HAL Id: hal-00137752
https://hal.archives-ouvertes.fr/hal-00137752
Preprint submitted on 21 Mar 2007
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.
Sharp well-posedness results for the generalized Benjamin-Ono equation with high nonlinearity
Stéphane Vento
To cite this version:
Stéphane Vento. Sharp well-posedness results for the generalized Benjamin-Ono equation with high
nonlinearity. 2007. �hal-00137752�
hal-00137752, version 1 - 21 Mar 2007
BENJAMIN-ONO EQUATION WITH HIGH NONLINEARITY
ST´EPHANE VENTO
Abstract. We establish the local well-posedness of the generalized Benjamin- Ono equation∂tu+H∂x2u±uk∂xu= 0 inHs(R),s >1/2−1/kfork≥12 and without smallness assumption on the initial data. The conditions >1/2−1/k is known to be sharp since the solution mapu0 7→uis not of classCk+1 on Hs(R) fors <1/2−1/k. On the other hand, in the particular case of the cubic Benjamin-Ono equation, we prove the ill-posedness inHs(R),s <1/3.
1. Introduction and statement of the results
1.1. Introduction. Our purpose in this paper is to study the initial value problem for the generalized Benjamin-Ono equation
(gBO)
∂
tu + H∂
x2u ± u
k∂
xu = 0, x, t ∈ R , u(x, t = 0) = u
0(x), x ∈ R ,
where k ∈ N \ {0}, H is the Hilbert transform defined by Hf (x) = 1
π pv 1 x ∗ u
(x) = F
−1− i sgn(ξ) ˆ f (ξ) (x)
and with initial data u
0belonging to the Sobolev space H
s( R ) = (1−∂
x2)
−s/2L
2( R ).
The case k = 1 was deduced by T.B. Benjamin [1] and later by H. Ono [14] as a model in internal wave theory. The Cauchy problem for the Benjamin-Ono equation has been extensively studied. It has been proved in [16] that (BO) is globally well- posed (i.e. global existence, uniqueness and persistence of regularity of the solution) in H
s( R ) for s ≥ 3, and then for s ≥ 3/2 in [15] and [5]. Recently, T. Tao [17] proved the well-posedness of this equation for s ≥ 1 by using a gauge transformation. More recently, combining a gauge transformation with a Bourgain’s method, A.D. Ionescu and C.E. Kenig [4] shown that one could go down to L
2( R ), and this seems to be, in some sense, optimal. It is worth noticing that all these results have been obtained by compactness methods. On the other hand, L. Molinet, J.-C. Saut and N. Tzvetkov [10] proved that, for all s ∈ R , the flow map u
07→ u is not of class C
2from H
s( R ) to H
s( R ). Furthermore, building suitable families of approximate solutions, H. Koch and N. Tzvetkov proved in [9] that the flow map is not even uniformly continuous on bounded sets of H
s( R ), s > 0. As an important consequence of this, since a Picard iteration scheme would imply smooth dependance upon the initial data, one see that such a scheme cannot be used to get solutions in any space continuously embedded in C([0, T ], H
s( R )).
Key words and phrases. NLS-like equations, Cauchy problem.
1
For higher nonlinearities, that is for k ≥ 2, the picture is a little bit different. It turns out that one can get local well-posedness results throught a Picard iteration scheme but for small initial data only. This seems mainly due to the fact that the smoothing properties of the linear group V (·) associated to the linear (BO) equation is just sufficient to recover the lost derivative in the nonlinear term, but does not allow to get the required contraction factors. On the other hand, for large initial data, one can prove local well-posedness by compactness methods together with a gauge transformation. Unfortunately, this usually requires more smoothness on the initial data. We summurize now the known results about the Cauchy problem for (gBO) equations when k ≥ 2.
In the case of the modified Benjamin-Ono equation (k = 2), C.E. Kenig and H.
Takaoka [8] have recently obtained the global well-posedness in the energy space H
1/2( R ). This have been proved thanks to a localized gauge transformation com- bined with a L
2xTestimate of the solution. This result is known to be sharp since the solution map u
07→ u is not C
3in H
s( R ), s < 1/2 (see [12]).
For (gBO) with cubic nonlinearity (k = 3), the local well-posedness is known in H
s( R ), s > 1/3 for small initial data [12] but only in H
s( R ), s > 3/4, for large initial data. Moreover, the ill-posedness has been proved in H
s( R ), s < 1/6 [12].
In this paper, we show the ill-posedness of the cubic Benjamin-Ono equation in H
s( R ), s < 1/3, which turns out to be optimal according to the above results.
When k ≥ 4, by a scaling argument, one can guess the best Sobolev space in which the Cauchy problem is locally well-posed, that is, the critical indice s
csuch that (gBO) is well-posed in H
s( R ) for s > s
cand ill-posed for s < s
c. Recall that if u(x, t) is a solution of the equation then u
λ(x, t) = λ
1/ku(λx, λ
2t) (λ > 0) solves (gBO) with initial data u
λ(x, 0) and moreover
ku
λ(·, 0)k
H˙s= λ
s+k1−12ku(·, 0)k
H˙s.
Hence the ˙ H
s( R ) norm is invariant if and only if s = s
k= 1/2 − 1/k and one can conjecture that s
c= s
k.
In the case of small initial data, this limit have been reached by L. Molinet and F. Ribaud [12]. This result is almost sharp in the sense that the flow map u
07→ u is not of class C
k+1from H
s( R ) to C([0, T ], H
s( R )) at the origin when s < s
k, [11]. This lack of regularity is also described by H.A. Biagioni and F. Linares in [2] where they established, using solitary waves, that the flow map is not uniformly continuous in ˙ H
sk( R ), k ≥ 2.
For large initial data, the local well-posedness of (gBO) is only known in H
s( R ), s ≥ 1/2, whatever the value of k. This have been proved in [11] by using the gauge transformation
(1.1) u 7−→
GP
+(e
−iR−∞x uku),
together with compactness methods. Note also that very recently, in the particular case k = 4, N. Burq and F. Planchon [3] derived the local well-posedness of (gBO) in the homogeneous space ˙ H
1/4( R ).
In this paper, our aim is to improve the results obtained in [11] for large initial
data. We show that for all k ≥ 12, (gBO) is locally well-posed in H
s( R ), s > s
k.
Our proofs follow those of [11] : we perform the gauge transformation w = G(u)
of a smooth solution u of (gBO) and derive suitable estimates for w. The main interest of this transformation is to obtain an equation satisfied by w where the nonlinearity u
ku
xis replaced by terms of the form P
+(u
kP
−u
x) in which one can share derivatives on u with derivatives on u
k. Working in the surcritical case, this allows to get a contraction factor T
νin our estimates. It is worth noticing that ν = ν(s) verifies lim
s→skν(s) = 0, and this explains why our method fails in the critical case s = s
k. On the other hand, the restriction k ≥ 12 appears when we estimate the integral term
P
+e
−iR−∞x uku Z
x−∞
u
k−2Hu
xx(see section 3.2). This term doesn’t seem to have a ”good structure” since the bad interaction
Q
ju Z
x−∞
(P
ju)
k−2HP
ju
xxforbids the share of the antiderivative R
x−∞
with other derivatives.
1.2. Main results. Our main results read as follows.
Theorem 1. Let k ≥ 12 and u
0∈ H
s( R ) with s > 1/2 − 1/k. Then there exist T = T (s, k, ku
0k
Hs) > 0 and a unique solution u ∈ C([0, T ]; H
s( R )) of (gBO) such that
kD
s+1/2xuk
L∞xL2T
< ∞, (1.2)
kD
s−1/4xuk
L4xL∞T< ∞, (1.3)
kP
0uk
L2xL∞T< ∞.
(1.4)
Moreover, the flow map u
07→ u is Lipschitz on every bounded set of H
s( R ).
As mentioned previously, these results are in some sense almost sharp. However, the critical case s = s
kremains open. We will only consider the most difficult case, that is the lowest values for s. More precisely we will prove Theorem 1 for s
k< s < 1/2.
In the case k = 3, we have the following ill-posedness result.
Theorem 2. Let k = 3 and s < 1/3. There does not exist T > 0 such that the Cauchy problem (gBO) admits an unique local solution defined on the interval [0, T ] and such that the flow map u
07→ u is of class C
4in a neighborhood of the origin from H
s( R ) to H
s( R ).
This result implies that we cannot solve (gBO) with k = 3 in H
s( R ), s < 1/3 by a contraction method on the Duhamel formulation. Recall that for small initial data [12], we have local well-posedness in H
s( R ) for s > 1/3. In view of this, we can conjecture that (gBO) is locally well-posed in H
s( R ), s > 1/3.
The remainder of this paper is organized as follows. In section 2, we first derive
some linear estimates on the free evolution operator associated to (gBO) and we
define our resolution space. Then we give some technical lemmas which will be
used for nonlinear estimates. In section 3 we introduce the gauge transformation
and derive the needed nonlinear estimates. The section 4 is devoted to the proof
of Theorem 1. Finally we prove our ill-posedness result in the Appendix.
The author is grateful to Francis Ribaud for several useful comments on the subject.
1.3. Notations. For two positive numbers x, y, we write x . y to mean that there exists a C > 0 which does not depend on x and y, and such that x ≤ Cy. In the sequel, this constant may depend on s and k. We also use ν = ν(s, k) to denote a positive power of T which may differ at each occurrence.
Our resolution space is constructed thanks to the space-time Lebesgue spaces L
pxL
qTand L
qTL
pxendowed for T > 0 and 1 ≤ p, q ≤ ∞ with the norm
kf k
LpxLqT
= kf k
LqT([0;T])
Lpx(R)
and kf k
LqTLpx
= kf k
Lpx(R)LqT([0;T])
. When p = q we simplify the notation by writing L
pxT.
The well-known operators F (or ˆ ·) and F
−1(or ˇ ·) are the Fourier operators defined by ˆ f (ξ) = R
R
e
−ixξf (x)dx. The pseudo-differential operator D
αxis defined by its Fourier symbol |ξ|
α. Let P
+and P
−be the Fourier projections to [0, +∞[
and ] − ∞, 0]. Thus one has
iH = P
+− P
−.
Let η ∈ C
0∞( R ), η ≥ 0, supp η ⊂ {1/2 ≤ |ξ| ≤ 2} with P
∞−∞
η(2
−kξ) = 1 for ξ 6= 0.
We set p(ξ) = P
j≤−3
η(2
−jξ) and consider, for all k ∈ Z , the operators Q
kand P
krespectively defined by
Q
k(f ) = F
−1(η(2
−kξ) ˆ f (ξ)) and P
k(f ) = F
−1(p(2
−kξ) ˆ f (ξ)).
Therefore we have the standard Littlewood-Paley decomposition
(1.5) f = X
j∈Z
Q
j(f ) = P
0(f ) + X
j≥−2
Q
j(f ) = P
0(f ) + ˜ P (f ).
We also need the operators P
≤kf = X
j≤k
Q
jf, P
≥kf = X
j≥k
Q
jf.
We finally introduce the operators ˜ P
+= P
+P ˜ and ˜ P
−= P
−P ˜ in order to obtain the smooth decomposition
(1.6) f = ˜ P
−(f ) + P
0(f ) + ˜ P
+(f ).
2. Linear estimates and technical lemmas
2.1. Linear estimates and resolution space. Recall that (gBO) is equivalent to its integral formulation
(2.1) u(t) = V (t)u
0∓ 1 k + 1
Z
t 0V (t − τ)∂
x(u
k+1)(τ)dτ,
where V (t) = F
−1e
itξ|ξ|F is the generator of the free evolution. Let us now gather the well-known estimates on the group V (·) in the following lemma.
Lemma 1. Let ϕ ∈ S( R ), then
kV (t)ϕk
L∞TL2x. kϕk
L2, (2.2)
kD
x1/2V (t)ϕk
L∞xL2T. kϕk
L2, (2.3)
kD
x−1/4V (t)ϕk
L4xL∞T. kϕk
L2.
(2.4)
Moreover, for 0 < T < 1, we have
kP
0V (t)ϕk
L2xL∞T. kP
0ϕk
L2. (2.5)
The estimate (2.2) is straightforward whereas the proof of the Kato smoothing effect (2.3) and the maximal in time inequality (2.4) can be found in [6]. Estimate (2.5) has been proved in [7].
These estimates motivate the definition of our resolution space.
Definition 1. For s
k< s < 1/2, we define the space X
Ts= {u ∈ S
′( R
2), kuk
XTs<
∞} where 0 < T < 1 and
(2.6) kuk
XTs= kuk
L∞THsx+ kD
xs+1/2uk
L∞xL2T+ kD
s−1/4xuk
L4xL∞T+ kP
0uk
L2xL∞T. Thus lemma 1 implies immediately that for all ϕ ∈ S( R ) and 0 < T < 1,
(2.7) kV (t)ϕk
XTs. kϕk
Hs.
We now give some families of norms which are controlled by the X
Tsnorm. This will be usefull to derive some nonlinear estimates in the sequel.
Definition 2. A triplet (α, p, q) ∈ R ×[2, ∞]
2is said to be 1-admissible if (α, p, q) = (1/2, ∞, 2) or
(2.8) 4 ≤ p < ∞, 2 < q ≤ ∞, 2 p + 1
q ≤ 1
2 , α = 1 p + 2
q − 1 2 . Proposition 1. If (α − s, p, q) is 1-admissible, then for all u in X
Ts, (2.9) kD
xαuk
LpxLqT. kuk
XTs.
Proof : The inequality
(2.10) kD
s+1/2xuk
L∞xL2T. kuk
XTsyields the result when (α, p, q) = (1/2, ∞, 2). Assume now (α, p, q) 6= (1/2, ∞, 2).
Let r ∈ [4; p]. Then according to Sobolev embedding theorem, kD
s+1/r−1/2xuk
LrxL∞T. kD
xs−1/4uk
L4xL∞T. kuk
XsT. By interpolation with (2.10) we get for all 0 ≤ θ ≤ 1
kD
xs+12−(1−1r)θuk
Lr/θx L2/(1−θ)T
. kuk
XTs.
We deduce (2.9) by taking θ = r/p since the assumption r ≥ 4 is equivalent to
2
p
+
1q≤
12.
We list now all the norms needed for the nonlinear estimates.
Corollary 1. For u ∈ X
Ts, the following quantities are bounded by kuk
XTs. N
1= kuk
LpxL∞T, 4 ≤ p ≤ (
12− s)
−1, N
2= T
−νkuk
L3kxT
, N
3= T
−νkuk
Lk/(1−s)x L2k/sT
, N
4= T
−νkuk
Lk( 13+s)
−1 x Lk( 13−
s 2)−1 T
, N
5= T
−νkuk
L3k/4sx Lk( 12−
2s 3)−1 T
, N
6= T
−νkuk
Lk(1−
s 3)−1 x L6k/sT
, N
7= T
−νkuk
Lk+1x L2k(k+1)T
, N
8= T
−νkuk
L(k−1)( 56−
s 3)−1
x L(k−1)( 2
s 3−1
6)−1 T
, N
9= kD
1−2s+6εxuk
L( 32−3s)
−1
x L1/3εT
, N
10= kD
sxuk
L6xT, N
11= kD
s+1/2−3εxuk
L1/εx L( 12−2ε)
−1 T
, N
12= kD
1/2xuk
L3/sx L( 12−
2s 3)−1 T
,
where ε, ν > 0 are small enough.
Proof :
(i) Let 4 ≤ p ≤ (
12− s)
−1. By separating low and high frequencies, kuk
LpxL∞T. kP
0uk
L2xL∞T+ k P D ˜
s+1/p−1/2xuk
LpxL∞T. kuk
XTs.
Here we used that ˜ P is continuous on L
pxL
qT, 1 ≤ p, q ≤ ∞, and the 1- admissibility of (1/p − 1/2, p, ∞).
(ii)-(vii) We evaluate the norm of the form N = kuk
LpxLqTwith p > 2 and q < ∞. Fix δ > 0 small enough so that α = s − s
k− 2δ > 0 and
1q− δ > 0. Then using the previous decomposition, Bernstein and H¨ older inequalities, we get
N . T
νkP
0uk
L2xL∞T+ T
νk P D ˜
xαuk
LpxL( 1q−δ)
−1 T
.
One complete the proof by noticing that the triplet (α − s, p, (
1q− δ)
−1) is 1-admissible.
(viii) Following the same idea, we write N
8. T
νkP
0uk
L2xL∞T+ T
νk P D ˜
k
k−1(s−sk−2δk)
x
uk
L(k−1)( 56−
s 3)−1
x L(k−1)( 2
s 3−1
6−δ)−1 T
for an appropriate δ > 0. Once again, (
k−1k(s − s
k− 2
kδ) − s, (k − 1)(
56−
s
3
)
−1, (k − 1)(
2s3−
16− δ)
−1) is 1-admissible.
(ix)-(xii) Note finally that the triplets (1 − 3s + 6ε, (
23− 3s)
−1, 1/3ε), (0, 6, 6), (1/2 − 3ε, 1/ε, (
21− 2ε)
−1) and (1/2 − s, 3/s, (
12−
2s3)
−1) are 1-admissible.
We now turn to the non-homogenous estimates. Let us first recall the following result found in [11].
Lemma 2. Let (α
1, α
2) ∈ R
2, (ν
1, ν
2) ∈ R
2+, and 1 ≤ p
1, q
1, p
2, q
2≤ ∞ such that for all ϕ ∈ S( R ),
kD
αx1V (t)ϕk
Lpx1Lq1T
. T
ν1kϕk
L2, kD
αx2V (t)ϕk
Lpx2Lq2T
. T
ν2kϕk
L2. Then for all f ∈ S( R
2),
(2.11) D
αx2Z
t0
V (t − τ)f (τ)dτ
L∞TL2x
. T
ν2kf k
Lp¯2 xLqT¯2,
(2.12) D
αx1+α2Z
t 0V (t − τ)f (τ)dτ
Lp1x LqT1
. T
ν1+ν2kf k
Lpx¯2Lq¯2T
provided min(p
1, q
1) > max(¯ p
2, q ¯
2) or (q
1= ∞ and p ¯
2, q ¯
2< ∞), where p ¯
2and q ¯
2are defined by 1/ p ¯
2= 1 − 1/p
2and 1/ q ¯
2= 1 − 1/q
2. Using lemma 2 we infer the following result.
Lemma 3. For all f ∈ S( R
2), the quantity Z
t0
V (t − τ )f (τ)dτ
XTs
can be esti- mated by
(2.13) kf k
L( 56+
s3)−1 x L( 56−
2s3)−1 T
, kD
sxf k
L6/5xT
, kD
xs−1/2f k
L1xL2T
, kD
s+1/4xf k
L4/3x L1T
. Moreover,
(2.14) D
xs+1/2Z
t0
V (t − τ)f (τ)dτ
L∞xL2T
. kD
xsf k
L1TL2x.
Proof : (2.13) follows from (2.11)-(2.12) since the triplets (s, (
16−
3s)
−1, (
16+
2s
3
)
−1), (0, 6, 6), (1/2, ∞, 2) and (−1/4, 4, ∞) are 1-admissible. Inequality (2.14) is proved in [11], proposition 2.8.
2.2. Technical lemmas. In this subsection, we recall some useful lemmas which allow to share derivatives of various expressions in L
pxL
qTnorms. One can find proofs of lemmas 4-8 in [11, 7].
Here f and g denote two elements of S( R ).
Lemma 4. If α > 0 and 1 < p, q < ∞, then kD
xα(f g)k
LpxLqT. kf k
Lpx1Lq1T
kD
xαgk
Lpx2Lq2T
+ kgk
Lp˜1x LqT˜1
kD
αxf k
Lp˜2 x LqT˜2where 1 < p
1, p
2, q
2, p ˜
1, p ˜
2, q ˜
2< ∞, 1 < q
1, q ˜
1≤ ∞, 1/p
1+1/p
2= 1/ p ˜
1+1/ p ˜
2= 1/p and 1/q
1+ 1/q
2= 1/ q ˜
1+ 1/ q ˜
2= 1/q.
Moreover the cases (p
1, q
1) = (∞, ∞) and (˜ p
1, q ˜
1) = (∞, ∞) are allowed.
Lemma 5. If 0 < α < 1 and 1 < p, q < ∞ then kD
αxF(f )k
LpxLqT
. kF
′(f )k
Lpx1Lq1T
kD
xαf k
Lpx2Lq2T
where 1 < p
1, p
2, q
2< ∞, 1 < q
1≤ ∞, 1/p
1+ 1/p
2= 1/p and 1/q
1+ 1/q
2= 1/q.
Lemma 6. If 0 < α < 1, 0 ≤ β < 1 − α and 1 < p, q < ∞, then kD
βx([D
αx, f]g)k
LpxLqT. kgk
Lpx1Lq1T
kD
xα+βf k
Lpx2Lq2T
where 1 < p
1, q
1, p
2, q
2< ∞, 1/p
1+ 1/p
2= 1/p and 1/q
1+ 1/q
2= 1/q.
Moreover, if β > 0 then q
1= ∞ is allowed.
Lemma 7. If α > 0, β ≥ 0 and 1 < p, q < ∞ then kD
αxP
+(f P
−D
xβg)k
LpxLqT. kD
γx1f k
Lpx1Lq1T
kD
γx2gk
Lpx2Lq2T
where 1 < p
1, q
1, p
2, q
2< ∞, 1/p
1+ 1/p
2= 1/p, 1/q
1+ 1/q
2= 1/q and γ
1≥ α, γ
1+ γ
2= α + β.
As in [11], we introduce the bilinear operator G defined by G(f, g) = F
−11
2 Z
R
ξ
1(ξ − ξ
1)
iξ [sgn(ξ
1) + sgn(ξ − ξ
1)] ˆ f (ξ
1)ˆ g(ξ − ξ
1)dξ
1.
We easily verify that
(2.15) G(f, f) = ∂
x−1(f
xHf
x) = ∂
x−1(−i(P
+f
x)
2+ i(P
−f
x)
2) and
(2.16) G(f, g) = ∂
x−1(−iP
+f
xP
+g
x+ iP
−f
xP
−g
x).
Lemma 8. If 0 ≤ α ≤ 1 and 1 < p, q < ∞ then kD
xαG(f, g)k
LpxLqT. kD
xγ1f k
Lpx1Lq1T
kD
γx2gk
Lpx2Lq2T
where 0 ≤ γ
1, γ
2≤ 1, γ
1+ γ
2= α + 1, 1 < p
1, q
1, p
2, q
2< ∞, 1/p
1+ 1/p
2= 1/p and 1/q
1+ 1/q
2= 1/q.
We will also need the following lemma in order to treat low frequencies in the
integral term.
Lemma 9. If α ≥ 0 and 1 ≤ p, q ≤ ∞ then kP
0(f D
αxg)k
LpxLqT. kD
γx1f k
Lpx1Lq1T
kD
γx2gk
Lpx2Lq2T
+ kP
0f k
Lp˜1x LqT˜1
kD
αxP
0gk
Lp˜2 x LqT˜2where γ
1, γ
2≥ 0, α = γ
1+γ
2, 1 < p
i, q
i, p ˜
i, q ˜
i< ∞, 1/p
1+1/p
2= 1/ p ˜
1+1/ p ˜
2= 1/p and 1/q
1+ 1/q
2= 1/ q ˜
1+ 1/ q ˜
2= 1/q.
Proof : We split the product f D
αxg as follows :
(2.17) f D
αxg = P
+f P
+D
αxg + P
+f P
−D
xαg + P
−f P
+D
xαg + P
−f P
−D
αxg.
It is sufficient to consider the contribution of the first two terms. For the first one, we remark that
P
0[P
+f P
+(D
xαg)] = P
0[P
0(P
+f )P
0(P
+D
αxg)]
and thus using the continuity of P
0on L
pxL
qT,
kP
0[P
+f P
+(D
xαg)]k
LpxLqT. kP
0(P
+f )P
0(P
+D
αxg)k
LpxLqT. kP
0f k
Lp˜1x LqT˜1
kD
αxP
0gk
Lp˜2 x LqT˜2.
For the second term in (2.17) we have typically contributions of the form P
0[P
0(P
+f )P
0(P
−D
xαg)]
which are treated as above, and P
0[ ˜ P
+f P ˜
−D
αxg]. Using decomposition (1.5), one can write
P
0( ˜ P
+f P ˜
−D
αxg) = P
0X
j∈Z
Q
j( ˜ P
+f )P
j( ˜ P
−D
αxg) + X
j∈Z
P
j( ˜ P
+f )Q
j( ˜ P
−D
αxg) +P
0X
|p|≤2
X
j∈Z
Q
j( ˜ P
+f )Q
k−j( ˜ P
−D
αxg) .
By a careful analysis of the various localisations, we get P
0( ˜ P
+f P ˜
−D
αxg) = P
0h X
|p|.1
X
j∈Z
Q
j( ˜ P
+f )Q
j+p( ˜ P
−D
αxg) i .
Here we define the operators Q
λj= 2
−λjD
λxQ
j. It follows that P
0( ˜ P
+f P ˜
−D
xαg) = P
0h X
|p|.1
X
j∈Z
Q
−γj 1( ˜ P
+D
xγ1f )Q
γj+p1( ˜ P
−D
γx2g) i .
Thus using Cauchy-Schwarz and H¨ older inequalities, and Littlewood-Paley theo- rem,
kP
0( ˜ P
+f P ˜
−D
αxg)k
LpxLqT. X
|p|.1
h X
j∈Z
|Q
−γj 1P ˜
+D
γx1f |
21/2X
j∈Z
|Q
γj1P ˜
−D
xγ2g|
21/2i
LpxLqT
. X
j∈Z
|Q
−γj 1P ˜
+D
γx1f |
212Lpx1LqT1
X
j∈Z
|Q
γj1P ˜
−D
γx2g|
212Lpx2LqT2
. kD
γx1f k
Lpx1Lq1T
kD
γx2gk
Lpx2Lq2T
.
3. Nonlinear estimates
3.1. Gauge transformation. By a rescaling argument, it is sufficient to solve
(3.1) u
t+ Hu
xx= 2u
ku
x(equation with minus sign in front of the nonlinearity could be treated in the same way). If u ∈ C([0, T ]; H
∞( R )) is a smooth solution, we define the gauge transfor- mation
1(3.2) w = P
+(e
−iFu), F = F (u) = Z
x−∞
u
k(y, t)dy.
The rest of this subsection is devoted to the proof of the following estimate.
Proposition 2. Let be k ≥ 12 and s
k< s < 1/2. Let u ∈ C([0, T ]; H
∞( R )) be a solution of the Cauchy problem associated to (3.1) with initial data u
0∈ H
∞( R ).
Then there exist ν = ν(s, k) > 0 and a positive nondecreasing polynomial function p
ksuch that
kuk
XTs. ku
0k
Hs+ T
νp
k(kuk
XTs)kuk
XTs+(ku
0k
kHs+ T
νp
k(kuk
XTs)kuk
XTs)kD
xs+1/2wk
L∞xL2T. (3.3)
Proof : We start by splitting u according to (1.6). Then, using that |P
+u| =
|P
−u| (since u is real), we deduce
(3.4) kuk
XTs. kP
0uk
XTs+ k P ˜
+uk
XTs.
For the low frequencies, we use the Duhamel formulation of (gBO), lemma 3 and (2.7) to get
kP
0uk
XTs. kP
0u
0k
Hs+ kP
0D
x1/2u
k+1k
L1xL2T
. ku
0k
Hs+ ku
k+1k
L1xL2T
. ku
0k
Hs+ T
νkuk
k+1Lk+1x L2k(k+1)T
. ku
0k
Hs+ T
νkuk
XsT.
Now we consider the second term in the right-hand side of (3.4). As mentioned in [11], ˜ P
+u satisfies the dispersive equation
∂
t( ˜ P
+u) + H∂
x2( ˜ P
+u) = ˜ P
+(e
iFu
kw
x) − P ˜
+(e
iFu
k∂
xP
−(e
−iFu)) + i P ˜
+(u
2k+1).
Thus, according to lemma 3 k P uk ˜
XTs. kV (t)u
0k
XTs+
Z
t 0V (t − τ) ˜ P
+(e
iFu
kw
x)(τ)dτ
Xs T+kD
sxu
2k+1k
L6/5xT
+ k P ˜
+(e
iFu
k∂
xP
−(e
−iFu))k
L( 56+
s3)−1 x L( 56−
2s3)−1 T
. ku
0k
Hs+ A + B + C.
Obviously,
B . ku
2kk
L3/2 xTkD
sxuk
L6xT
. kuk
2kL3kxT
kD
xsuk
L6xT
. T
νkuk
2k+1Xs T.
1we can also setF= 12Rx
−∞ukin the non-rescaled caseut+Huxx=ukux.
Term C has a structure P
+(f P
−g
x) thus by lemma 7 C . kD
x1/2(e
iFu
k)k
L6/5x L3T
kD
x1/2(e
−iFu)k
L3/sx L( 12−
2s 3)−1 T
. C
1C
2. Using lemmas 4-5, we infer C
1. kD
1/2xu
kk
L6/5x L3T
+ kD
x1/2e
iFk
L( 12−s)
−1
x L2/sT
ku
kk
L( 13+s)
−1 x L( 13−
s 2)−1 T
. kuk
k−1L(k−1)( 56−
s 3)−1
x L(k−1)( 2
s 3−1
6)−1 T
kD
1/2xuk
L3/sx L( 12−
2s 3)−1 T
+kD
−1/2x(u
ke
iF)k
L( 12−s)
−1
x L2/sT
kuk
kLk( 13+s)
−1 x Lk( 13−
s 2)−1 T
. T
νkuk
kXTs+ T
νku
kk
L1/(1−s)x L2/sT
kuk
kXTs. T
νkuk
kXsT
+ T
νkuk
2kXs(3.5)
Tand in the same way C
2. kD
1/2xuk
L3/sx L( 12
−2s 3)−1 T
+ kD
1/2xe
−iFk
L( 4
s 3−1
2)−1 x L( 12−
2s 3)−1 T
kuk
L( 12−s)−1
x L∞T
. kD
1/2xuk
L3/sx L( 12−
2s 3)−1 T
+ ku
kk
L3/4sx L( 12−
2s 3)−1 T
kuk
L( 12−s)−1
x L∞T
. kuk
XsT+ T
νkuk
k+1Xs T. (3.6)
Combining (3.5) and (3.6), C is bounded by C . T
ν(kuk
k+1XsT
+ kuk
2k+1XsT
+ kuk
3k+1XsT
) . T
νp
k(kuk
XsT)kuk
XTs. In order to study the contribution of A, we decompose e
iFu
kw
xas
e
iFu
kw
x= D
1/2x(e
iFu
kHD
1/2xw) − [D
1/2x, e
iFu
k]HD
1/2xw.
Therefore, according to lemma 3, and using the fact that ˜ P
+is continuous on L
1xL
2T, A . kD
sx(e
iFu
kHD
1/2xw)k
L1xL2T+ k[D
1/2x, e
iFu
k]HD
x1/2wk
L( 56+
s 3)−1 x L( 56−
2s 3)−1 T
. A
1+ A
2.
Note that A
1cannot be treated by lemma 4, so we use lemma A.13 in [7]. This leads to
A
1. kD
xs(e
iFu
k)k
L(1−
s 3)−1
x L3/2sT
kD
x1/2(e
−iFu)k
L3/sx L( 12−
2s 3)−1 T
+ ku
kk
L1xL∞TkD
s+1/2xwk
L∞xL2T
. A
11C
2+ A
k12kD
s+1/2xwk
L∞xL2T
.
By lemma 4 we bound the contribution of A
11by A
11. kD
xsu
kk
L(1−
s 3)−1
x L3/2sT
+ kD
sxe
iFk
L3/2sx L6/sT
ku
kk
L1/(1−s)x L2/sT
. kuk
k−1L(k−1)( 56−
s 3)−1
x L(k−1)( 2
s 3−1
6)−1 T
kD
sxuk
L6xT+ kuk
kLk(1−
s 3)−1
x L6k/sT
kuk
kLk/(1x −s)L2k/sT
. T
νkuk
kXTs+ T
νkuk
2kXTs.
To treat A
12= kuk
LkxL∞Twe use the Duhamel formulation of (gBO) and lemma 2, A
12. kV (t)u
0k
LkxL∞T+
Z
t 0V (t − τ)∂
xu
k+1(τ)dτ
LkxL∞T
. ku
0k
Hs+ kD
sxk+1/2+3εu
k+1k
L(1x−ε)−1L( 12+2ε)
−1 T
and setting ε
′=
13(s − s
k) − ε > 0 it follows that kD
sxk+1/2+3εu
k+1k
L(1−ε)x −1L( 12+2ε)
−1 T
. kD
s+1/2−3εx ′uk
L1/εx ′L( 12−2ε
′)−1 T
ku
kk
L(1−
1
3(s−sk))−1
x L
3 2(s−sk)−1 T
. T
νkuk
XTskuk
kLk(1−
1
3(s−sk))−1
x L∞T
. T
νkuk
k+1Xs T.
Finally, according to lemma 6 we write A
2. kD
x1/2(e
iFu
k)k
L6/5x L3T
kD
x1/2(e
−iFu)k
L3/sx L( 12−
2s 3)−1 T
. C
1C
2. T
νp
k(kuk
XTs)kuk
XTs, witch complete the proof of (3.3).
3.2. Estimate of kD
s+1/2xwk
L∞xL2T. Now our aim is to estimate the term kD
s+1/2xwk
L∞xL2Twhich appears in (3.3). More precisely we will prove the following proposition.
Proposition 3. Let k ≥ 12 and s
k< s < 1/2. For all solution u ∈ C ([0, T ]; H
∞( R )) of (3.1) with initial data u
0∈ H
∞( R ), we have the following bound,
(3.7) kD
xs+1/2wk
L∞xL2T. p
k(ku
0k
Hs)ku
0k
Hs+ T
νp
k(kuk
XTs)kuk
XTswhere p
kis a positive nondecreasing polynomial function.
Proof : Following [11], we see that w satisfies the equation w
t+ Hw
xx= P
+[2e
−iF(−ku
kP
−u
x− iP
−u
xx)]
−ik(k − 1)P
+e
−iFu Z
x−∞
u
k−2u
xHu
x. (3.8)
Thus using the Duhamel formulation of (3.8) and lemma 3 we infer kD
s+1/2xwk
L∞xL2T. kD
s+1/2xV (t)w(0)k
L∞xL2T+kP
+[2e
−iF(−ku
kP
−u
x− iP
−u
xx)]k
L( 56+
s 3)−1 x L( 56−
2s 3)−1 T
+ D
s+1/2xZ
t0
V (t − τ)P
+e
−iFu Z
x−∞
u
k−2u
xHu
xdτ
L∞xL2T