r,l
,
r,l
) around which the equations are linearized satisfy the Rankine-
Hugoniot conditions, see (5.3). As in [11], we dene the critical set of space-
frequency variables in the following way:
0
c
:=
_
(t, x
1
, , )
s.t.
_
i
(v
r
+v
l
)(t, x
1
, 0)
2
, iV
1
(t, x
1
), iV
2
(t, x
1
)
__
.
This is exactly the set of space variables on the boundary and frequencies so
that the Lopatinskii determinant vanishes. Moreover, if the perturbation (

U
r,l
,
r,l
)
is suciently small (in the L
norm), we have
v
l
(t, x
1
, 0) > V
1
(t, x
1
) >
(v
r
+v
l
)(t, x
1
, 0)
2
> V
2
(t, x
1
) > v
r
(t, x
1
, 0),
(t, x
1
) .
Therefore, the critical set
0
c
does not intersect the set of poles on the boundary of
the space domain:
0
c
(
p
x
2
= 0) = .
Another important feature of the critical set
0
c
is that it admits a neighborhood in
which the symbol A is diagonalizable. To be more precise, there exists a neighbor-
hood `
0
c
of
0
c
in R
2
and a mapping Q
0
on `
0
c
(with values in the set of 44
invertible matrices and homogeneous of degree 0 with respect to (, )) such that
(5.31) Q
0
(z)A(z)Q
0
(z)
1
= diag(
r
(z),
+
r
(z),
l
(z),
+
l
(z)),
z = (t, x
1
, , ) `
0
c
,
where
r
(resp.
+
r
) is the eigenvalue with negative (resp. positive) real part of
A
r
when > 0. (The denition of
:
l
is similar). Note that the matrix Q
0
has
the same block diagonal structure as A, see (5.30). The symbol Q
0
belongs to the
class
0
2
(see Appendix B for the precise denition).
Since
0
c
does not intersect
p
x
2
= 0, we may assume that the neighbor-
hood `
0
c
does not intersect
p
x
2
= 0 either.
The key point in the derivation of an energy estimate is to understand how
the singularities at the boundary (that is, the set
0
c
) propagate in the interior do-
main. Following [11], we shall show that the singularities propagate along the
two bicharacteristic curves associated with the (real) symbols 0
r,l
, provided that
these curves do not reach the poles of A or the points where A stops being diago-
nalizable. These ideas motivate the following result.
Proposition 5.4. Assume that the perturbation (

U
r,l
,
r,l
) is small in W
2,
()
and has compact support. Then one can choose the neighborhood `
0
c
such that there
exists an open set `
c

satisfying the following properties:
`
c
x
2
= 0 = `
0
c
and `
c
p
= .
The symbol A dened by (5.28)(5.29)(5.30) is diagonalizable on the set `
c
. In
other words, (5.31) holds on all `
c
, and not only on the trace `
0
c
.
For all z = (t, x
1
, x
2
, , ) `
c
, one has
r
(z)
+
r
(z) and
l
(z)
+
l
(z).
The solutions of the hamiltonian system of ODEs
dt
dx
2
=
h
(t, x
1
, x
2
, , ), (5.32a)
dx
1
dx
2
=
h
(t, x
1
, x
2
, , ), (5.32b)
d
dx
2
=
h
t
(t, x
1
, x
2
, , ), (5.32c)
d
dx
2
=
h
x
1
(t, x
1
, x
2
, , ), (t, x
1
, , , )
x
2
=0
`
0
c
, (5.32d)
are dened for all x
2
0 and remain in `
c
, both for h = 0
r
and h = 0
l
.
These solutions are referred to as bicharacteristic curves.
Following [11], we now construct a (real) weight that vanishes on the bichar-
acteristic curves, and that satises a linear transport equation. For all z =
(t, x
1
, , ) R
2
, dene
(5.33) (z) :=
_
+
(v
r
+v
l
)(t, x
1
, 0)
2
_
( V
1
(t, x
1
))( V
2
(t, x
1
)),
and extend to the whole set R
2
as a homogeneous mapping of degree 1 with
respect to (, ). The velocities V
1,2
are those dened above, and correspond to
the critical speeds for which the Lopatinskii determinant vanishes. The symbol
thus belongs to the class
1
2
. It is straightforward to check that
0
c
= z = (t, x
1
, , ) R
2
s.t. +i(z) = 0.
Using Proposition 5.4, it is possible to construct solutions
r,l
of the linear
transport equations
x
2
r
+
r
, 0
r
= 0, (5.34a)
x
2
l
+
l
, 0
l
= 0, (5.34b)
r
x
2
=0
=
l
x
2
=0
= , (5.34c)
where a, b stands for the Poisson bracket:
a, b :=
a
b
t
+
a
b
x
1
a
t
b

a
x
1
b
.
As a matter of fact, both
r
and
l
are well-dened in the neighborhood of the
bicharacteristic curves starting from the critical set
0
c
. This is because
r,l
are
constant along the bicharacteristic curves dened by (5.32), provided these curves
are globally dened! Shrinking `
0
c
and `
c
, if necessary, we may assume that
r
and
l
are dened in the whole open set `
c
. The key point is that
r
vanishes
on the bicharacteristic curve originating from
0
c
and associated with the symbol
0
r
. (A similar result holds for
l
). Far from these bicharacteristic curves, both
r
and
l
are bounded from below.
Up to now, the symbols Q
0
,
r
,
l
are only dened microlocally, that is,
locally in the frequency space. To circumvent this diculty, we now introduce
cut-o functions. We x, once and for all, two nonnegative cut-o functions
(with values in [0, 1])
c
and
p
such that

c
and
p
are smooth, that is, C
and homogeneous of degree 0 with respect

to (, ). They thus belong to the class
0
k
for any integer k.
The support of
c
is contained in the open set `
c
, and
c
1 in a neighbor-
hood of the bicharacteristic curves originating from
0
c
.
The support of
p
does not intersect the support of
c
, that is,
c
p
0.
Moreover,
p
1 in a neighborhood of the poles
p
.
Eventually, we dene
u
:= 1
c
p
and observe that
u
is supported far from
the bicharacteristic curves and far fromthe poles. As a consequence, the support of
u
on the boundary only consists of points for which the uniform Lopatinskii
condition holds. We shall therefore be able to use standard Kreiss symmetrizers
(as constructed in Section 4) to derive an energy estimate for T
u
W.
The end of this section is devoted to the proof of the a priori energy estimate
(5.27). We thus x W = (W
+
, W
) H
2
() and dene the source terms
F
+
:= T
A
r
0
+iA
r
1
+A
r
0
C
r
W
+
+I
2
x
2
W
+
H
1
(), (5.35a)
F
:= T
A
l
0
+iA
l
1
+A
l
0
C
l
W
+I
2
x
2
W
H
1
(), (5.35b)
G := T
W
nc
x
2
=0
H
3/2
(R
2
). (5.35c)
We rst show how to estimate the trace of T
c
W
nc
; then we show how to esti-
mate the trace of T
u
W
nc
, using Kreiss symmetrizers. Eventually, we show how
to estimate the trace of T
p
W
nc
. In the rst two cases, the rst step in the analysis
consists in deriving an equation that only involves W
nc
, that is, in eliminating W
:
1
in the paradierential equations (5.26). Once we have derived this noncharacteris-
tic equation, we apply the strategy of [11]. At the very end of the proof, we show
how to absorb the microlocalization errors.
5.5. Derivation of energy estimates: The bad frequencies. We dene
W
+
c
:= T
c
W
+
,
and compute the equation satised by W
+
c
. Starting from (5.35a), we obtain
I
2
x
2
W
+
c
= I
2
T
x
2
c
W
+
+T
c
F
+
T
c
(T
A
r
0
+iA
r
1
+A
r
0
C
r
W
+
).
Then we apply the rules of symbolic calculus (Theorem B.6) to get
(5.36) T
A
r
0
+iA
r
1
W
+
c
+T
A
r
0
C
r
W
+
c
+T
r
W
+
+I
2
x
2
W
+
c
= T
c
F
+
+R
1
W
+
,
where R
1
is an operator of order 1, and r is a symbol in the class
0
1
that
vanishes in a neighborhood of the bicharacteristic curves. Namely, r is dened by
the following formula:
r :=
1
i
c
, A
r
0
+iA
r
1

x
2
c
I
2
,
and is thus a linear combination of derivatives of
c
. Therefore, r is supported far
from the bicharacterstic curves originating from the critical set.
To avoid overloading the paper with unuseful notations, we shall denote by
m
a generic symbol in the class
m
1
, that may vary from line to line, or within
the same line, and whose exact expression is not useful. Moreover, we denote by
r any symbol in
0
1
that vanishes in a neighborhood of the bicharacteristic curves.
The notation R
m
is also used to denote a generic operator of order m. At last,
we denote the components of the vectors W
+
c
, W
+
in the following way:
W
+
c
:= (w
+
1
, w
+
2
, w
+
3
)
T
, W
+
:= (W
+
1
, W
+
2
, W
+
3
)
T
.
Some tedious computations lead to
A
r
1
=
_
_
_
_
_
_
_
_
_
_
v
r
c
2
r
x
1
r
)
c
2
r
x
1
r
)
r

x
2
r
2c
r
x
1
r
)
2
x
2
r
c
r
x
1
r
)
_
v
r

c
r

x
1
x
1
r
)
_
0
r

x
2
r
2c
r
x
1
r
)
2
0

x
2
r
c
r
x
1
r
)
_
v
r
+
c
r

x
1
x
1
r
)
_
_
_
_
_
_
_
_
_
_
_
and we also have
A
r
0
= diag
_
1,

x
2
r
c
r
x
1
r
)
,

x
2
r
c
r
x
1
r
)
_
.
The rst scalar equation in (5.36) thus reads
T
+iv
r
w
+
1
+T
ic
2
r
/
r
x
1
r
)
(w
+
2
+w
+
3
)
+
3
_
i=1
T
0
w
+
i
+T
r
W
+
= T
c
F
+
1
+R
1
W
+
.
Since the support of
c
is included in the open set `
c
(and does not intersect the
poles
p
), we can choose two smooth cut-o functions
1
and
2
such that

1
1 on the support of
c
, and
2
1 on the support of
1
.

2
(and therefore
1
) is supported in `
c
.

1
and
2
are C
and homogeneous of degree 0 with respect to (, ).

It is clear that the properties of the cut-o function
2
imply
2
/(+iv
r
)
1
2
.
We apply the operator T
2
/(+iv
r
)
to the previous equality and, after repeated
applications of Theorem B.6, we obtain:
T
2
w
+
1
+T
2
ic
2
r
/(+iv
r
)
r
x
1
r
)
(w
+
2
+w
+
3
)
+
3
_
i=1
T
1
w
+
i
+T
r
2
/(+iv
r
)
W
+
= T
2
/(+iv
r
)
T
c
F
+
1
+R
2
W
+
.
Now, we note that
T
2
w
+
1
= T
2
T
c
W
+
1
= T
c
W
+
1
+R
2
W
+
1
= w
+
1
+R
2
W
+
1
,
and we are led to the following relation:
(5.37) w
+
1
= T
2
ic
2
r
/(+iv
r
)
r
x
1
r
)
(w
+
2
+w
+
3
) +
3
_
i=1
T
1
w
+
i
+ T
2
r/(+iv
r
)
W
+
+T
2
/(+iv
r
)
T
c
F
+
1
+R
2
W
+
.
It is important to note that there is a term in w
+
1
(of degree 1) in the right-hand
side of (5.37).
In the second equation of (5.36), w
+
1
appears both in a term of order 1, say
T
1
w
+
1
with
1

1
2
, and in a term of order 0, say T
0
w
+
1
with
0

0
1
. We
rst use the expression (5.37) of w
+
1
in the term T
1
w
+
1
. The second equation of
(5.36) thus reads
T
1
2
w
+
2
+T
1
3
w
+
3
+
3
_
i=1
T
0
w
+
i
+T
r
W
+
+
x
2
w
+
2
= R
0
T
c
F
+
1
+T
c
F
+
2
+R
1
W
+
,
where
1
2,3

1
2
. We use once more the expression (5.37) in the term T
0
w
+
1
just above (recall that
0

0
1
so we can apply the rules of symbolic calculus).
Collecting the dierent terms, we are led to an equation that can be written under
the following form:
T
1
2
w
+
2
+T
1
3
w
+
3
+
3
_
i=2
T
0
w
+
i
+T
r
W
+
+
x
2
w
+
2
= R
0
T
c
F
+
1
+T
c
F
+
2
+R
1
W
+
.
In this equation, all the rst and zeroth order terms in w
+
1
have been eliminated.
Performing similar computations to eliminate w
+
1
in the last equation of (5.36),
we obtain a system of two equations that reads
(5.38)
x
2
_
_
w
+
2
w
+
3
_
_
= T
A
r
2
_
_
w
+
2
w
+
3
_
_
+T
E
r
_
_
w
+
2
w
+
3
_
_
+T
r
W
+
+R
0
F
+
+R
1
W
+
,
where E
r

0
1
, A
r
2

1
2
, and A
r
2
A
r
in the region
2
1. Recall that
the (singular) symbol A
r
is dened by (5.28)(5.29). Moreover, the symbol r in
(5.38) belongs to
0
1
and is identically zero in the region
c
1, and R
0
(resp.
R
1
) is an operator of order 0 (resp. 1).
We are now reduced to the noncharacteristic case, for which we follow the
analysis of [11]. Indeed, since the symbol A
r
2
equals A
r
in the region
2
1,
we can diagonalize A
r
2
in this region, and its eigenvalues are exactly
:
r
. More
precisely, in the region
2
1, we have the following relation:
Q
r
0
A
r
2
=
_
_
r
0
0
+
r
_
_
. , .
D
r
1
Q
r
0
.
Recall that on the set `
c
, and therefore also in the region
2
1, we have
r

+
r
thanks to Proposition 5.4. The following lemma can thus be proved as
in [11] (to avoid overloaded equations, we denote x
0
:= t,
0
:= , and
1
:=
the variables used in the tangential symbolic calculus).
Lemma 5.5. There exist a symbol Q
r
1

1
1
and a diagonal symbol D
r
0

0
1
,
that are dened in the region
2
1, such that
(Q
r
0
+Q
r
1
)(A
r
2
+E
r
) +
x
2
Q
r
0
+
1
i
1
_
j=0
(
j
Q
r
0

x
j
A
r
j
D
r
1

x
j
Q
r
0
) (D
r
1
+D
r
0
)(Q
r
0
+Q
r
1
)
is a symbol of degree 1 and regularity 1 (at least in the region
2
1).
In terms of symbolic calculus, Lemma 5.5 means nothing but
(Q
r
0
+Q
r
1
) (
x
2
A
r
2
E
r
) = (
x
2
D
r
1
D
r
0
) (Q
r
0
+Q
r
1
).
In other words, the change of basis (Q
r
0
+ Q
r
1
) diagonalizes both the rst order
term A
r
2
and the zeroth order term E
r
. We thus wish to prove an estimate for
Z
+
:= T
1
(Q
r
0
+Q
r
1
)
_
_
w
+
2
w
+
3
_
_
,
since this new vector will satisfy a paradierential equation with diagonal symbols.
Observe the role of the cut-o function
1
, whose support is contained in the
region
2
1, and that also satises
1
1 on the support of
c
. (We recall
that the vector (w
+
2
, w
+
3
) is microlocalized on the support of
c
).
Starting from (5.38) and using Lemma 5.5, as well as Theorem B.6, we com-
pute the equation satised by the vector Z
+
:
(5.39)
x
2
Z
+
= T
D
r
1
Z
+
+T
D
r
0
Z
+
+T
r
W
+
+R
0
F
+
+R
1
W
+
,
where

D
r
1
(resp.

D
r
0
) is an extension to the whole set of D
r
1
(resp. D
r
0
).
These extensions can be chosen such that
D
r
1
=
_
_
r
0
0
+
r
_
_
=
_
_
e
r
+ih
r
0
0 e
+
r
+ih
+
r
_
_
,
D
r
0
=
_
_
d
r
0
0 d
+
r
_
_
,
with e
r
, e
+
r

0
2
, h
r
, h
+
r

1
2
, and d
r
, d
+
r

0
1
. Moreover, the symbols e
r
, e
+
r
,
h
r
, h
+
r
are real valued and there exists a constant c > 0 such that
e
r
c < 0, e
+
r
c > 0.
The second equation in (5.39) reads
x
2
Z
+
2
= T
+
r
Z
+
2
+T
d
+
r
Z
+
2
+T
r
W
+
+R
0
F
+
+R
1
W
+
.
For this scalar equation, we choose
2,
:= Op(
2
+
2
0
+
2
1
) as a symmetrizer, that
is, we multiply the equation by
2,
Z
+
2
and integrate over . Some elementary
manipulations, whose details can be found in [11], yield the L
2
(H
1
) estimate of
Z
+
2
:
(5.40)
Z
+
2
2
1,
+
_
_
Z
+
2
(0)
_
_
2
1,
F
+
2
1,
+
W
+
2
0
+
r
W
+
2
1,
_
.
The rst equation in (5.39) reads
(5.41)
x
2
Z
+
1
= T
r
Z
+
1
+T
r
Z
+
1
+T
r
W
+
+R
0
F
+
+R
1
W
+
.
We rst choose the identity as a symmetrizer, and derive the following L
2
estimate:
Z
+
1
2
0
C
2
_
_
Z
+
1
(0)
_
_
2
0
(5.42)
+
C
F
+
2
1,
+
W
+
2
0
+
r
W
+
2
1,
_
.
Recall that in the preceding paragraph, we have constructed a symbol
r
that
satises the transport equation:
_
x
2
r
+
r
, h
r
= 0, x
2
> 0,
r
x
2
=0
= ,
where h
r
is the imaginary part of the eigenvalue
r
. This symbol
r
is well-
dened in the open set `
c
, thanks to Proposition 5.4. Since we have extended the
eigenvalue
r
(and thus h
r
) to the whole set , we can also extend
r
to
. Of course, we do not change the value of
r
on `
c
, since the solution
r
to the transport equation is constant on the bicharacteristic curves (5.32). With
slight abuse of notations, we still denote
r
the extension of
r
to the whole set
. This extension belongs to the class
1
2
and
x
2
r

1
1
.
We now choose S := (T
r
)
r
as a symmetrizer for (5.41). (We reproduce
below the calculations of [11] since this is really the key point in the analysis).
Standard integration by parts yields rst of all:
(5.43)
_
_
T
Z
+
1
(0)
_
_
2
0
= T
_
+
0
(
x
2
S)Z
+
1
, Z
+
1
) dx
2
+2T
_
+
0
ST
r
Z
+
1
, Z
+
1
) dx
2
+2T
_
+
0
ST
r
Z
+
1
, Z
+
1
) dx
2
+2T
_
+
0
ST
r
W
+
, Z
+
1
) dx
2
+2T
_
+
0
SR
0
F
+
, Z
+
1
) dx
2
+2T
_
+
0
SR
1
W
+
, Z
+
1
) dx
2
,
where the notation a, b) stands for the tangential scalar product in L
2
:
a, b) :=
_
R
2
a(t, x
1
)b(t, x
1
) dt dx
1
.
The three last integrals on the right hand-side of (5.43) are easily estimated using
Cauchy-Schwarz and Youngs inequalities ( is a positive number to be xed later
on):
2T
_
+
0
T
r
T
r
W
+
, T
r
Z
+
1
) dx
2

C
r
W
+
2
1,
+
r
Z
+
1
2
0
,
2T
_
+
0
T
r
R
0
F
+
, T
r
Z
+
1
) dx
2

C
F
+
2
1,
+
r
Z
+
1
2
0
,
2T
_
+
0
T
r
R
1
W
+
, T
r
Z
+
1
) dx
2

C
W
+
2
0
+
r
Z
+
1
2
0
.
The rules of symbolic calculus give T
r
T
r
= T
r
T
r
+R
0
, which yields the upper
bound
2T
_
+
0
ST
r
Z
+
1
, Z
+
1
) dx
2
C
r
Z
+
1
2
0
+CZ
+
1
0
T
r
Z
+
1
0
(C +)
r
Z
+
1
2
0
+
C
Z
+
1
2
0
(C +)
r
Z
+
1
2
0
+
C
W
+
2
0
.
Using the denition of S, we obtain
x
2
S = (T
x
2
r
)
r
+(T
r
)
x
2
r
.
From the basic estimates above, we already get the relation
(5.44) 2T
_
+
0
T
x
2
r
Z
+
1
+T
r
T
r
Z
+
1
, T
r
Z
+
1
) dx
2
_
_
T
Z
+
1
(0)
_
_
2
0
+(C +4)
r
Z
+
1
2
0
+
C
F
+
2
1,
+
r
W
+
2
1,
+
W
+
2
0
_
.
Now, we decompose
r
as
r
= e
r
+ ih
r
, where e
r
, h
r
take real values,
e
r

0
2
, and h
r

1
2
. Because the symbol
r
satises the transport equation
x
2
r
+
r
, h
r
= 0,
equation (5.44) yields
2T
_
+
0
T
r
T
r
Z
+
1
, T
r
Z
+
1
) dx
2
2T
_
+
0
T
i
r
,e
Z
+
1
, T
r
Z
+
1
) dx
2
2T
_
+
0
R
1
Z
+
1
, T
r
Z
+
1
) dx
2
2T
_
+
0
T
ih
r
T
r
Z
+
1
, T
r
Z
+
1
) dx
2
2T
_
+
0
R
0
Z
+
1
, T
r
Z
+
1
) dx
2
right-hand side of (5.44).
The rst term in the left-hand side is bounded from below thanks to G ardings
inequality (Theorem B.7). All the other terms are put on the right-hand side and
estimated using Cauchy-Schwarz and Youngs inequalities. In the end, we choose
an appropriate and obtain
r
Z
+
1
2
0
C
_
_
T
Z
+
1
(0)
_
_
2
0
+C
Z
+
1
2
0
+
C
F
+
2
1,
+
r
W
+
2
1,
+
W
+
2
0
_
.
We use (5.42) to estimate the term Z
+
1
2
0
in the right-hand side. Eventually,
we derive
r
Z
+
1
2
0
C
__
_
T
Z
+
1
(0)
_
_
2
0
+
_
_
Z
+
1
(0)
_
_
2
0
_
(5.45)
+
C
F
+
2
1,
+
r
W
+
2
1,
+
W
+
2
0
_
.
Recall that
r
is homogeneous of degree 1 with respect to the frequencies (, ),
so (5.45) has to be understood as a L
2
(H
1
) estimate of Z
+
1
far from the bicharac-
teristic curve (which is exactly the set where
r
vanishes).
A similar analysis enables us to derive an energy estimate for the vector
Z
:= T
1
(Q
l
0
+Q
l
1
)
T
c
_
_
W
2
W
3
_
_
,
where Q
l
0
diagonalizes the symbol A
l
2
:
Q
l
0
A
l
2
=
_
_
l
0
0
+
l
_
_
Q
l
0
,
and Q
l
1
is dened as in Lemma 5.5, mutatis mutandis. The nal estimates are
(5.46)
2
1,
+
_
_
Z
2
(0)
_
_
2
1,

C
2
1,
+
2
0
+
r
W
1,
2
_
,
2
0
+
l
Z
2
0
C
_
2
_
_
Z
1
(0)
_
_
2
0
+
_
_
T
1
(0)
_
_
2
0
_
+
C
2
1,
+
2
0
+
r
W
2
1,
_
.
The remaining part of the job is to estimate the traces of the incoming modes
(Z
+
1
and Z
1
) in terms of the outgoing modes (Z
+
2
and Z
2
) and G, knowing that
we have the relation
T
_
W
+
2
W
+
3
W
2
W
3
_
T
x
2
=0
= G.
This is nothing but the denition of G, see (5.35c). Observe that the rst col-
umn vector of Q
r
0
and the rst column vector of Q
l
0
span the stable subspace
1
(t, x
1
, x
2
, , ) (at least when the space-frequency variables belong to the set
`
c
). These column vectors are denoted by E
r
and E
l
. Following the proof of
Lemma 4.5 (see Appendix A), one can show that there exist some 2 2 invertible
matrices P
1
and P
2
such that
P
1
_
E
r
E
l
_
P
2
=
_
+i 0
0 1
_
.
Therefore, repeating the arguments of [11], one can show the following estimate
for the boundary terms:
(5.47)
2
__
_
Z
1
(0)
_
_
2
0
+
_
_
Z
+
1
(0)
_
_
2
0
_
+
_
_
T
1
(0)
_
_
2
0
+
_
_
T
Z
+
1
(0)
_
_
2
0
C
__
_
G
_
_
2
1,
+
_
_
Z
2
(0)
_
_
2
1,
+
_
_
Z
+
2
(0)
_
_
2
1,
+
_
_
W
nc
x
2
=0
_
_
2
0
_
.
Recall that the main ingredient of the proof is the microlocalized G ardings in-
equality (Theorem B.8), and the fact that is real valued.
Collecting (5.40), (5.42), (5.45), (5.46), and (5.47), we end up with the nal
estimate near the critical set:
(5.48)
_
2
1,
+
Z
+
2
2
1,
+
l
Z
2
0
+
r
Z
+
1
2
0
+
2
2
0
+
2
Z
+
1
2
0
_
+
__
_
Z
2
(0)
_
_
2
1,
+
_
_
Z
+
2
(0)
_
_
2
1,
+
_
_
T
1
(0)
_
_
2
0
+
_
_
T
Z
+
1
(0)
_
_
2
0
+
2
_
_
Z
1
(0)
_
_
2
0
+
2
_
_
Z
+
1
(0)
_
_
2
0
_
2
1,
+
2
0
+
r
W
2
1,
_
+
_
_
G
_
_
2
1,
+
_
_
W
nc
x
2
=0
_
_
2
0
.
Recall, for later use, that the vectors Z
:
are dened by the formulas
Z
+
:= T
1
(Q
r
0
+Q
r
1
)
T
c
_
_
W
+
2
W
+
3
_
_
, Z
:= T
1
(Q
l
0
+Q
l
1
)
T
c
_
_
W
2
W
3
_
_
,
and the matrices Q
r,l
0
are invertible on a neighborhood of the support of
1
. We
also recall the relation
1
c

c
.
Recall also that the components T
c
W
:
1
are given in terms of T
c
W
:
2,3
by the
relation (5.37). In particular, this relation yields an L
2
estimate for T
c
W
:
1
, and
an L
2
(H
1
) estimate far from the bicharacteristic curves. Namely, we can add the
norms
c
W
+
1
2
0
+
3
c
W
2
0
+
r
T
c
W
+
1
2
0
+
l
T
c
W
2
0
in the left-hand side of (5.48). We thus control the L
2
norm of the vector T
c
W,
and not only the noncharacteristic part of the vector. We also control the L
2
(H
1
)
norm far from the bicharacteristic curves that originate from the critical set.
5.6. Derivation of energy estimates: the good frequencies. To estimate the
trace of T
u
W
nc
, one rst computes the equation satised (in ) by the vector
T
u
W; then one eliminates the components that belong to the kernel of I
2
, and
obtains an equation involving only T
u
W
nc
. The equation is similar to (5.38).
For this reduced equation, one can construct Kreiss type symmetrizers, because
the uniform Lopatinskii condition is satised in a neighborhood of the support of
u
. The construction of the symmetrizer is achieved as in the constant coecients
case (see Section 4). Once again, we refer to [11] for a detailed derivation of energy
estimates, and we only give the result here. The estimate obtained by this method
reads:
(5.49)
u
W
2
1,
+
_
_
T
u
W
nc
(0)
_
_
2
1,
C
__
_
G
_
_
2
1,
+
_
_
W
nc
(0)
_
_
2
0
_
+
C
2
1,
+
2
0
+
r
W
2
1,
_
.
As in (5.48), the symbol r vanishes in a neighborhood of the bicharacteristic
curves (the symbol r in (5.49) is a linear combination of the derivatives of
u
,
and the cut-o function
u
is identically zero near the bicharacteristic curves).
5.7. Derivation of energy estimates: the poles. To derive an estimate for
T
p
W
:
, one starts from (5.35a)(5.35b) and computes an equation similar to
(5.36). Then one changes basis, as was done in the constant coecients case. In
the end, one derives a maximal L
2
(H
1
) estimate because the uniform Lopatinskii
condition is satised near the poles. The energy estimate is thus similar to the one
corresponding to the good frequencies:
(5.50)
p
W
2
1,
+
_
_
T
p
W
nc
(0)
_
_
2
1,
C
__
_
G
_
_
2
1,
+
_
_
W
nc
(0)
_
_
2
0
_
+
C
2
1,
+
2
0
+
r
W
2
1,
_
.
5.8. Proof of Theorem 5.1. We now patch together the microlocalized en-
ergy estimates, and show that the estimate (5.27) holds. We rst note that,
adding (5.48), (5.49), (5.50), we are able to control the norms
3
W
2
0
and
2
|W
nc
x
2
=0
|
2
0
. Namely, we rst obtain (up to choosing large enough):
Left-hand side of (5.48)(5.49)(5.50) (5.51)
C
_
_
G
_
_
2
1,
+
C
2
1,
+
r
W
2
1,
_
.
In view of (5.51), the only thing to show is how to absorb the term T
r
W
1,
.
Recall that the symbol r is identically zero in the regions where
c
,
u
or
p
are
equal to 1. We may thus decompose r as a linear combination of the form
r =
u
u
+
p
p
+
c
_
_
_
_
_
_
_
_
_
r
0 0
0
r
0 O
0 0
1,
l
0 0
O 0
l
0
0 0
1,
_
_
_
_
_
_
_
_
_
_
_
_
_
_
1 0
0
1
Q
r
0
1 0
0
1
Q
l
0
_
_
_
_
_
c
.
The matrices
c,u,p
have a block diagonal structure. We are thus able to absorb
the term T
r
W
1,
, thanks to the left-hand sides of (5.48)(5.49)(5.50). We
thus obtain (5.27), since the left-hand sides of (5.48)(5.49)(5.50) are bounded
from below by
c
_
2
0
+
2
_
_
W
nc
x
2
=0
_
_
2
0
_
, c > 0.
Thanks to Proposition 5.3, the estimate (5.12) for the variable coecients lin-
earized operators also holds. This completes the proof of Theorem 5.1.
6. CONCLUDING REMARKS
In this paper, we have proved a linear stability result for a wide class of rectilinear
compressible vortex sheets. To summarize, once we are given a suitable pertur-
bation of a rectilinear supersonic vortex sheet, the linearized coecients around
this perturbation satises an a priori estimate with loss of one derivative (in the
tangential variables). We have also proved that the paralinearized version of the
linearized equations satises the same a priori estimate. The constants appearing
in the energy estimates are uniform with respect to the W
2,
norm of the coe-
cients.
To prove the local in time existence of nonconstant vortex sheets, the next
step will be to build an iteration scheme that takes into account this loss of regu-
larity. In view of [1, 13], there is a strong hope that a Nash-Moser type iteration
scheme might answer the problem. However, special attention should be paid, at
each step, to the relations (5.3), (5.4), and (5.5), that are crucial in the proof of
Theorem 5.1. The verication of the local existence of (supersonic) vortex sheets
is postponed to a future work.
The one-dimensional stability of contact discontinuities has received a gen-
eral treatment in [8] and [9]. Unfortunately, the isentropic Euler equations do
not admit contact discontinuities in one space dimension. However, it would
be interesting to determine whether the present analysis extends to some contact
discontinuities for the general Euler equations, and see the connections with the
one-dimensional analysis.
Acknowledgements. The authors thank Sylvie Benzoni-Gavage and Denis
Serre for providing them with early versions of [4]. This has been very useful
to understand the symmetrizer construction near the poles. J.-F. C. thanks the
Mathematics Department of Brescia University for its hospitality during his post-
doctoral stay. Research of the authors was supported by the EU nanced network
HYKE, HPRN-CT-2002-00282.
APPENDIX A. PROOF OF INTERMEDIATE RESULTS
1.1. The proof of Proposition 4.3. Using (4.9) and (4.14), we obtain
_
E
r
E
l
_
=
_
( +iv
r
)(c
1
( +iv
r
)
r
) ( +iv
l
)(c
1
( +iv
l
)
l
)
c
r
( +iv
l
)(c
r
( +iv
r
)) c
l
( +iv
r
)(c
l
( +iv
l
))
_
for all (, ) . This gives the following expression for the Lopatinskii determi-
nant:
(, ) = c
2
( +iv
r
c
r
)( +iv
l
c
l
)(
r
2
)(
r
+
l
).
Recall that
r
and
l
have negative real part when has positive real part, and
satisfy the dispersion relations (4.13a)(4.13b). Because v
r
= v
l
, we have the
identity
r
(, ) =
l
(, ).
With the above expression for , it is easy to check that (, ) = (, ). We
shall thus only consider nonnegative values of in all this section: 0.
One rst checks that both expressions
( +iv
r
c
r
) and ( +iv
l
c
l
)
do not vanish for any (, ) , because of (4.13a)(4.13b).
Clearly, the sum
r
+
l
can not vanish when has positive real part, since
both numbers have negative real part. When is purely imaginary, one extends
r,l
by continuity. If = i iR satises ( +v
r
)
2
c
2
2
, one has
r
=
1
c
2
( +v
r
)
2
R.
If ( +v
r
)
2
> c
2
2
, we use Cauchy-Riemann relations to derive
r
= i sgn( +v
r
)
1
c
2
( +v
r
)
2
2
iR.
The calculations are almost the same as those done in [3]. For
l
, one just changes
v
r
into v
l
= v
r
and derives similar formulas. Then using the dispersion rela-
tions (4.13a)(4.13b), we easily check that
r
+
l
vanishes if and only if = 0
(and therefore 0). For > 0, this gives the following values for the eigen-
modes:
r
= i
v
2
r
c
2
1 =
l
iR.
Recall that v
r
> c
2.
It now remains to determine whether the expression (
r
2
) may vanish.
If = 0, one has
r
=
l
= /c, and, therefore,
r
l
0. We thus assume
0 (that is, > 0) and introduce the reduced expressions
V :=

i
,
r,l
:=

r,l
i
.
Assume that
r
l
= 1. Using (4.13) and (
r
l
)
2
= 1, we obtain the following
polynomial equation for V:
V
4
2(c
2
+v
2
r
)V
2
+v
2
r
(v
2
r
2c
2
) = 0.
This is a polynomial equation of degree 2 for the unknown V
2
, whose roots are
real and distinct. If (3.11) holds, both roots are positive. Let us denote these roots
by V
2
1
and V
2
2
, where 0 < V
1
< V
2
. One has
V
2
1
= c
2
+v
2
r
c
_
c
2
+4v
2
r
,
V
2
2
= c
2
+v
2
r
+c
_
c
2
+4v
2
r
.
We rst show that the root V
2
does not yield any instability. If V = V
2
, one
has V :v
r
c. Because > 0, we obtain
r
=
1
c
2
(V
2
+v
r
)
2
1 and
l
=
1
c
2
(V
2
v
r
)
2
1,
so
r
l
1. The Lopatinskii determinant does not vanish when V = V
2
. A
similar argument shows that does not vanish either for V = V
2
.
Now we show that V = V
1
is a root of the Lopatinskii determinant. One rst
checks that V
1
+v
r
> c, and V
1
v
r
< c. Hence, for > 0 and = iV
1
, we
nd
r
=
1
c
2
(V
1
+v
r
)
2
1 and
l
=
1
c
2
(V
1
v
r
)
2
1.
These relations yield
r
l
< 0 and (
r
l
)
2
= 1, that is, we have
r
l
= 1.
For V = V
1
, the Lopatinskii determinant vanishes. The same argument holds for
V = V
1
.
This completes the rst part of the proof of Proposition 4.3. What remains
to show is that the roots of the Lopatinskii determinant are simple. We rst show
that near = 0 and = 1/v
r
(this is the only point of such that = 0 and
> 0), we have
r
+
l
= h(, ),
for an appropriate C
function h. Since 0 near = 0, we have
r
+
l
= i(
r
+
l
),
where the notations are those introduced earlier. Near V = 0, both functions
r
and
l
are analytic with respect to V and satisfy
2
r
=
1
c
2
(V +v
r
)
2
1,
2
l
=
1
c
2
(V v
r
)
2
1.
We thus obtain
_
d
r
dV
+
d
l
dV
_
V=0
=
2v
r
c
r
(0)
0.
Using a classical factorization property of holomorphic functions, we obtain
r
+
l
= VH(V),
where H is holomorphic near 0 and H(0) 0. This yields
r
+
l
= H
_

i
_
.
The factorization result for is proved near the root = 0.
As regards the situation near those roots of the form (:iV
1
, ) , it is
entirely similar and we shall not detail the proof. (The proof is similar because
r
and
l
are still holomorphic with respect to V in a neighborhood of V
1
and V
1
).
The result is that the Lopatinskii determinant admits a factorization that reads
(, ) = ( iV
1
)h(, ) or (, ) = ( +iV
1
)h(, ),
where h is C
and does not vanish near the roots of . This completes the proof.
1.2. The proof of Lemma 4.5. In the proof of Proposition 4.3, we have seen
that the matrix (E
r
E
l
) has the following expression:
_
E
r
E
l
_
=
_
( +iv
r
)(c
1
( +iv
r
)
r
) ( +iv
l
)(c
1
( +iv
l
)
l
)
c
r
( +iv
l
)(c
r
( +iv
r
)) c
l
( +iv
r
)(c
l
( +iv
l
))
_
,
for all (, ) . We have also seen that the quantity ( + iv
r
c
r
) does
not vanish for (, ) . Let us now consider a neighborhood ` of a point
(
0
,
0
) such that
0
= 0. Up to shrinking `, the quantity ( +iv
r
) does
not vanish in `. As a consequence, the upper left corner coecient of (E
r
E
l
)
does not vanish in `. We write
_
E
r
E
l
_
=
_
1

2
3

4
_
.
Then, the relation (4.15) can be rewritten as =
1
4

2
3
, and
1
does not
vanish in `. The identity
_
1/
1
0
3
/
1
1
_
_
E
r
E
l
_
_
1
2
0
1
_
=
_
1 0
0
_
is a straightforward verication. In particular, this identity yields the estimate
_
E
r
E
l
_
Z
2
min(1,
2
)Z
2
,
for all Z
C
2
. Using Proposition 4.3, the Lopatinskii determinant can be
factorized near (
0
,
0
):
(, ) = h(, ), h(
0
,
0
) 0.
Since is the real part of , we obtain (, ) , for a suitable constant
> 0 (still up to shrinking the neighborhood `). This last inequality yields
_
E
r
E
l
_
Z
2

2
Z
2
,
for all Z
C
2
and all (, ) `.
Lemma 4.5 is thus proved when (
0
,
0
) satises
0
= 0. The other points
where vanishes are those points (
0
,
0
) such that
0
= :iV
1
0
. Near those
points, the upper left corner coecient of (E
r
E
l
) still does not vanish. We
can again conclude that in an appropriate neighborhood ` of (
0
,
0
) (with,
for instance,
0
= iV
1
0
), one has
_
E
r
E
l
_
Z
2
min(1,
2
)Z
2
.
Now we use the factorization
(, ) = ( iV
1
)h(, ), h(
0
,
0
) 0,
to conclude. This completes the proof of Lemma 4.5.
APPENDIX B. PARADIFFERENTIAL CALCULUS WITH A PARAMETER
In this appendix, we collect the main results of the paradierential calculus of
Bony and Meyer [5, 26] that we use in this paper, see [25] for the introduction of
the parameter. We refer to these papers for the proofs of the results stated below.
We rst recall the classication of paradierential symbols.
Denition B.1. A paradierential symbol of degree m R and regularity k
(k N) is a function a(x, , ) : R
2
R
2
[0, +[ C
NN
such that a is C
with respect to and for all N

2
, there exists a constant C
verifying
(, ), |
a(, , )|
W
k,
(R
2
)
C
m,
() = C
(
2
+
2
)
(m)/2
.
The set of paradierential symbols of degree m and regularity k is denoted by
m
k
.
We denote by
m
k
the subset of paradierential symbols a
m
k
such that for a
suitable ]0, 1[ one has
(, ), Supp
x
a(, , ) R
2
/ (
2
+
2
)
1/2
.
Of course, the symbols in
m
k
are C
functions with respect to both variables

x and , and for all a
m
k
, we have the estimates
(x, , ),
a(x, , ) C
,
m+,
().
Thus any symbol a
m
k
belongs to H ormanders class S
m
1,1
[16] and denes an
operator Op
(a) on the Schwartz class : by the usual formula

u :, Op
(a)u(x) :=
1
(2)
2
_
R
2
e
ix
a(x, , )
u() d.
We shall use the following terminology:
Denition B.2. A family of operators P
dened for 1 will be said of

order m (m R) if the operators P
are uniformly bounded from H

s+m
to
H
s
:
1, u H
s+m
, |P
u|
s,
C(s, m)|u|
s+m,
.
The following theorem is crucial for the sequel of the analysis.
Theorem B.3. If a
m
k
, the family Op
(a) is of order m.
The regularization of symbols in the class
m
k
is achieved by a convolution
with admissible cut-o functions.
Denition B.4. Let : R
2
R
2
[1, +[ [0, +[ be a C
function such
that the following estimates hold for all , N
2
:
(, , ),
(, , ) C
,
,
().
We shall say that is an admissible cut-o function if there exist real numbers
0 <
1
<
2
< 1 satisfying
(, , ) = 1 if
1
(
2
+
2
)
1/2
,
(, , ) = 0 if
2
(
2
+
2
)
1/2
.
An example of cut-o function is the following: let be a nonnegative C
function on R
2
R such that
2
1
+
1
2

2
2
+
2
2
(
1
,
1
) (
2
,
2
),
_
_
(, ) = 1 if (
2
+
2
)
1/2
1
2
,
(, ) = 0 if (
2
+
2
)
1/2
1.
We dene a function (, ) := (/2, /2) (, ). Then the function
0
dened by
0
(, , ) :=
_
p0
(2
2p
, 0)(2
p
, 2
p
)
is an admissible cut-o function (one can take
1
=
1
16
and
2
=
1
2
).
If is an admissible cut-o function, the inverse Fourier transform K
of
(, , ) satises
(, ), |
(, , )|
L
1
(R
2
)
C
,
().
These L
1
bounds for the derivatives
enable us to establish the following

proposition.
Proposition B.5. Let be an admissible cut-o function. The mapping
a
a
(x, , ) :=
_
R
2
K
(x y, , )a(y, , ) dy
is continuous from
m
k
to
m
k
for all m.
If a
m
1
, then a
a

m1
0
. In particular, if
1
and
2
are two admissible
cut-o functions and a
m
1
, then
1
a

2
a

m1
0
.
Fixing an admissible cut-o function , we dene the paradierential opera-
tor T
,
a
by the formula
T
,
a
:= Op
a
).
If
1
and
2
are two admissible cut-o functions and a
m
1
, then Proposition
B.5 and Theorem B.3 show that the family T
1
,
a
T
2
,
a
is of order (m1).
The symbolic calculus is based on the following theorem.
Theorem B.6.
Let a
m
1
and b
m
1
. Then ab
m+m
1
and the family
T
,
a
T
,
b
T
,
ab

1
is of order m+m
1 for all admissible cut-o function .

Let a
m
1
. Then for all admissible cut-o function , the family
(T
,
a
)
T
,
a

1
is of order m1.
Let a
m
2
and b
m
2
. Then ab
m+m
2
and the family
T
,
a
T
,
b
T
,
ab
T
,
i
j
a
x
j
b
1
is of order m+m
2 for all admissible cut-o function .

Let a
m
2
. Then the family
(T
,
a
)
T
,
a
T
,
i
j

x
j
a
1
is of order m2 for all admissible cut-o function .
The next theorem is the parameter version of G ardings inequality.
Theorem B.7. Let a
2m
1
and let be and admissible cut-o function. As-
sume that there exists a constant c > 0 such that
(x, , ), Ta(x, , ) c
2m,
()I.
Then there exists
0
1 such that

0
, u H
m
, TT
,
a
u, u)
H
m
,H
m
c
2
_
_
u
_
_
2
m,
.
We also have a microlocalized version of G ardings inequality.
Theorem B.8. Let a
2m
1
,
0
1
and be and admissible cut-o function.
Assume that there exists

0
1
and a constant c > 0 such that

0,

= , and
(x, , ),

2
(x, , )Ta(x, , ) c
2
(x, , )
2m,
()I.
Then there exists
0
1 and C > 0 such that
TT
,
a
T
,
u, T
,
u)
H
m
,H
m
c
2
_
_
T
,
u
_
_
2
m,
C
_
_
u
_
_
2
m1,
,

0
, u H
m
.
We now study the case of paraproducts: they are dened by the particular
choice of
0
as cut-o function. We shall write T
a
instead of T
0
,
a
for the parad-
ierential operators obtained after convolution by the function
0
. We have the
following important result.
Theorem B.9. Let a W
1,
(R
2
), u L
2
(R
2
), and 1. Then we have
|auT
a
u|
0

C
|a|
W
1,
(R
2
)
|u|
0
,
|a
x
j
uT
a

x
j
u|
0
C|a|
W
1,
(R
2
)
|u|
0
,
for a suitable constant C that is independent of (a, u, ).
If a W
2,
(R
2
), we have
|auT
a
u|
1,

C
|a|
W
2,
(R
2
)
|u|
0
,
|a
x
j
uT
a

x
j
u|
1,
C|a|
W
2,
(R
2
)
|u|
0
,
for a suitable constant C that is independent of (a, u, ).
We can extend the paradierential calculus to symbols dened on a half-space
in the following way: we still denote by
m
k
the set of symbols a(x
0
, x
1
, x
2
, , )
dened on (R
d
[0, +[ \ 0) such that the mapping x
2
a(, x
2
, ) is
bounded into
m
k
. We dene the paradierential operator T
a
by the formula
u C
c
(
), x
2
0, (T
a
u)(, x
2
) := T
a(x
2
)
u(, x
2
).
Using Theorem B.9 and integrating with respect to x
2
, we obtain for all symbols
a W
1,
() and all u L
2
() the estimates:
auT
a
u
0

C
|a|
W
1,
()
u
0
,
a
x
j
uT
a

x
j
u
0
C|a|
W
1,
()
u
0
, j = 0, 1.
REFERENCES
[1] S. ALINHAC, Existence dondes de rarefaction pour des systèmes quasi-lineaires hyperboliques mul-
tidimensionnels, Comm. Partial Dierential Equations 14 (1989), 173230. MR 90h:35147b
(French, with English summary)
[2] MIGUEL ARTOLA and ANDREW J. MAJDA, Nonlinear development of instabilities in supersonic
vortex sheets. I. The basic kink modes, Phys. D 28 (1987), 253281,
http://dx.doi.org/10.1016/0167-2789(87)90019-4. MR 88i:76025
[3] S. BENZONI-GAVAGE, Stability of multi-dimensional phase transitions in a van der Waals uid,
Nonlinear Anal. 31 (1998), 243263. MR 99b:76070
[4] S. BENZONI-GAVAGE and D. SERRE, First order systems of hyperbolic partial dierential equa-
tions with applications (In preparation).
[5] JEAN-MICHEL BONY, Calcul symbolique et propagation des singularites pour les equations aux
derivees partielles non lineaires, Ann. Sci.

Ecole Norm. Sup. (4) 14 (1981), 209246. MR
84h:35177 (French)
[6] JACQUES CHAZARAIN and ALAIN PIRIOU, Introduction to the theory of linear partial dieren-
tial equations, Studies in Mathematics and its Applications, vol. 14, North-Holland Publishing
Co., Amsterdam, 1982, ISBN 0-444-86452-0. MR 83j:35001
[7] JEAN-YVES CHEMIN, Fluides parfaits incompressibles, Ast erisque, 1995. MR 97d:76007
(French, with French summary)
[8] ANDREA CORLI, Asymptotic analysis of contact discontinuities, Ann. Mat. Pura Appl. (4) 173
(1997), 163202. MR 99f:35128
[9] ANDREA CORLI and MONIQUE SABL
E-TOUGERON, Stability of contact discontinuities un-

der perturbations of bounded variation, Rend. Sem. Mat. Univ. Padova 97 (1997), 3560. MR
98h:35149
[10] JEAN-FRANC OIS COULOMBEL, Weak stability of nonuniformly stable multidimensional shocks,
SIAM J. Math. Anal. 34 (2002), 142172 (electronic),
http://dx.doi.org/10.1137/S0036141001392803. MR 2004c:35274
[11] J.-F. COULOMBEL, Weakly stable multidimensional shocks, Annales de lIHP, Analyse Non
Lin eaire (to appear).
[12] J. A. FEJER and JOHN W. MILES, On the stability of a plane vortex sheet with respect to three-
dimensional disturbances, J. Fluid Mech. 15 (1963), 335336. MR 27 #4455
[13] JACQUES FRANCHETEAU and GUY M
ETIVIER, Existence de chocs faibles pour des systèmes

quasi-lineaires hyperboliques multidimensionnels, Ast erisque, 2000. MR 2002c:35183 (French)
[14] DAVID HATEAU, Instabilite des feuilles de tourbillon avec conduction de chaleur, C. R. Acad. Sci.
Paris S er. I Math. 330 (2000), 629633,
http://dx.doi.org/10.1016/S0764-4442(00)00230-5. MR 2001b:76039 (French)
[15] REUBEN HERSH, Mixed problems in several variables, J. Math. Mech. 12 (1963), 317334. MR
26 #5304
[16] LARS H
ORMANDER, Lectures on nonlinear hyperbolic dierential equations, Math ematiques &

Applications (Berlin) [Mathematics & Applications], vol. 26, Springer-Verlag, Berlin, 1997,
ISBN 3-540-62921-1. MR 98e:35103
[17] HEINZ-OTTO KREISS, Initial boundary value problems for hyperbolic systems, Comm. Pure Appl.
Math. 23 (1970), 277298. MR 55 #10862
[18] P. D. LAX, Hyperbolic systems of conservation laws. II , Comm. Pure Appl. Math. 10 (1957), 537
566. MR 20 #176
[19] P. D. LAX and R. S. PHILLIPS, Local boundary conditions for dissipative symmetric linear dieren-
tial operators, Comm. Pure Appl. Math. 13 (1960), 427455. MR 22 #9718
[20] ANDREW MAJDA, The Stability of Multidimensional Shock Fronts, Mem. Amer. Math. Soc.,
vol. 41, 1983. MR 84e:35100
[21] , The Existence of Multidimensional Shock Fronts, Mem. Amer. Math. Soc., vol. 43,
1983. MR 85f:35139
[22] ANDREW J. MAJDA and ANDREA L. BERTOZZI, Vorticity and incompressible ow, Cambridge
Texts in Applied Mathematics, vol. 27, Cambridge University Press, Cambridge, 2002, ISBN 0-
521-63057-6, 0-521-63948-4. MR 2003a:76002
[23] ANDREW MAJDA and STANLEY OSHER, Initial-boundary value problems for hyperbolic equa-
tions with uniformly characteristic boundary, Comm. Pure Appl. Math. 28 (1975), 607675. MR
53 #13857
[24] GUY M
ETIVIER, The block structure condition for symmetric hyperbolic systems, Bull. London
Math. Soc. 32 (2000), 689702,
http://dx.doi.org/10.1112/S0024609300007517. MR 2001i:35198
[25] , Stability of multidimensional shocks, Advances in the Theory of Shock Waves, Progr.
Nonlinear Dierential Equations Appl., vol. 47, Birkh auser Boston, Boston, MA, 2001, pp. 25
103. MR 2002m:35148
[26] YVES MEYER, Remarques sur un theorème de J.-M. Bony, Rend. Circ. Mat. Palermo (2) (1981),
120. MR 83b:35169 (French)
[27] JOHN W. MILES, On the disturbed motion of a plane vortex sheet, J. Fluid Mech. 4 (1958), 538
552. MR 20 #4394
[28] JAMES V. RALSTON, Note on a paper of Kreiss, Comm. Pure Appl. Math. 24 (1971), 759762.
MR 58 #29326
[29] DENIS SERRE, Systems of conservation laws. 1, Cambridge University Press, Cambridge, 1999,
ISBN 0-521-58233-4. MR 2000g:35142
[30] , Systems of conservation laws. 1, Cambridge University Press, Cambridge, 1999, ISBN
0-521-58233-4. MR 2000g:35142
JEAN-FRANC OIS COULOMBEL:
CNRS &Universit e de Lille 1
Laboratoire de Math ematiques, UMR 8524,
Cit e scientique
59655 Villeneuve DAscq Cedex, France
E-MAIL: jfcoulom@ano.univ-lille1.fr
PAOLO SECCHI:
Dipartimento di Matematica - Facolt` a di Ingegneria
Via Valotti 9, 25133 Brescia, Italy
E-MAIL: paolo.secchi@ing.unibs.it
2000 MATHEMATICS SUBJECT CLASSIFICATION: Author: Please supply.
KEY WORDS AND PHRASES: Author: Please supply.
Received: September 17th, 2003.

Jean-Francois Coulombel and Paolo Secchi - The Stability of Compressible Vortex Sheets in Two Space Dimensions

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Jean-Francois Coulombel and Paolo Secchi - The Stability of Compressible Vortex Sheets in Two Space Dimensions

Uploaded by

Copyright:

Available Formats

The Stability of Compressible Vortex Sheets

in Two Space Dimensions

denotes the jump of a quantity q across the interface (see [29]).

are smooth on the xed do-

should also satisfy

). One should always

), and obvious denitions for the operators L

is an operator that only acts on

() in an entirely similar way.

W 0. This rst reduction

mapping Q dened on such that Q has values

, homogeneous of degree 1, and has the additional property:

(, ) of (, ) has dimension 2, and is

(, ) is dened for all (, ) ,

function h dened on ` such that

is the classical adjoint matrix).

, real valued mappings b

real valued mapping that vanishes at (

of admits a continuous extension at

compact manifold, there exists a nite cov-

mappings on the whole

. In view of the results in

in the linearized equations and

are dened in (5.8) and (5.10).

is the same, namely:

. Of course, we shall derive this estimate using the new

W is a solution to the (weighted) linearized equations (5.14), (5.16). From now

is dened by (5.16). Observe that in the

, there is no more zeroth order term in .

norm), there exists a constant c > 0

). Using (5.24) and (5.25), and choosing

and to substitute the cor-

are decoupled). An important issue

and homogeneous of degree 0 with respect

and homogeneous of degree 0 with respect to (, ).

function h. Since 0 near = 0, we have

with respect to and for all N

functions with respect to both variables

(a) on the Schwartz class : by the usual formula

dened for 1 will be said of

are uniformly bounded from H

enable us to establish the following

1 for all admissible cut-o function .

2 for all admissible cut-o function .

E-TOUGERON, Stability of contact discontinuities un-

ETIVIER, Existence de chocs faibles pour des syst`emes

ORMANDER, Lectures on nonlinear hyperbolic dierential equations, Math ematiques &

You might also like