HAL Id: hal-01238617
https://hal.archives-ouvertes.fr/hal-01238617
Preprint submitted on 6 Dec 2015
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
The singular limit of the Water-Waves equations in the rigid lid regime
Benoît Mésognon-Gireau
To cite this version:
Benoît Mésognon-Gireau. The singular limit of the Water-Waves equations in the rigid lid regime.
2015. �hal-01238617�
The singular limit of the Water-Waves equations in the rigid lid regime
Mésognon-Gireau Benoît
∗Abstract
The re-scaled Water-Waves equations depend strongly on the ratio ε between the amplitude of the wave and the depth of the water. We investigate in this paper the convergence as ε goes to zero of the free surface Euler equations to the so called rigid lid model. We first prove that the only solutions of this model are zero. Due to the conservation of the Hamiltonian, the solutions of the free surface Euler equations converge weakly to zero, but not strongly in the general case, as ε goes to zero. We then study this default of convergence. More precisely, we show a strong convergence result of the solutions of the water waves equations in the Zakharov-Craig-Sulem formulation to the solutions of the linear water-waves equations. It is then easy to observe these latter converge weakly to zero. The simple structure of this system also allows us to explain the mechanisms of the weak convergence to zero. Finally, we show that this convergence to the rigid lid model also holds for the solutions of the Euler equations. To this end we give a new proof of the equivalence of the free surface Euler equations and of the Zakharov-Craig-Sulem equation by building an extension of the velocity and pressure fields.
1 Introduction
We recall here classical formulations of the Water Waves problem. We then shortly introduce the mean- ingful dimensionless parameters of this problem, and introduce the so called rigid lid equations. We then present the main result of this paper.
1.1 Formulations of the Water Waves problem
The Water Waves problem puts the motion of a fluid with a free surface into equations. We recall here two equivalent formulations of the Water Waves equations for an incompressible and irrotationnal fluid.
1.1.1 Free surface d-dimensional Euler equations
The motion, for an incompressible, inviscid and irrotationnal fluid occupying a domain Ω
tdelimited below by a flat bottom and above by a free surface is described by the following quantities:
– the velocity of the fluid U “ p V, w q , where V and w are respectively the horizontal and vertical components;
– the free top surface profile ζ;
∗UMR 8553 CNRS, Laboratoire de Mathématiques et Applications de l’Ecole Normale Supérieure, 75005 Paris, France.
Email: benoit.mesognon-gireau@ens.fr
– the pressure P.
All these functions depend on the time and space variables t and p X, z q P Ω
t. The domain of the fluid at the time t is given by
Ω
t“ tp X, z q P R
d`1, ´ H
0ă z ă ζ p t, X qu ,
where H
0is the typical depth of the water. The unknowns p U, ζ, P q are governed by the Euler equations:
$ ’
&
’ %
B
tU ` U ¨ ∇
X,zU “ ´
ρ1∇ P ´ ge
zin Ω
tdiv p U q “ 0 in Ω
tcurlp U q “ 0 in Ω
t.
(1.1)
We denote here ´ ge
zthe acceleration of gravity, where e
zis the unit vector in the vertical direction, and ρ the density of the fluid. Here, ∇
X,zdenotes the d ` 1 dimensional gradient with respect to both variables X and z.
These equations are completed by boundary conditions:
$ ’
&
’ %
B
tζ ` V ¨ ∇ ζ ´ w “ 0 U ¨ n “ 0 on t z “ ´ H
0u P “ P
atmon t z “ ζ p t, X qu .
(1.2)
In these equations, V and w are the horizontal and vertical components of the velocity evaluated at the surface. The vector n in the last equation stands for the normal upward vector at the bottom p X, z “ ´ H
0q . We denote P
atmthe constant pressure of the atmosphere at the surface of the fluid. The first equation of (1.2) states the assumption that the fluid particles do not cross the surface, while the last equation of (1.2) states the assumption that they do not cross the bottom. The equations (1.1) with boundary conditions (1.2) are commonly referred to as the free surface Euler equations.
1.1.2 Craig-Sulem-Zakharov formulation
Since the fluid is by hypothesis irrotational, it derives from a scalar potential:
U “ ∇
X,zΦ.
Zakharov remarked in [17] that the free surface profile ζ and the potential at the surface ψ “ Φ
|z“ζfully determine the motion of the fluid, and gave an Hamiltonian formulation of the problem. Later, Craig-Sulem, and Sulem ([8] and [9]) gave a formulation of the Water Waves equation involving the Dirichlet-Neumann operator. The following Hamiltonian system is equivalent (see [12] and [1] for more details) to the free surface Euler equations (1.1) and (1.2):
$ &
%
B
tζ ´ Gψ “ 0 B
tψ ` gζ ` 1
2 | ∇ ψ |
2´ p Gψ ` ∇ ζ ¨ ∇ ψ q
22 p 1 ` | ∇ ζ |
2q “ 0, (1.3)
where the unknowns are ζ (free surface profile) and ψ (velocity potential at the surface) with t as time variable and X P R
das space variable. The fixed bottom profile is b, and G stands for the Dirichlet- Neumann operator, that is
Gψ “ G r ζ s ψ “ a
1 ` | ∇ ζ |
2B
nΦ
|z“ζ,
where Φ stands for the potential, and solves Laplace equation with Neumann (at the bottom) and Dirichlet (at the surface) boundary conditions:
# ∆
X,zΦ “ 0 in tp X, z q P R
dˆ R , ´ H
0` b p X q ă z ă ζ p X qu
φ
|z“ζ“ ψ, B
nΦ
|z“´H0“ 0, (1.4)
with the notation, for the normal derivative
B
nΦ
|z“´H0“ ∇
X,zΦ p X, ´ H
0q ¨ n
where n stands for the normal upward vector at the bottom p X, ´ H
0q . See also [12] for more details.
1.1.3 Dimensionless equations
Since the properties of the solutions depend strongly on the characteristics of the fluid, it is more convenient to non-dimensionalize the equations by introducing some characteristic lengths of the wave motion:
(1) The characteristic water depth H
0;
(2) The characteristic horizontal scale L
xin the longitudinal direction;
(3) The characteristic horizontal scale L
yin the transverse direction (when d “ 2);
(4) The size of the free surface amplitude a
surf; Let us then introduce the dimensionless variables:
x
1“ x L
x, y
1“ y L
y, ζ
1“ ζ a
surf, z
1“ z H
0,
and the dimensionless variables:
t
1“ t t
0, Φ
1“ Φ Φ
0, where
t
0“ L
x? gH
0, Φ
0“ a
surfH
0L
xa gH
0.
After rescaling, several dimensionless parameters appear in the equation. They are a
surfH
0“ ε, H
02L
2x“ µ, L
xL
y“ γ,
where ε, µ, γ are commonly referred to respectively as "nonlinearity", "shallowness" and "transversality"
parameters.
For instance, the Zakharov-Craig-Sulem system (1.3) becomes (see [12] for more details) in dimen- sionless variables (we omit the "primes" for the sake of clarity):
$ ’
&
’ % B
tζ ´ 1
µ G
µ,γr εζ s ψ “ 0 B
tψ ` ζ ` ε
2 | ∇
γψ |
2´ ε µ
p G
µ,γr εζ s ψ ` εµ ∇
γζ ¨ ∇
γψ q
22 p 1 ` ε
2µ | ∇
γζ |
2q “ 0,
(1.5)
where G
µ,γr εζ s ψ stands for the dimensionless Dirichlet-Neumann operator,
G
µ,γr εζ s ψ “ a
1 ` ε
2| ∇
γζ |
2B
nΦ
|z“εζ“ pB
zΦ ´ µ ∇
γp εζ q ¨ ∇
γΦ q
|z“εζ,
where Φ solves the Laplace equation with Neumann (at the bottom) and Dirichlet (at the surface) boundary conditions #
∆
µ,γΦ “ 0 in tp X, z q P R
dˆ R ´ 1 ă z ă εζ p X qu
φ
|z“εζ“ ψ, B
nΦ
|z“´1“ 0. (1.6) We used the following notations:
∇
γ“
tpB
x, γ B
yq if d “ 2 and ∇
γ“ B
xif d “ 1
∆
µ,γ“ µ B
2x` γ
2µ B
y2` B
z2if d “ 2 and ∆
µ,γ“ µ B
x2` B
z2if d “ 1 and
B
nΦ
|z“´1“ pB
zΦ ´ µ ∇
γp βb q ¨ ∇
γΦ q
|z“´1.
1.2 The rigid lid model
The Euler system (1.1),(1.2), can be written in the dimensionless variables:
$ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ &
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
%
B
tV ` ε p V ¨ ∇
γ` 1
µ w B
zq V “ ´ ∇
γP in Ω
tB
tw ` ε p V ¨ ∇
γ` 1
µ w B
zq w “ ´pB
zP q in Ω
tB
tζ ` εV ¨ ∇
γζ ´ 1 µ w “ 0
∇
µ,γ¨ U “ 0 in Ω
tcurl
µ,γp U q “ 0 in Ω
tU ¨ n “ 0 for z “ ´ 1 P “ εζ for z “ εζ.
(1.7)
We changed the pressure into hydrodynamic pressure P “ P ´ p z ´ εζ q and set P
atm“ 0. The rigid lid model models the motion of the fluid as if the top surface was fixed at z “ 0 (see for instance [6]).
But to derive this model, we do not brutally take ε “ 0 in the Euler system (1.7) but rather start by a change of time scale, and a change of scale for the pressure:
t
1“ εt and
P
1“ P ε .
Formally, if one takes the limit ε goes to zero in the newly scaled Euler equations, one finds the following so called rigid lid equations (we omit the "primes" for the sake of clarity):
$ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
&
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ %
B
tV ` p V ¨ ∇
γ` 1
µ w B
zq V “ ´ ∇
γP in Ω B
tw ` p V ¨ ∇
γ` 1
µ w B
zq w “ ´pB
zP q in Ω w “ 0
∇
µ,γ¨ U “ 0 in Ω curl
µ,γp U q “ 0 in Ω U ¨ n “ 0 for z “ ´ 1 P “ ζ for z “ 0
(1.8)
where Ω is now the fixed domain
Ω “ tp X, z q P R
d`1, ´ 1 ď z ď 0 u .
1.3 Reminder on the local existence for the Water-Waves equations with flat bottom
We briefly give some reminders about the local well-posedness theory for the Water-Waves equations and their local existence (see [12] Chapter 4 for a complete study, and also [11]). After scaling
t
1“ εt
as for the rigid lid equation, one gets from (1.5) the equations:
$ ’
&
’ %
ε B
tζ ´ 1
µ G r εζ s ψ “ 0 ε B
tψ ` ζ ` ε
2 | ∇
γψ |
2´ ε µ
p G r εζ s ψ ` εµ ∇
γζ ¨ ∇
γψ q
22 p 1 ` ε
2µ | ∇
γζ |
2“ 0.
(1.9)
As explained above, we prove later that the solutions of the full Water-Waves equations (1.9) converge strongly to the solutions of the linearized equations (2.23). The strategy is to treat the non-linear terms of the Water-Waves equations as a perturbation of the linearized equation.
Let t
0ą d { 2 and N ě t
0` t
0_ 2 ` 3 { 2 (where a _ b “ sup p a, b q ). The energy for the Water-Waves equations is the following (see Section 1.5 for the notations):
E
Np U q “ | Pψ |
Ht0`3{2` ÿ
|α|ďN
| ζ
pαq|
2` | ψ
pαq|
2where ζ
pαq, ψ
pαqare the so called Alinhac’s good unknowns:
@ α P N
d, ζ
pαq“ B
αζ, ψ
pαq“ B
αψ ´ εw B
αζ, and where P is the differential operator defined by
P “ | D
γ| p 1 ` ? µ | D
γ|q
1{2.
The operator P is of order 1 { 2 and acts as the square roots of the Dirichlet-Neumann operator, since there exists M
0“ C p
hmin1, | ζ |
Ht0`1q where C is a non decreasing function of its arguments, such that
1
M
0| P |
22ď p ψ, 1
µ Gψ q
2ď M
0| Pψ |
22. (1.10)
We consider solutions U “ p ζ, ψ q of the Water-Waves equations in the following space:
E
TN“ t U P C pr 0, T s ; H
t0`2ˆ H
.2p R
dqq , E
Np U p . qq P L
8pr 0, T squ .
The following quantity, called the Rayleigh-Taylor coefficient plays an important role in the Water-Waves problem:
a p ζ, ψ q “ 1 ` ε p ε B
t` εV ¨ ∇
γq w “ ´ ε P
0ρag pB
zP q
|z“εζ.
We can now state the local existence result by Alvarez-Samaniego Lannes (see [2] and [12] Chapter 3):
Theorem 1.1 Let t
0ą d { 2,N ě t
0` t
0_ 2 ` 3 { 2. Let U
0“ p ζ
0, ψ
0q P E
0N. Let ε, γ be such that 0 ď ε, γ ď 1,
and moreover assume that:
D h
miną 0, D a
0ą 0, 1 ` εζ
0ě h
minand a p U
0q ě a
0.
Then, there exists T ą 0 and a unique solution U
εP E
TNto (1.9) with initial data U
0. Moreover, 1
T “ C
1, and sup
tPr0;Ts
E
Np U
εp t qq “ C
2with C
i“ C p E
Np U
0q , 1 h
min, 1
a
0q for i “ 1, 2.
Remark 1.2 Note that the time existence provided by Theorem 1.1 does not depend on ε. Actually, the full Theorem in non flat bottom (see [12]) states that the time of existence for the equations in the original scaling (1.5) is of size
ε_1βwhere β is the size of the topography. In our case, β “ 0 and therefore one gets a time of existence of size
1εfor (1.5), and of size 1 for the rescaled equations (1.9). In the case of a non flat bottom, this long time result stands true in presence of surface tension (see [15]).
1.4 Main result
We investigate in this paper the rigorous limit ε goes to zero in the rescaled Euler equation (1.7) in view of the mathematical justification of the derivation of the rigid lid model (1.8). However, as one shall see in Section 2.1, the only solutions of the rigid lid equations are trivially null. Though the rigid lid model does not then seem of much interest, it implies that the solutions of the rescaled Euler equations (1.7) converge weakly to zero as ε goes to zero. In [4], the same limit in the rigid lid regime is investigated by Bresch and Métivier for the Shallow-Water equations with large bathymetry, which consist in an asymptotic model for the Water-Waves problem in the shallow water regime (µ small). They prove that the rescaled equations are well-posed on a time interval independent on ε (which is equivalent to a large time of existence of size
1εfor the system written in the original variables) and they rigorously pass to the limit as ε goes to zero, and prove the weak convergence of the solutions.
In [15], a similar long time existence result is proved for the Water-Waves equations with large bathymetry and with surface tension. The local existence result (1.1) implies that (we don’t give a precise statement) that there exists T ą 0, and a unique solution p ζ, ψ q P C
1pr 0;
Tεs ; H
Nˆ H
N`1{2p R
dqq to the Water-Waves equations (1.5). Moreover, one has:
1
T “ C
1p ζ
0, ψ
0q , |p ζ, ψ q|
C1pr0;Tεs;HNˆHN`1{2pRdqqď C
2p ζ
0, ψ
0q , (1.11) where C
iare continuous functions of their arguments, and are independent on µ, ε. The bound on the solutions given by (1.11) is uniform with respect to ε, which allows by a compactness argument to extract a convergent sub-sequence of solutions as ε goes to zero. The limit should be zero (at least for ψ) as we discussed above. The question is to understand if the convergence is strong in C pr 0; T s ; H
N´1ˆ H
N´1{2p R
dqq , i.e. "globally in time and space". In this paper, we complete the study of the rigid lid limit for the Water-Waves problem, and give an answer to this question.
Since the equations (1.5) have the structure of a Hamiltonian equation, the following quantity is conserved:
1
2µ p G ψ, ψ q
2` p ζ, ζ q
2. (1.12)
We recall that G is symmetric and positive for the L
2scalar product. Therefore, even after the time change of scale
t
1“ εt
and the limit ε goes to zero, the Hamiltonian is conserved and the strong convergence in L
2of the
unknowns cannot be (except in particular cases). In Section 2.2, we precisely try to highlight the default
of compactness that prevents the strong convergence in dimension 1, in presence of a flat bottom. As
one shall see, the default comes from the linear operator of the Water-Waves equations which has quite a
similar behavior to the wave equation. In Section 3, we prove the equivalence between the Water-Waves equations and the Euler equations, which completes the study of the weak but not strong convergence to zero in Sobolev spaces for the solutions of the Euler equation (1.7) in the rigid lid regime. To sum up, we give here the plan of this article:
– In Section 2.2, we prove that the solutions of the linearized equation does not converge strongly in L
2p R
dq at a fixed time, in the rigid lid scaling p t
1“ εt q for d “ 1, 2.
– In Section 2.3, we prove the strong convergence in L
8pr 0; T r ; L
2ˆ H
1{2p R
dqq of the solutions of the full Water-Waves equation in rigid lid scaling to the solutions of the linearized equations, for d “ 1. It proves that the solutions of the full Water-Waves equations do not converge strongly in L
8pr 0; T r ; L
2ˆ H
1{2p R
dqq to zero, for d “ 1.
– In Section 3, we prove the equivalence between the free-surface Euler equations (1.7) and the Water- Waves equations (1.5) for d “ 1, 2. We show that the solutions of Euler equations converge weakly as ε goes to zero, to the solutions of the rigid lid equation (1.8). The convergence is therefore not strong, at least for d “ 1, according to the preceding points.
1.5 Notations
We introduce here all the notations used in this paper.
1.5.1 Operators and quantities
Because of the use of dimensionless variables (see before the "dimensionless equations" paragraph), we use the following twisted partial operators:
∇
γ“
tpB
x, γ B
yq if d “ 2 and ∇
γ“ B
xif d “ 1
∆
µ,γ“ µ B
2x` γ
2µ B
y2` B
z2if d “ 2 and ∆
µ,γ“ µ B
x2` B
z2if d “ 1
∇
µ,γ“
tp ? µ B
x, γ ? µ B
y, B
zq if d “ 2 and
tp ? µ B
x, B
zq if d “ 1
∇
µ,γ¨ “ ?
µ B
x` γ ?
µ B
y` B
zif d “ 2 and ?
µ B
x` B
zif d “ 1 curl
µ,γ“
tp ?
µγ B
y´ B
z, B
z´ ?
µ B
x, B
x´ γ B
yq if d “ 2.
Remark 1.3 All the results proved in this paper do not need the assumption that the typical wave lengths are the same in both directions, ie γ “ 1. However, if one is not interested in the dependence of γ, it is possible to take γ “ 1 in all the following proofs. A typical situation where γ ‰ 1 is for weakly transverse waves for which γ “ ? µ; this leads to weakly transverse Boussinesq systems and the Kadomtsev–Petviashvili equation (see [13]).
For all α “ p α
1, .., α
dq P N
d, we write
B
α“ B
xα11... B
αxddand
| α | “ α
1` ... ` α
d. We use the classical Fourier multiplier
Λ
s“ p 1 ´ ∆ q
s{2on R
ddefined by its Fourier transform as
F p Λ
su qp ξ q “ p 1 ` | ξ |
2q
s{2p F u qp ξ q
for all u P S
1p R
dq . The operator P is defined as
P “ | D
γ|
p 1 ` ? µ | D
γ|q
1{2(1.13)
where
F p f p D q u qp ξ q “ f p ξ q F p u qp ξ q
is defined for any smooth function f of polynomial growth and u P S
1p R
dq . The pseudo-differential operator P acts as the square root of the Dirichlet-Neumann operator (see (1.10)).
We denote as before by G
µ,γthe Dirichlet-Neumann operator, which is defined as followed in the scaled variables:
G
µ,γψ “ G
µ,γr εζ s ψ “ a
1 ` ε
2| ∇
γζ |
2B
nΦ
|z“εζ“ pB
zΦ ´ µ ∇
γp εζ q ¨ ∇
γΦ q
|z“εζ, where Φ solves the Laplace equation
# ∆
γ,µΦ “ 0
Φ
|z“εζ“ ψ, B
nΦ
|z“´1“ 0.
For the sake of simplicity, we use the notation G r εζ s ψ or even Gψ when no ambiguity is possible.
1.5.2 The Dirichlet-Neumann problem
In order to study the Dirichlet-Neumann problem (1.4), we need to map Ω
tinto a fixed domain (and not on a moving subset). For this purpose, we introduce the following fixed strip:
S “ R
dˆ p´ 1; 0 q and the diffeomorphism
Σ
εt: S Ñ Ω
tp X, z q ÞÑ p 1 ` εζ p X qq z ` εζ p X q . (1.14) It is quite easy to check that Φ is the variational solution of (1.4) if and only if φ “ Φ ˝ Σ
εtis the variational solution of the following problem:
# ∇
µ,γ¨ P p Σ
εtq ∇
µ,γφ “ 0
φ
z“0“ ψ, B
nφ
z“´1“ 0, (1.15)
and where
P p Σ
εtq “ | det J
Σεt| J
Σ´ε1 tt
p J
Σ´ε1 tq ,
where J
Σεtis the jacobian matrix of the diffeomorphism Σ
εt. For a complete statement of the result, and a proof of existence and uniqueness of solutions to these problems, see [12] Chapter 2.
We introduce here the notations for the shape derivatives of the Dirichlet-Neumann operator. More precisely, we define the open set Γ Ă H
t0`1p R
dq as:
Γ “ t Γ “ ζ P H
t0`1p R
dq , D h
0ą 0, @ X P R
d, εζ p X q ` 1 ´ ě h
0u and, given a ψ P H
.s`1{2p R
dq, the mapping:
G r ε ¨s : Γ ÝÑ H
s´1{2p R
dq Γ “ p ζ, b q ÞÝÑ G r εζ s ψ.
We can prove the differentiability of this mapping. See Appendix A for more details. We denote
d
jG p h, k q ψ the j-th derivative of the mapping at p ζ, b q in the direction p h, k q . When we only differentiate
in one direction, and no ambiguity is possible, we simply denote d
jG p h q ψ or d
jG p k q ψ.
1.5.3 Functional spaces
The standard scalar product on L
2p R
dq is denoted by p , q
2and the associated norm | ¨ |
2. We will denote the norm of the Sobolev spaces H
sp R
dq by | ¨ |
Hs.
We introduce the following functional Sobolev-type spaces, or Beppo-Levi spaces:
Definition We denote H 9
s`1p R
dq the topological vector space
H 9
s`1p R
dq “ t u P L
2locp R
dq , ∇ u P H
sp R
dqu endowed with the (semi) norm | u |
H9s`1pRdq“ | ∇ u |
HspRdq.
Just remark that H 9
s`1p R
dq{ R
dis a Banach space (see for instance [10]).
The space variables z P R and X P R
dplay different roles in the equations since the Euler formulation (1.1) is posed for p X, z q P Ω
t. Therefore, X lives in the whole space R
d(which allows to take fractional Sobolev type norms in space), while z is actually bounded. For this reason, we need to introduce the following Banach spaces:
Definition The Banach space p H
s,kpp´ 1, 0 q ˆ R
dq , | . |
Hs,kq is defined by H
s,kpp´ 1, 0 q ˆ R
dq “
č
k j“0H
jpp´ 1, 0 q ; H
s´jp R
dqq , | u |
Hs,k“ ÿ
k j“0| Λ
s´jB
zju |
2.
2 The rigid lid limit for the Water-Waves equations
We prove in this section that the solutions of the linearized Water-Waves equations in flat bottom converge weakly but non strongly in L
2as ε goes to zero, in the rigid lid regime. We then prove the strong convergence in L
8tL
2xof the solutions of the full Water-Waves equations to the solutions of the linearized equations, in the same regime, in dimension 1.
Remark 2.1 The hypothesis "d “ 1" can be removed, if one can prove a dispersive estimate for the linear operator of the Water-Waves equation of the form of Theorem 2.9, in dimension d “ 2. The flat bottom hypothesis seems however less easy to remove, due to technical reasons, such as the complexity of the asymptotic expansion of G with respect to the surface when the bottom is non flat.
2.1 Solutions of the rigid lid equations
We recall the formulation of the rigid lid equations with a flat bottom (see also [6] for reference):
$ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
&
’ ’
’ ’
’ ’
’ ’
’ ’
’ ’
’ %
B
tV ` p V ¨ ∇
γ` 1
µ w B
zq V “ ´ ∇
γP in Ω B
tw ` p V ¨ ∇
γ` 1
µ w B
zq w “ ´pB
zP q in Ω w “ 0
∇
µ,γ¨ U “ 0 in Ω curl
µ,γp U q “ 0 in Ω U ¨ e
z“ 0 for z “ ´ 1 P “ ζ for z “ 0
(2.16)
where Ω is the fixed domain
Ω “ tp X, z q P R
d`1, ´ 1 ď z ď 0 u .
We prove here that in fact, there is only one trivial solution to the system (2.16):
Lemma 2.2 Let us consider the following div-curl problem:
$ ’
’ ’
&
’ ’
’ %
∇
µ,γ¨ U “ 0 in Ω curl
µ,γp U q “ 0 in Ω U ¨ e
z“ 0 for z “ ´ 1 w “ 0
(2.17)
with the above notations. Then, the unique solution in H 9
1p Ω q to the system (2.17) is U “ 0.
Proof The vector field U “ 0 is of course a solution to the system (2.17). Let us now consider a solution U to this problem. The curl free condition over U states that there exists a potential Φ such that U “ ∇
µ,γΦ in Ω. The divergence free condition provides ∆
µ,γΦ “ 0 in Ω. With the boundary conditions (recall that w is the vertical component of the velocity at the surface), the potential Φ satisfies the following Laplace equation:
$ ’
&
’ %
∆
µ,γΦ “ 0 in Ω
∇
µ,γΦ
|z“0¨ n “ 0
∇
µ,γΦ
|z“´1¨ n “ 0,
where n denotes the upward normal vector in the vertical direction. It is well known (see for instance [12] Chapter 2 and Appendix A) that the unique solution to this system in H 9
1p Ω q satisfies ∇
µ,γΦ “ 0.
1l
Let us explain this result on a physical point of view. The rigid-lid equations are obtained after scaling the time
t
1“ εt
in the free surface Euler equations, and passing to the limit as ε goes to zero. Therefore, in the original time variable, t “
tε1, the rigid lid limit consists in letting the amplitude of the waves goes to zero AND moving forward in time at a rate
1ε. At the limit, after an infinite time, all the interesting components of the waves moved to ˘8 and there only remains a static fluid (U “ 0) with a flat surface. It is possible that some vortices remain, but they are not seen by the model (curl p U q “ 0). One could generalize the study lead in this paper to the Water-Waves equation with vorticity (see [7]).
2.2 The linearized equation around ζ “ 0 in the rigid lid regime
We prove here that the solutions to the Water-Waves equations do not converge strongly in L
2as ε goes to zero. To this purpose, we study the lack of compactness induced by the linear operator of the Water-Waves equations. We recall that in the rigid lid regime, we perform a time scaling
t
1“ εt.
The Water-Waves equations can be written in dimensionless form, and in the rigid lid scaling:
$ ’
&
’ %
ε B
tζ ´ 1
µ G r εζ s ψ “ 0 ε B
tψ ` ζ ` ε
2 | ∇
γψ |
2´ ε µ
p G r εζ s ψ ` εµ ∇
γζ ¨ ∇
γψ q
22 p 1 ` ε
2µ | ∇
γζ |
2“ 0.
(2.18)
1In fact the result stands even if the bottom is non flat.
Since we are interested in the convergence as ε goes to zero, we write this system under the form:
$ &
%
ε B
tζ ´ 1
µ G
0ψ “ εf ε B
tψ ` ζ “ εg.
(2.19) We will treat the non-linear terms f, g present in (2.19) as a perturbation of the linearized equations.
To this purpose, we start to study the solutions of the linearized system:
# ε B
tζ ´
µ1G
0ψ “ 0
ε B
tψ ` ζ “ 0, (2.20)
where G
0“ G r 0 s is the Dirichlet-Neumann operator in flat bottom and flat surface, and is given by (see for instance [12] Chapter 1):
G
0ψ “ ?
µ | D
γ| tanh p ?
µ | D
γ|q ψ (2.21)
for all ψ P S p R
dq . It is easy to check that the solution ζ of (2.20) satisfies the following equation:
ε
2B
tζ ` 1
µ G
0ζ “ 0. (2.22)
The equation (2.22) looks like a wave equation, except that G
0is not exactly an order two operator (since it acts like an order one operator for high frequencies). In the case of the wave equation on R
$ ’
&
’ %
ε
2B
tu ´ B
2xu “ 0 u p x, 0 q “ u
0p ε B
tu qp x, 0 q “ u
1, with u
0, u
1smooth, the solutions have components of the form
u
0p x ´ 1
ε t q ` u
0p x ` 1 ε t q
(plus other terms we do not detail). Each of these two components converges pointwise to 0 as ε goes to zero (since u
0p t, x q ÝÑ
|x|Ñ`8
0) and does not converge strongly in L
2p R q , since for instance its L
2p R q norm does not depend on ε. The same behavior, combined with some dispersive effects stands for (2.22).
We start to give an existence result for the linearized equations (2.20). Note that it is not difficult to prove formally that if p ζ, ψ q is the solution of (2.20), then the following quantity, called "Hamiltonian"
is conserved through time:
1
2µ p G
0ψ, ψ q
2` 1 2 | ζ |
22. Therefore, we introduce the following operator:
P “ | D
γ| p 1 ` ? µ | D
γ|q
1{2.
Note that P has the same behavior as the square root of G
0defined by (2.21).
Proposition 2.3 Let s ě 1. Let also ζ
0P H
sp R
dq and ψ
0be such that Pψ
0P H
sp R
dq . Then, there exists a unique solution p ζ, ψ q to the equation:
$ ’
&
’ %
ε B
tζ ´
1µG
0ψ “ 0 ε B
tψ ` ζ “ 0
p ζ p 0, X q , ψ p 0, X qq “ p ζ
0p X q , ψ
0p X qq
(2.23)
such that p ζ, Pψ q P C p R , H
sp R
dqq X C
1p R , H
s´1p R
dqq . Moreover, one has:
@ t P R ,
ˆ ζ ψ
˙
p t q “ e
´tεLˆ ζ
0ψ
0˙
where
e
tL“ 1 2
ˆ 1 ´ iω p D q
´
iωpDq11
˙
e
´iωpDqt` 1 2
ˆ 1 iω p D q
1
iωpDq
1
˙
e
iωpDqt(2.24)
and
ω p D q “
d | D
γ| tanh p ? µ | D
γ|q
? µ . (2.25)
Remark 2.4 The notation e
tLrefers to the linear operator of the Water-Waves, defined by L “
ˆ 0 ´
µ1G
01 0
˙ .
Proof Let us fix s ě 1, and p ζ
0, ψ
0q P H
sp R
dq
2. If p ζ, ψ q is a solution to (2.23), then ζ satisfies the
following equation: $
’ ’
’ &
’ ’
’ %
ε
2B
2tζ ` 1
µ G
0ζ “ 0 ζ p x, 0 q “ ζ
0p ε B
tζ qp x, 0 q “ ζ
1,
(2.26)
in the space C p R , H
sp R
dqqX C
1p R , H
s´1p R
dqq , with ζ
1“
µ1G
0ψ
0. As usual, we take the Fourier transform in space of the equation (2.26) and we denote ζ p p t, ξ q the Fourier transform of ζ with respect to the variable X for a fixed t. The distribution B x
tζof S
1p R ˆ R
dq is equal to the distribution B
tζ p using the Banach- Steinhaus theorem. We therefore get the following equation in the distributional sense of S
1p R ˆ R
dq :
$ ’
&
’ %
ε
2B
t2ζ p `
?1µ| ξ
γ| tanh p ? µ | ξ
γ|q ζ p “ 0 ζ p p ξ, 0 q “ ζ p
0p ξ q
p ε B
tζ p qp ξ, 0 q “ ζ p
1p ξ q .
It is then easy to check that the unique solution to this equation is given by
ζ p p ξ, t q “ p ζ p
0p ξ q ´ iω p ξ q ψ x
0p ξ q
2 q exp
iωpξqεt`p ζ p
0p ξ q ` iω p ξ q ψ x
0p ξ q
2 q exp
´iωpξqεt(2.27) with
ω p ξ q “
d | ξ
γ| tanh p ? µ | ξ
γ|q
? µ
such that
µ1G
0“ ω p D q
2. The desired regularity is easy to get by noticing that ω p ξ q ψ x
0p ξ q “ | ξ
γ|
p 1 ` ? µ | ξ
γ|q
1{2ψ x
0d tanh p ? µ | ξ
γ|qp 1 ` ? µ | ξ
γ|q
? µ | ξ
γ|
“ Pψ z
0d tanh p ? µ | ξ
γ|qp 1 ` ? µ | ξ
γ|q
? µ | ξ
γ| .
with Pψ
0P H
sp R
dq and that there exists C
1, C
2ą 0 independent of µ, ε such that:
@ ξ P R
d, C
1ď
d tanh p ? µ | ξ
γ|qp 1 ` ? µ | ξ
γ|q
? µ | ξ
γ| ď C
2.
The same method applies for ψ and one gets that if p ζ, ψ q satisfies (2.23), then necessarily:
ψ p p t, ξ q “ x ψ
0cos p ω p ξ q t
ε q ´ sin p ω p ξ q
tεq
ω p ξ q ζ p
0. (2.28)
It is easy to check that p ζ, ψ q given by (2.27), (2.28) is a solution to the system (2.20). l
The oscillating component of the solution of the linearized Water-Waves equations (2.23) appears in the explicit formula (2.27), but we need to prove an oscillating phase method type result in order to conclude to the weak convergence to zero and strong non-convergence. The phase ω is actually not differentiable at ξ “ 0, which prevent the direct use of standards results on stationary phase methods.
Proposition 2.5 Let t ą 0 and u P C
01p R
dq . Then, one has ż
Rd
exp
iεtωpξqu p ξ q dξ ÝÑ
εÑ0
0.
Proof One can compute:
@ ξ ‰ 0, p ∇ ω qp ξ q “ tanh p| ξ
γ|q ` | ξ
γ|p 1 ´ tanh
2p| ξ
γ|qq 2 a
| ξ
γ| tanh p| ξ
γ|q
ξ
| ξ
γ| . (2.29)
It is easy to see that ∇ ω does not vanish for ξ ‰ 0. It is also easy to show that ω is not differentiable in ξ “ 0. But the derivative ∇ ω stay bounded as ξ goes to zero. We treat separately the cases d “ 1 and d ą 1.
- Case d “ 1 One writes ż
R
exp
itεωpξqu p ξ q dξ “ ż
0´8
exp
itεωpξqu p ξ q dξ ` ż
`80
exp
itεωpξqu p ξ q dξ. (2.30) Since ω is right and left differentiable in ξ “ 0, according to the expression (2.29), one can use standard methods of stationary phase. Indeed, one can write:
ż
0´8
exp
itεωpξqu p ξ q dξ “ ε ti
ż
0´8
1 ω
1p ξ q
d
dξ p exp
itεωpξqq u p ξ q dξ
“ ε ti
u p 0 q
ω
1´p 0 q e
iεtωp0q´ ε ti
ż
0´8
e
itεωpξqd dξ p u p ξ q
ω
1´p ξ q q dξ
where ω
1`p 0 q and ω
1´p 0 q are respectively the right and left derivatives of ω at ξ “ 0. We then get the desired result. The same goes for the second integral of (2.30).
- Case d ą 1 We use the standard integrating by part method for the truncated integral:
ż
|ξ|ěδ
e
iεtωpξqu p ξ q dξ “ ż
|ξ|ěδ
ε it
1
| ω
1p ξ q|
2ÿ
d j“1B B ξ
jω p ξ q B
B ξ
jp e
itεωpξqq u p ξ q dξ
“ ε it
ż
|ξ|“δ
1
| ω
1p ξ q|
2ÿ
d j“1B B ξ
jω p ξ q e
itεωpξqu p ξ q n
jdσ p ξ q
´ ε it
ż
|ξ|ěδ
ÿ
d j“1B B ξ
jp 1
| ω
1p ξ q|
2B B ξ
jω p ξ q u p ξ qq e
itεωpξqdξ (2.31) where n
jstands for the j-th component of the normal external vector at the surface | ξ | “ δ. Now, one can check that:
1
| ω
1p ξ q|
2B B ξ
jω p ξ q “ 2ξ
ja | ξ
γ| tanh p| ξ
γ|q
| ξ
γ| tanh p| ξ
γ|q ` p 1 ´ tanh
2p| ξ
γ|qq| ξ
γ|
2and thus this function is bounded as ξ goes to zero. Moreover, one can check by computation that
| B B ξ
jp 1
| ω
1p ξ q|
2B
B ξ
jω p ξ qq| ď C
| ξ |
as ξ goes to zero, and consequently this function is integrable at ξ “ 0 (remember that d ą 1 here).
Therefore, one can pass the limit as δ goes to zero in the formula (2.31) and get ż
Rd
e
itεωpξqu p ξ q dξ “ ´ ε it
ż
Rd
ÿ
d j“1B B ξ
jp 1
| ω
1p ξ q|
2B B ξ
jω p ξ q u p ξ qq e
iεtωpξqdξ and we get the desired result.
l
Now it is easy to prove the following result:
Theorem 2.6 Let s ě 1. Let p ζ
0, ψ
0q be such that p ζ
0, Pψ
0q P H
sp R
dq. Then, for all t P R , the solution p ζ, ψ qp t q of the system (2.20) converges weakly to zero, and does not converge strongly, in L
2p R
dq .
Proof We prove it for ζ. We first assume that ζ, p ψ p P C
01p R
dq . The weak convergence to zero is a consequence of the explicit formulation of the solution given by (2.27) and the Proposition 2.5. To prove the strong non convergence, one computes:
| ζ p t, . q|
22“ 1
p 2π q
d| ζ p p t, . q|
22“ 1
p 2π q
dż
Rd
1 2
` p 1 ` cos p 2ω p ξ q t
ε qq| ζ p
0p ξ q|
2` ω p ξ q
2p 1 ´ cos p 2ω p ξ q t
ε qq| ψ x
0p ξ q|
2˘
` Re p ζ p
0x ψ
0qp ξ q sin p 2ω p ξ q t ε q dξ
where we expanded the square modulus of the explicit formulation of ζ p given by (2.27). We can then conclude, using Proposition 2.5 to the following convergence:
| ζ |
22ÝÑ
εÑ0
1
2 p| ζ
0|
22` | ω p D q ψ
0|
22q and thus ζ p t, . q does not converge strongly to zero.
For the general case, one proceeds by density of C
01p R
dq in L
2p R
dq . l
Remark 2.7 As explained above, if ζ, ψ is a solution to (2.23), then the Hamiltonian 1
2µ p G
0ψ, ψ q
2` 1 2 | ζ |
22is conserved through time. Since
1µp G
0ψ, ψ q
2„ | Pψ |
22, the strong convergence of both ζ and ψ to zero
cannot happen (except for zero initial conditions). However, it would have been possible that some
transfers of energy occur between ζ and ψ, with one of the unknown strongly converging to zero. The
Theorem 2.6 states that such behavior does not happen.
2.3 Lack of strong convergence for the full Water-Waves equations in dimen- sion 1
In all this section, we work with a flat bottom and in dimension 1:
d “ 1, b “ 0.
The study in dimension 2 should however not be hard to do, if one gets a dispersive estimate of the form of Theorem 2.9 for the Water-Waves operator in dimension 2. In the previous section, we proved that the solutions of the linearized Water-Waves equations (2.20) weakly converge as ε goes to zero in L
2p R q , and do not converge strongly. We now establish a similar result for the solutions of the full nonlinear Water-Waves system (1.9). To this purpose, we write this system under the form:
# ε B
tζ ´
µ1G
0ψ “ εf ε B
tψ ` ζ “ εg with
f “ 1
µε p G r εζ s ψ ´ G
0ψ q and
g “ ´p 1
2 | ∇
γψ |
2´ 1 µ
p G r εζ s ψ ` εµ ∇
γζ ¨ ∇
γψ q
22 p 1 ` ε
2µ | ∇
γζ |
2q .
Remark 2.8 As suggested by the notations, the quantity f is of size O p 1 q with respect to ε (although it has a
1εfactor), as given by the asymptotic extension of G r εζ, 0 s given later in Proposition 2.13. We are going to treat the non-linear terms f, g as perturbations of the linear equation as ε goes to zero.
For this proof, we need to use a dispersive estimate for the linear Water-Waves equations with flat bottom, proved in [14]:
Theorem 2.9 Let
ω :
$ ’
&
’ %
R ÝÑ R ξ ÞÝÑ
d | ξ | tanh p ? µ | ξ |q
? µ .
Then, there exists C ą 0 independent on µ such that, for all µ ą 0:
@ t ą 0, @ ϕ P S p R q | e
itωpDqϕ |
8ď C p 1 µ
1{41
p t { ? µ q
1{8` 1
p t { ? µ q
1{2qp| ϕ |
H1` | x B
xϕ |
2q .
Remark 2.10 The dispersion in
t18stated by this Theorem is absolutely not optimal. Actually, one should expect a decay of order
t11{3in dimension 1 (see for instance [14] for variants of Theorem 2.9).
However, the
t11{8decay suffices to prove the main result of this section.
In view of use of Theorem 2.9, we also need a local existence result in weighted Sobolev spaces, proved also in [14]. For all N P N , we define E
xNby
E
xN“ E
Np ζ, ψ q ` ÿ
1ď|α|ďN´2
| xζ
pαq|
22` | Pxψ
pαq|
22.
Theorem 2.11 Let us consider the assumption of Theorem 1.1, and then consider p ζ, ψ q the unique solution provided by the Theorem 1.1, of the Water-Waves equation (1.9). If p ζ
0, ψ
0q P E
xN, then one has
p ζ, ψ q P L
8pr 0; T s , E
xNq , with
p ζ, ψ q
L8pr0;Ts,ExNqď C
2. We now state the main result of this section:
Theorem 2.12 Let p ζ
0, ψ
0q be such that p ζ
0, ψ
0q P E
xN. Let denote T ą 0, p ζ
W W, ψ
W Wq the solution of the Water-Waves equations (1.9) in p ζ, ψ q P L
8pr 0; T s , E
xNq given by Theorem 1.1 on r 0; T s and p ζ
L, ψ
Lq the solution of the linearized equation (2.20) given by Proposition 2.3, both with initial condition p ζ
0, ψ
0q . Then, one has:
|p ζ
L, Pψ
Lq ´ p ζ
W W, Pψ
W Wq|
L8pp0;Tq;L2pRqqď p ε
1{8µ
3{16` ε
1{2µ
1{4q C
2(2.32) where C
2is given by Theorem 1.1.
Proof In all this proof, we will denote by C
Nany constant of the form C
N“ C p E
xNp ζ
0, ψ
0q , 1
h
min, 1
a
0q (2.33)
where C is a non decreasing function of its arguments. Let us define p ζ, ψ q “ p ζ
W W, Pψ
W Wq´p ζ
L, Pψ
Lq which is defined on r 0; T s . We use the evolution operator e
tεLdefined by (2.24) to write:
B
tp e
εtLˆ ζ
ψ
˙
q “ e
εtLF p ˆ ζ
ψ
˙ qp t q where
F p ˆ ζ
ψ
˙ q “
˜
1µε
p G p εζ q ψ ´ G
0ψ q
´
12| ∇
γψ |
22´
1µpGpεζqψ`εµ∇γζ¨∇γψq2
2p1`ε2µ|∇γζ|2q
¸
(2.34) and thus one has
@ t P r 0; T s ,
ˆ ζ ψ
˙
“ ż
t0
e
s´tε LF p ˆ ζ
ψ
˙
qp s q ds. (2.35)
We set A “ ˆ I 0
0 P
˙
and look for a estimate of the L
2norm of A
tp ζ, ψ q . The proof consists in using the decay estimate of the linear operator e
tLof the Water-Waves equation of Theorem 2.9. More precisely, the operator F is "almost" bilinear (at least up to a O p ε q order term), which allows us to control the L
2norm of the integral (2.35) by writing estimates of the form
| ż
t0
Ae
s´tε LF p ˆ ζ
ψ
˙
qp s q ds |
2ď | ż
t0
Ae
s´tε LB p ˆ ζ
ψ
˙ ,
ˆ ζ ψ
˙
qp s q ds |
2` εC
where C does not depend on ε, and with B a bilinear operator. The proof then consists in using a Strichartz type of estimate, using the dispersive nature of e
itL, given by Theorem 2.9 to get a control of the remaining integral of the form
| ż
t0
e
s´tε LB p ˆ ζ
ψ
˙ ,
ˆ ζ ψ
˙
qp s q ds |
2ď p ε q
1{8t
αwith α ą 0.
We start to write F under the form F “ B ` εR where B is bilinear, and R is a least of size O p 1 q with respect to ε. To this purpose, recalling that F is given by (2.34), we get inspired by the following Proposition, which gives an asymptotic extension of G r εζ, 0 s (see [12] Proposition 3.44) with respect to ε:
Proposition 2.13 Let t
0ą d { 2, s ě 0 and k “ 0, 1. Let ζ P H
s`pk`1q{2X H
t0`2p R
dq be such that:
D h
miną 0, @ X P R
d, 1 ` εζ p X q ě h
minand ψ P H
s`k{2p R
dq . We get:
| G ψ ´ G
0ψ ´ ε G
1|
Hs´1{2ď ε
2µ
3`k4C p 1 h
min, µ
max, | ζ |
Ht0`1, | ζ |
Hs`pk`1q{2q| Pψ |
Hs`k{2, where G
1“ ´ G
0p ζ p G
0¨qq ´ µ ∇
γ¨ p ζ ∇
γ¨q .
We therefore write 1
µε p G p εζ q ´ G
0q ψ “ 1 µ G
1´ 1
µε p G
1` G
0´ G p εζ qq ψ (2.36) where the second term of the right hand side satisfies the following estimate, using Proposition 2.13 with k “ 1 and s “ 1 { 2:
| 1
µε p G
1` G
0´ G p εζ qq|
2ď C p 1 h
min, µ
max, | ζ |
Ht0`1, | ζ |
H1{2`1q| Pψ |
H1. (2.37) The second component of F (recall that it is given by (2.34)) is easier to decompose, and using Proposition A.1 one gets
´ 1
2 | ∇
γψ |
2´ 1 µ
p G p εζ q ψ ` εµ ∇
γζ ¨ ∇
γψ q
22 p 1 ` ε
2µ | ∇
γζ |
2q “ ´ 1
2 | ∇
γψ |
2´ 1
µ p G p εζ q ψ q
2` εR p ψ q (2.38) with
| R p ψ q|
2Lď C
N, (2.39)
where C
Nis given by (2.33). Using (2.36) with the control (2.37), and (2.38) with the control (2.39), one gets:
F p ˆ ζ
ψ
˙ q “ B p
ˆ ζ ψ
˙ q ,
ˆ ζ ψ
˙ qq ` εR
ˆ ζ ψ
˙
q (2.40)
with
B p ˆ ζ
ψ
˙ q ,
ˆ ζ ψ
˙ qq “
ˆ
1µ
G
1ψ
´
12| ∇
γψ |
2´
µ1G p εζ q ψ q
2˙
(2.41) and
| R ˆ ζ
ψ
˙
q|
2ď C
N. (2.42)
We therefore get, using (2.40) and (2.42) (recall that C
Nis a constant of the form (2.33)):
| ż
t0
Ae
s´tε LF p ˆ ζ
ψ
˙
qp s q ds |
2ď | ż
t0
e
s´tε LAB p s q ds |
2` εC
N.
We denote in the following lines p¨ , ¨q
L2xthe L
2scalar product with respect to the space variable. One writes:
| ż
t0
e
s´tε LAB p s q ds |
22“ | ż
t0