Global well-posedness of the Cauchy problem

for a fifth-order KP-I equation in

anisotropic Sobolev spaces
arXiv:1712.09334v1 [math.AP] 23 Dec 2017

Yongsheng Li a , Wei Yanb , Yimin Zhangc

a
School of Mathematics, South China University of Technology,
Guangzhou, Guangdong 510640, P. R. China
b
School of Mathematics and Information Science,
Henan Engineering Laboratory for Big Data Statistical Analysis and Optimal Control,
Henan Normal University, Xinxiang, Henan 453007, P. R. China

c
Department of Mathematics, Wuhan University of Technology, Wuhan430070, China
Wuhan, Hubei 430070, P. R. China

Abstract In this paper, we consider the Cauchy problem for the fifth-order KP-I
equation

1
ut + ∂x5 u + ∂x−1 ∂y2 u + ∂x (u2 ) = 0.
2

Firstly, we establish the local well-posedness of the problem in the anisotropic Sobolev
spaces H s1 , s2 (R2 ) with s1 > − 89 and s2 ≥ 0. Secondly, we establish the global well-
posedness of the problem in H s1 , 0 (R2 ) with s1 > − 47 . Our result improves considerably
the results of Saut and Tzvetkov (J. Math. Pures Appl. 79(2000), 307–338.) and Li and
Xiao (J. Math. Pures Appl. 90(2008), 338–352.) and Guo, Huo and Fang (J. Diff. Eqns.
263 (2017), 5696–5726).

Emails: yshli@scut.edu.cn (YS Li); yanwei19821115@sina.cn (W Yan); zhangyimin@whut.edu.cn

(YM Zhang)
Keywords: fifth-order KP-I equation; Cauchy problem; Anisotropic Sobolev spaces

Short Title: Cauchy problem for fifth-order KP-I equation

Corresponding Author: Wei Yan

Email Address: yanwei19821115@sina.cn

AMS Subject Classification: 35Q53; 35B30

2
 Global well-posedness of the Cauchy problem
for a fifth-order KP-I equation in
anisotropic Sobolev spaces

Yongsheng Li a , Wei Yanb , Yimin Zhangc

a
School of Mathematics, South China University of Technology,
Guangzhou, Guangdong 510640, P. R. China
b
School of Mathematics and Information Science,
Henan Engineering Laboratory for Big Data Statistical Analysis and Optimal Control,
Henan Normal University, Xinxiang, Henan 453007, P. R. China

c
Department of Mathematics, Wuhan University of Technology, Wuhan430070, China
Wuhan, Hubei 430070, P. R. China

Abstract
In this paper, we consider the Cauchy problem for the fifth-order KP-I equation

1
ut + ∂x5 u + ∂x−1 ∂y2 u + ∂x (u2 ) = 0.
2

Firstly, we establish the local well-posedness of the problem in the anisotropic Sobolev
spaces H s1 , s2 (R2 ) with s1 > − 89 and s2 ≥ 0. Secondly, we establish the global well-
posedness of the problem in H s1 , 0 (R2 ) with s1 > − 47 . Our result improves considerably
the results of Saut and Tzvetkov (J. Math. Pures Appl. 79(2000), 307–338.) and Li and
Xiao (J. Math. Pures Appl. 90(2008), 338–352.) and Guo, Huo and Fang (J. Diff.
Eqns. 263 (2017), 5696–5726).

1. Introduction

This paper is devoted to studying the Cauchy problem for the fifth-order KP-I equa-

Emails: yshli@scut.edu.cn (YS Li); yanwei19821115@sina.cn (W Yan); zhangyimin@whut.edu.cn

(YM Zhang)

3
tion

1
ut + ∂x5 u + ∂x−1 ∂y2 u + ∂x (u2 ) = 0, (1.1)
2
u(x, y, 0) = u0 (x, y) (1.2)

in anisotropic Sobolev space H s1,s2 (R2 ).

(1.1) appears as a model describing certain long dispersive waves (see [1, 31, 32]). It
is considered as the higher-order version of the following KP equation

1
ut + α∂x3 u + ∂x−1 ∂y2 u + ∂x (u2 ) = 0, (1.3)
2

where the coefficient α may be either positive or negative. The KP equation (1.3) occurs
in physical contexts as models for the propagation of dispersive long waves with weak
transverse effects and is regarded as the two-dimensional extensions of the Korteweg-
de-Vries equation (see [30]). When α < 0, (1.3) is known as the KP-I equation. When
α > 0, (1.3) is known as the KP-II equation.
Several people have studied its Cauchy problem for (1.3), see [4, 15, 16, 25–29, 37, 49–
54] for the KP-II equation (1.3) with α > 0, and see [3, 7, 8, 13, 17–19, 21–24, 33, 39–
42, 44, 56] for the KP-I equation (1.3) with α < 0.
For the KP-II equation, by using the Fourier restriction norm method, Bourgain
[4] established the global well-posedness of its Cauchy problem in L2 (R2 ) and L2 (T2 ).
Takaokao and Tzvetkov [51] and Isaza and Mejı́a [25] established the local well-posedness
of KP-II equation in H s1,s2 (R2 ) with s1 > − 31 and s2 ≥ 0. Takaoka [49] established the
local well-posedness of KP-II equation in H s1 ,0 (R2 ) with s1 > − 12 with the assumption
that

− 21 +ǫ
|ξ| Fx u0 <∞

L2

for the suitable chosen ǫ. By introducing some resolution spaces, Hadac et al. [16]
established the small data global well-posedness and scattering result of KP-II equation
1
in the homogeneous anisotropic Sobolev space Ḣ − 2 , 0 (R2 ) defined in [16] and arbitrary
1
large initial data local well-posedness in both homogeneous Sobolev space Ḣ − 2 , 0 (R2 ) and
1
inhomogeneous anisotropic Sobolev space H − 2 , 0 (R2 ). Recently, by using new bilinear

4
estimates, Koch and Li [37] established the global well-posedness and scattering for the
KP-II equation in three space dimensions with small initial data.
For the KP-I equation, Kenig, Molinet, Saut and Tzvetkov studied its Cauchy prob-
lem and periodic boundary value problem and showed that the problems are globally
well-posed in the second energy spaces on both R2 and T2 (see [33, 41, 42]). Molinet
et al. [40] proved that the Picard iterative method does not work for the KP-I equation
in standard Sobolev space and in anisotropic Sobolev space, since the flow map fails to
be real-analytic at the origin in these spaces. Ionescu et al. [18] established the global
well-posedness of KP-I in the natural energy space E 1 with the aid of some resolution
spaces and bootstrap inequality and the energy estimates. Molinet et al. [43] established
the local well-posedness of the Cauchy problem for the KP-I equation in H s, 0 (R2 ) with
s > 23 . Guo et al. [13] established the local well-posedness of the Cauchy problem for
the KP-I equation in H 1, 0 (R2 ). Zhang [56] established the local well-posedness of the
1
periodic KP-I initial value problem in the Besov type space B2,1
2
(T2 ).
Saut and Tzvetkov [46] established the global well-posedness of the Cauchy problem
for the fifth order KP-II equation

1
ut − ∂x5 u + α∂x3 u + ∂x−1 ∂y2 u + ∂x (u2 ) = 0, α ∈ R, (1.4)
2
in H s1 ,s2 (R2 ) with s1 > − 14 , s2 ≥ 0. Isaza et al. [20] established the local well-posedness
of the fifth-order KP-II equation in H s1 ,s2 (R2 ) with s1 > − 45 , s2 ≥ 0 and globally well-
posed in H s1 ,0 (R2 ) with s1 > − 47 with the aid of I-method.
By using the Fourier restriction norm method and the Cauchy-Schwartz inequalities
as well as some calculus inequalities, Saut and Tzvetkov [47] established the global well-
posedness of Cauchy problem for the fifth order KP-I equation (1.1) with initial data
u0 ∈ L2 (R2 ) and finite energy. By using the Fourier restriction norm method and
the dyadic decomposed Strichartz estimates, Chen et al. [6] established the local well-
posedness of the problem (1.1)(1.2) in the interpolated energy space E s with 0 < s ≤ 1,
where

s
E s = u0 ∈ E s : ku0 kE s = 1 + |ξ|2 + |µ/ξ| Fxy u0 (ξ, µ) L2 < ∞ .


In particular, Chen et al. established the global well-posedness of the problem (1.1)(1.2)

5
in the energy space E 1 . By using the Fourier restriction norm method and sufficiently
exploiting the geometric structure of the resonant set of (1.1) to deal with the high-
high frequency interaction, Li and Xiao [38] established the global well-posedness of
the Cauchy problem (1.1)(1.2) in L2 (R2 ). Guo et al. [12] established the local well-
posedness of the Cauchy problem for (1.1) in H s,0 (R2 ) with s ≥ − 43 . Yan et al. [55]
proved that the Cauchy problem for (1.1) is locally well-posed in H s,0(R2 ) with s > − 34
and the Cauchy problem for (1.1) is globally well-posed in H s,0(R2 ) with s > − 23
6
with
the aid of I-method introduced in [9, 10]. The method of [55] establishing local well-
posedness is different from the method of [12]. Saut and Tzvetkov [48] have proved
that the Cauchy problem for (1.1) posed on T × R is globally well-posed in the energy
space. Compared to the fifth order KP-II equation, the structure of the fifth order KP-I
equation is complicated. The reason is that the resonant function of the fifth-order KP-I
equation does not possess the same good property as its of fifth-order KP-II equation.
More precisely, the resonant function of the fifth order KP-I equation is

RI (ξ1 , ξ2 , µ1, µ2 ) :=φ(ξ, µ) − φ(ξ1 , µ1 ) − φ(ξ2 , µ2 )

"  2 #
ξ1 ξ2 µ 1 µ 2
= 5ξ 2(ξ12 + ξ1 ξ2 + ξ22 ) − − (1.5)
ξ ξ1 ξ2
and the resonant function of the fifth order KP-II equation is

RII (ξ1 , ξ2, µ1 , µ2 ) :=φ(ξ, µ) − φ(ξ1, µ1 ) − φ(ξ2 , µ2 )

"  2 #
ξ1 ξ2 µ 1 µ 2
= 5ξ 2 (ξ12 + ξ1 ξ2 + ξ22 ) + − . (1.6)
ξ ξ1 ξ2
We remark that RI (ξ1 , ξ2 , µ1 , µ2) = 0 gives a surface, while RII (ξ1 , ξ2 , µ1 , µ2 ) will never
be zero away from the origin.
In this paper, motivated by [7, 20, 38, 47], by using the Fourier restriction norm
method introduced in [2, 5, 36, 45] and developed in [34, 35], the Cauchy-Schwartz
inequality and Strichartz estimates as well as suitable splitting of domains, we establish
the local well-posedness of the Cauchy problem for the fifth-order KP-I equation in the
anisotropic Sobolev spaces H s1 , s2 (R2 ) with s1 > − 89 and s2 ≥ 0; combining the local
well-posness result of this paper with the I-method introduced in [9, 10], we established
the global well-posedness of the problem in H s1 , 0 (R2 ) with s1 > − 74 . Thus, our result
considerably improves the result of [12, 38, 47].

6
 We introduce some notations before presenting the main results. Throughout this
paper, we assume that C is a positive constant which may vary from line to line. a ∼ b
means that there exist constants Cj > 0(j = 1, 2) such that C1 |b| ≤ |a| ≤ C2 |b|. a ≫ b
means that there exist a positive constant C ′ such that |a| > C ′ |b|. 0 < ǫ ≪ 1 means
that 0 < ǫ < 10−4 . We define

h·i := 1 + | · |,
5 µ2
φ(ξ, µ) := ξ + ,
ξ
σ := τ + φ(ξ, µ), σj = τj + φ(ξj , µj )(j = 1, 2),
1
Z
F u(ξ, µ, τ ) := 3 e−ixξ−iyµ−itτ u(x, y, t)dxdydt,
(2π) R2 3

1
Z
Fxy f (ξ, µ) := e−ixξ−iyµ f (x, y)dxdy,
2π R2
1
Z
−1
F u(ξ, µ, τ ) := 3 eixξ+iyµ+itτ u(x, y, t)dxdydt,
(2π) R2 3

1
Z
a
Dx u(x, y, t) := 3 |ξ|aF u(ξ, µ, τ )eixξ+iyµ+itτ dξdµdτ,
(2π) 2 R 3

1
Z Z
2
P u(x, y, t) := 3 F u(ξ, µ, τ )eixξ+iyµ+itτ dξdµdτ,
(2π) 2 |ξ|≥2 R2
1
Z
W (t)f := eixξ+iyµ+itφ(ξ,µ) Fxy f (ξ, µ)dξdµ.
2π R2
Let η be a bump function with compact support in [−2, 2] ⊂ R and η = 1 on (−1, 1) ⊂ R.
For each integer j ≥ 1, we define ηj (ξ) = η(2−j ξ) − η(21−j ξ), η0 (ξ) = η(ξ), ηj (ξ, µ, τ ) =
P
ηj (σ), thus, ηj (σ) = 1. ψ(t) is a smooth function supported in [0, 2] and equals 1 in
j≥0
[0, 1]. Let I ⊂ Rd , χI (x) = 1 if x ∈ I; χI (x) = 0 if x does not belong to I.
We define
Z Z  pr ! 1r
kf kLrt Lpxy := |f |p dxdy dt .
R R2

We denote by H s1,s2 (R2 ) the anisotropic Sobolev space as follows:

n ′
o
s1 ,s2 2 2 s1 s2
H (R ) := u0 ∈ S (R ) : ku0 kH s1 ,s2 (R2 ) = khξi hµi Fxy u0 (ξ, µ)kL2 .
ξµ

The Bourgain space Xbs1 ,s2 is defined by

 
′
Xbs1 ,s2
:= u ∈ S (R3 ) : kukXbs1 ,s2 = hξis1 hµis2 hσib F u(ξ, µ, τ ) <∞ .

L2τ ξµ

7
The space Xbs1 ,s2 ([0, T ]) denotes the restriction of Xbs1 ,s2 onto the finite time interval
[0, T ] and is equipped with the norm
n o
s1 ,s2
kukXb 1 2 ([0,T ]) = inf kgkXb 1 2 : g ∈ Xb , u(t) = g(t) for t ∈ [0, T ] .
s ,s s ,s

For s < 0 and N ∈ N+ , N ≥ 20, we define an operator IN by F IN u(ξ, µ, τ ) =

M(ξ)F u(ξ, µ, τ ), where M(ξ) = 1 if |ξ| < N; M(ξ) = (|ξ|/N)s if |ξ| ≥ N.
The main results of this paper are as follows.
′
Theorem 1.1. (Local well-posedness) Let |ξ|−1Fxy u0 (ξ, µ) ∈ S (R2 ). Then, the
Cauchy problem for (1.1) is locally well-posed in H s1 , s2 (R2 ) with s1 > −9/8, s2 ≥ 0.

Remark 1. We only consider the case of −9/8 < s1 < 0, s2 ≥ 0. For s1 ≥ 0, s2 ≥ 0

the local well-posedness is proved by Li and Xiao [38]. Lemmas 3.1 and 3.2 are the key
ingredients in establishing the bilinear estimates in Lemmas 4.1 and 4.2. Once Lemma
4.1 is proven to be valid, then we can combine it and Lemma 2.6 with the fixed point
argument to obtain the local wellposedness. Since the phase function φ(ξ, µ) is singular
′
at ξ = 0, to define the derivative of W (t)u0 , the requirement |ξ|−1Fxy u0 (ξ, µ) ∈ S (R2 )
is necessary.
′
Theorem 1.2. (Global well-posedness) Let |ξ|−1Fxy u0 (ξ, µ) ∈ S (R2 ). Then the
Cauchy problem for (1.1) is globally well-posed in H s1 , 0 (R2 ) with s1 > −4/7.

Remark 2. We only consider the case of −4/7 < s1 < 0, s2 ≥ 0. The case of
s1 ≥ 0, s2 ≥ 0 is proved by Li and Xiao [38]. For the fifth order KP-II equation, Isaza,
López and Mejı́a [29] obtained the same result about the global well-posedness, that
is, the Cauchy problem for the fifth order KP-II equation is also globally well-posed in
H s1 , s2 (R2 ) with s1 > − 74 , s2 ≥ 0.
The rest of the paper is arranged as follows. In Section 2, we give some preliminaries.
In Section 3, we establish two L2 bilinear estimates. In Section 4, we establish three
bilinear estimates. In Section 5, we prove the local well-posedness. In Section 6, we
firstly prove Lemma 6.1 which is a variation of Theorem 1.1, then, we apply Lemmas
6.1, 4.2 and 2.6 to prove Theorem 1.2.

2. Preliminaries

8
 This section is devoted to present Lemmas 2.1–2.6.

Lemma 2.1. Let b > |a| ≥ 0. Then, we have

Z b
dx 1
1 ≤ Cb , (2.1)
2

−b hx + ai 2
dt
Z
γ ht − aiγ
≤ Chai−γ , γ > 1, (2.2)
hti
ZR
dt − 21
1 ≤ Chai , γ ≥ 1, (2.3)
γ
R hti |t − a| 2
Z K 1
dx K2
1 1 ≤ C 1 . (2.4)
−K |x| 2 |a − x| 2 |a| 2

Proof. The conclusion of (2.1) is given in (2.4) of Lemma 2.1 in [20]. (2.2)-(2.3) can
be seen in Proposition 2.2 of [47]. (2.4) can be seen in [15, Page 6562] .
This completes the proof of Lemma 2.1.

Lemma 2.2. Let T ∈ (0, 1) and s1 , s2 ∈ R and − 21 < b′ ≤ 0 ≤ b ≤ b′ + 1. Then, for

h ∈ Xbs′1 ,s2 , we have

kψ(t)S(t)φkX s1 ,s2 ≤ CkφkH s1 , s2 , (2.5)

 Z t b
ψ t ′
≤ CT 1+b −b khkX s′1 , s2 .

S(t − τ )h(τ )dτ (2.6)
T 0
s ,s
X 1 2
b
b

For the proof of Lemma 2.2, we refer readers to [5, 11, 34] and [14, Lemma 1.7 and
Lemma 1.9].

Lemma 2.3. Let b > 12 . Then,

1
Dx u ≤ CkukX 0,0 . (2.7)
4
L4t L4xy (R3 ) b

For the proof of Lemma 2.3, we refer readers to [15, Theorem 3.1].

Lemma 2.4. Let

 
 µ1 µ2 2 

 ξ1 ξ2
|σ − σ1 − σ2 | = ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12) −  − 

 ξ  ξ1 ξ2  
|ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12)|
≥
4

9
and

F P 1 (u1 , u2 )(ξ, µ, τ )
4
Z 2
Y
= χ|ξ1 |≤ |ξ2 | (ξ1 , µ1 , τ1 , ξ, µ, τ ) F uj (ξj , µj , τj )dξ1dµ1 dτ1 .
R3 4
j=1

For b > 21 , we have

1

|Dx |−1 u2 0,0 .

P 41 (u1 , u2) ≤ C |Dx | 2 u1 (2.8)

L2txy Xb0,0 Xb

Proof. Let

1
f1 (ξ1 , µ1, τ1 ) = |ξ1 | 2 hσ1 ib F u1(ξ1 , µ1 , τ1 ), f2 (ξ2 , µ2 , τ2 ) = |ξ2 |−1 hσ2 ib F u2 (ξ2 , µ2, τ2 ).

To obtain (2.8), it suffices to prove that

1
|ξ1 |− 2 |ξ2|f1 (ξ1 , µ1 , τ1 )f2 (ξ2 , µ2 , τ2 )
Z


2
dξ 1 dµ 1 dτ1

R3
2
Lτ ξµ
Q
hσj ib
j=1
2
Y
≤C kfj kL2τ ξµ . (2.9)
j=1

To obtain (2.9), by duality, it suffices to prove that

1
|ξ1|− 2 |ξ2 |f1 (ξ1 , µ1 , τ1 )f2 (ξ2 , µ2 , τ2 )f (ξ, µ, τ )
Z


2
dξ 1 dµ 1 dτ1

R3
2
Lξµτ
Q
hσj ib
j=1
2
Y
≤Ckf kL2τ ξµ kfj kL2τ ξµ . (2.10)
j=1

We define

|ξ1 |−1 |ξ2|2

Z
I(ξ, µ, τ ) := 2
dξ1 dµ1 dτ1 . (2.11)
R3 Q b
hσj i
j=1

For fixed (ξ, µ, τ ), we make the change of variables L : (ξ1 , µ1 , τ1 ) −→ (∆, σ1 , σ2 ), where

∆ := ξξ1ξ2 (5ξ 2 − 5ξξ1 + 5ξ12),

σ1 := τ1 + φ(ξ1 , µ1 ), σ2 := τ2 + φ(ξ2 , µ2 ).

10
By using a direct computation, since σ = τ + φ(ξ, µ), we have

(ξ1 µ2 − µ1 ξ2 )2
σ1 + σ2 − σ = −∆ + . (2.12)
ξξ1 ξ2

Thus, we have the Jacobian determinant equals

 
∂(∆, σ1 , σ2 ) 2 2

2 2

µ1 µ2
= − 10 ξ1 − ξ2 ξ1 + ξ2 −
∂(ξ1 , µ1 , τ1 ) ξ1 ξ2
  12
1 ξ
ξ12 ξ22 ξ12 ξ22



= − 10 − + (σ1 + σ2 − σ + ∆) 2 . (2.13)
ξ1 ξ2

Notice that it is possible to divide the integration into a finite number of open subsets
Wi such that L is an injective C 1 -function in Wi with non-zero Jacobian determinant.
From (2.13), since |ξ1 | ≤ |ξ42 | and |∆| ∼ |ξ1 ||ξ2 |4 , we have
    
 ∂(∆, σ1 , σ2 ) 
 =10  ξ 2 − ξ 2 ξ 2 + ξ 2 µ1 − µ2 




 ∂(ξ1 , µ1 , τ1 )   1 2 1 2
ξ1 ξ2 
   12 
 1 ξ
=10  ξ12 − ξ22 ξ12 + ξ22 (σ1 + σ2 − σ + ∆) 2



ξ1 ξ2 


1 1
∼|ξ1|−1 |ξ2 |2 |∆| 2 |σ1 + σ2 − σ + ∆| 2 . (2.14)

|∆|
Since |σ1 + σ2 − σ| ≥ 4
, by using the change of variables (ξ1 , µ1, τ1 ) −→ (∆, σ1 , σ2 ) and
(2.4), we have

|ξ2|2 |ξ1|−1
Z
I(ξ, µ, τ ) := χ|ξ1 |≤ |ξ2 | 2 dξ1 dµ1 dτ1
R3 4 Q 2b
hσj i
j=1
χ|∆|≤4|σ1+σ2 −σ|d∆dσ1 dσ2
Z
≤C 2
1 Q
R3 1
|∆| 2 |σ1 + σ2 − σ + ∆| 2 hσj i2b
j=1
!
χ|∆|≤4|σ1+σ2 −σ|d∆ dσ1 dσ2
Z Z
= 1 1 2
R2 R |∆| 2 |σ1 + σ2 − σ + ∆| 2 Q
hσj i2b
j=1
dσ1 dσ2
Z
≤C 2
≤ C. (2.15)
R2 Q 2b
hσj i
j=1

11
Combining (2.10) with (2.15), by using the Cauchy-Schwartz inequality twice, we have
1
|ξ2 ||ξ1|− 2 f1 (ξ1 , µ1 , τ1 )f2 (ξ2 , µ2 , τ2 )f (ξ, µ, τ )
Z
χ|ξ1 |≤ |ξ2 | 2
dξ1 dµ1 dτ1
R3 4 Q
hσj ib
j=1
  21 2
Y
≤C sup I(ξ, µ, τ ) kf kL2τ ξµ kfj kL2τ ξµ
ξ,µ,τ j=1
2
Y
≤Ckf kL2τ ξµ kfj kL2τ ξµ . (2.16)
j=1

This completes the proof of Lemma 2.4.

Lemma 2.5. Let

 
 µ1 µ2 2 

 ξ1 ξ2
|σ − σ1 − σ2 | = ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12) −  − 

 ξ  ξ1 ξ2  
|ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12)|
≥ ,
4
1
b> 2
and G(ξ1 , µ1 , τ1 , ξ, µ, τ ) = f1 (ξ1 , µ1, τ1 )f2 (ξ − ξ1 , µ − µ1 , τ − τ1 )f (ξ, µ, τ ), we have
1 2
|ξ1 |− 2 |ξ2 |G(ξ1 , µ1, τ1 , ξ, µ, τ )
Z 
  Y

2
dξ1dµ1 dτ1 dξdµdτ  ≤ Ckf kL2τ ξµ
 kfj kL2τ ξµ , (2.17)
R6
 Q
hσj i b j=1
j=1
Z 
− 21 2
 |ξ 1 | |ξ|G(ξ 1 , µ ,
1 1τ , ξ, µ, τ )  Y
dξ dµ dτ dξdµdτ ≤ Ckf k kfj kL2τ ξµ , (2.18)
 
1 1 1 2
Lτ ξµ
hσ1 ib hσib
 
 R6 
j=1
Z 
1 2
 |ξ|− 2 |ξ2 |G(ξ1 , µ1, τ1 , ξ, µ, τ )  Y
dξ1 dµ1 dτ1 dξdµdτ  ≤ Ckf kL2τ ξµ kfj kL2τ ξµ , (2.19)
 
hσibhσ2 ib

 R6 
j=1
Z 
− 21 2
 |ξ| |ξ 1 |G(ξ 1 , µ ,
1 1τ , ξ, µ, τ )  Y
dξ1 dµ1 dτ1 dξdµdτ  ≤ Ckf kL2τ ξµ kfj kL2τ ξµ . (2.20)
 
hσibhσ1 ib

 R6 
j=1

|ξ2 |
Proof. We firstly prove (2.17). When 4
≥ |ξ1 |, from Lemma 2.4, we have (2.17) is
|ξ2 | 1 1 1
valid. When 4
< |ξ1|, since |ξ1|− 2 |ξ2| ≤ C|ξ1 | 4 |ξ2 | 4 , from Lemma 2.3, we know that
(2.17) is valid. Let ξ1 = ξ1′ , µ1 = µ′1 , τ1 = τ1′ and −ξ2 = ξ ′ , −τ2 = τ ′ , −(µ − µ1 ) = µ′ and
−ξ = ξ ′ − ξ1′ , −µ = µ′ − µ′1 , −τ = τ ′ − τ1′ and σ2′ = τ2′ − φ′ (ξ2′ , µ′2 ), σ1 = σ1′ = τ1′ − φ(ξ1′ , µ′1 ).
Thus, −σ = σ2′ , σ1 = σ1′ . Let

H(ξ1′ , µ′1 , τ1′ , ξ ′, µ′ , τ ′ ) = f1 (ξ1′ , µ′1 , τ1′ )f2 (−ξ ′ , −µ′ , −τ ′ )f (−ξ2′ , −µ′2 , −τ2′ ).

12
To obtain (2.18), it suffices to prove that
Z 
′ − 21 ′ ′ ′ ′ ′ ′ ′
 |ξ1 | |ξ2||H(ξ1, µ1 , τ1 , ξ , µ , τ ) ′ ′ ′ ′ ′ ′ 
dξ1 dµ1 dτ1 dξ dµ dτ 

hσ1′ ib hσ2′ ib

 R6 
2
Y
≤Ckf kL2τ ξµ kfj kL2τ ξµ . (2.21)
j=1

Obviously, (2.21) follows from (2.17). By using a proof similar to (2.18), we obtain that
(2.19)-(2.20) are valid.
This ends the proof of Lemma 2.5.

Lemma 2.6. Let 0 < b1 < b2 < 21 . Then, we have

kχI (·)ukX 0,0 ≤ C kukX 0,0 , (2.22)

b1 b2

kχI (·)ukX 0,0 ≤ C kukX 0,0 . (2.23)

−b2 −b1

For the proof of Lemma 2.6, we refer the readers to Lemma 3.1. of [26].

3. L2 -bilinear estimates
Inspired by the idea of Lemma 5.1 of [18], we give the proof of Lemma 3.1. For k ∈ Z
and l, j ∈ R, we define

Dk,l,j := (ξ, µ, τ ) : |ξ| ∈ [2k−1 , 2k+1], |µ| ≤ 2l , |τ + φ(ξ, µ)| < 2j ,


[
Dk,∞,j := Dk,l,j .
l∈Z

Lemma 3.1. Assume α ∈ R and k1 , k2 , k3 ∈ Z, kmax = max {k1 , k2 , k3 } and kmin =

min {k1 , k2 , k3 } and j1 , j2 , j3 ∈ Z+ , jmax = max {j1 , j2 , j3 } and fi : R3 → R are L2
functions supported in Dki ,∞,ji (i = 1, 2, 3). We assume that

RI (ξ1, ξ2 , µ1 , µ2 ) ≤ 2k1 +k2 +k3 +2kmax −60 , (3.1)

jmax ≤ k1 + k2 + k3 + 2kmax − 60. (3.2)

(1) If |k1 − k2 | ≤ 5, k1 ≥ 20, then, we have

Z 3
j1 +j2 +j3 k3
− 74 (k1 +k2 )+
Y
(f1 ∗ f2 )f3 dξ1 dµ1 dτ1 ≤ C2 2 2 2 kfj kL2 . (3.3)
R3 j=1

13
(2) If k2 − 10 ≥ k1 and |k2 − k3 | ≤ 5, k2 ≥ 20, then, we have
Z 3
j1 +j2 +j3 k1
− 52 (k2 +k3 )−
Y
(f1 ∗ f2 )f3 dξ1 dµ1 dτ1 ≤ C2 2 2 2 kfj kL2 . (3.4)
R3 j=1

Proof. First we prove (3.3). From (5.4) of [18], we have

Z Z Z
(f1 ∗ f2 )f3 = ˜ ˜
(f1 ∗ f3 )f2 = (f˜2 ∗ f˜3 )f1 . (3.5)
R3 R3 R3

where f˜i = fi (−ξ, −µ, −τ )(i = 1, 2). Due to the symmetry in (3.5), without loss of
generality, we may assume j3 = max {j1 , j2 , j3 } . We define

fi# (ξi , µi , θi ) := fi (ξi , µi , θi − φ(ξi , µi ))(i = 1, 2, 3).

Obviously, kfi# kL2 = kfi kL2 . The left-hand side of (3.3) can be rewritten as follows:
Z 2
Y 
fi# (ξi , µi , θi ) f3# (ξ1 + ξ2 , µ1 + µ2 , RI (ξ1 , µ1 , ξ2 , µ2))dξ1 dξ2 dµ1 dµ2dθ1 dθ2 , (3.6)
6
R i=1

where RI (ξ1 , ξ2 , µ1 , µ2 ) is the resonant function defined as in (1.5). The functions fi#
(i = 1, 2) are supported in the sets
n o
(ξi , µi , θi ) : ξi ∈ I˜ki , µ ∈ R, |θi | ≤ 2ji

and f3# is supported in the set

n o
(ξ3 , µ3, θ3 ) : ξ3 ∈ I˜k3 , µ3 ∈ R, |θ3 | ≤ 2j3 .

We will prove that if gi : R2 −→ R are L2 functions supported in I˜ki × R(i = 1, 2)

and g : R2 −→ R are L2 functions supported in I˜k × R × [−2j , 2j ], then, we have
2
Z Y 
gj (ξj , µj ) g(ξ1 + ξ2 , µ1 + µ2 , RI (ξ1 , µ1 , ξ2 , µ2 ))dξ1dξ2 dµ1 dµ2
R4 j=1
2
j 7 k3 Y
≤C2 2 2− 4 (k1 +k2 )+ 2 kgkL2 kgj kL2 . (3.7)
j=1

When j ≤ k1 + k2 + k3 + 2kmax − 60 is valid, we may assume that the integral in the

left-hand side of (3.7) is taken over the set
 
µ1 µ2
R++ = (ξ1 , µ1 , ξ2 , µ2 ) : ξ1 + ξ2 ≥ 0, − ≥0 (3.8)
ξ1 ξ2

14
since other case can be proved similarly to case (3.8). We make the changes of variables
√ √
µ1 = 5ξ13 + β1 ξ1 , µ2 = − 5ξ23 + β2 ξ2 . (3.9)

From (3.9), we have

µ1 µ2 √  2 √
= 5 ξ1 + ξ22 + β1 − β2 = 5(ξ12 + ξ22 ) + β1 − β2 ≥ 0.

− (3.10)
ξ1 ξ2

From (3.10), we know that

√
β1 − β2 ≥ − 5(ξ12 + ξ22 ). (3.11)

By using the assumption upon g and the definition of RI (ξ1 , µ1, ξ2 , µ2 ), we infer that
 √ 
2
(β1 − β2 ) + 2 5(β1 − β2 )(ξ1 + ξ2 ) + 5ξ1 ξ2 (4ξ1 ξ2 − 3ξ ) ≤ 2j+k−k1−k2 +3 .
2 2 2 
(3.12)


The left hand side of (3.7) can be bounded by

k1 +k2
Z √ √
C2 h1 (ξ1 , 5ξ13 + β1 ξ1 )h2 (ξ2 , − 5ξ23 + β2 ξ2 )
S
√
× h(ξ1 + ξ2 , 5(ξ13 − ξ23 ) + β1 ξ1 + β2 ξ2 , R̃1 (ξ1 , β1 , ξ2, β2 )dξ1dξ2 dβ1 dβ2 , (3.13)

where

S = {(ξ1 , β1 , ξ2 , β2 ) : ξ1 + ξ2 ≥ 0, β1 − β2 satisfies (3.11) − (3.12)} . (3.14)

and

R̃1 (ξ1 , β1 , ξ2 , β2 )
√
=(β1 − β2 )2 + 2 5(β1 − β2 )(ξ12 + ξ22) + 5ξ1ξ2 (4ξ1 ξ2 − 3ξ 2 ). (3.15)

We define the functions hi : R2 −→ R supported in I˜ki × R(i = 1, 2)

k1 √
h1 (ξ1 , β1 ) = 2 2 g1 (ξ1 , 5ξ13 + β1 ξ1 ), (3.16)
k2 √
h2 (ξ2 , β2 ) = 2 2 g2 (ξ2 , − 5ξ23 + β2 ξ2 ). (3.17)

with khi kL2 ≈ kgi kL2 (i = 1, 2).

15
 To prove (3.7), it suffices to prove that
Z
k1 +k2
2 2 h1 (ξ1 , β1 )h2 (ξ2 , β2 )
S̃
√
× h(ξ1 + ξ2 , 5(ξ13 − ξ23) + β1 ξ1 + β2 ξ2 , R̃1 (ξ1 , β1 , ξ2 , β2 )dξ1 dξ2 dβ1 dβ2
2
j k3
− 47 (k1 +k2 )+
Y
≤C2 2
2 2 khkL2 khj kL2 . (3.18)
j=1

Combining (3.11) with (3.12), we have

√
5(B − B2 )
p 1 ≤ β1 − β2
ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) − B2
√
5(B + B2 )
≤ 2 p 1 . (3.19)
ξ1 + ξ22 + ξ 2(ξ 2 − ξ1 ξ2 ) + B2
where
2j+k−k1−k2 +3
B1 = ξ1 ξ2 3ξ12 + 2ξ1 ξ2 + 3ξ22 ,
 
B2 = . (3.20)
5
Now we claim that the following inequality is valid
 √ 
 5B 1

β −  ≤ B3 , (3.21)
 
p
2 2
 2 2
 ξ1 + ξ2 + ξ (ξ − ξ1 ξ2 ) 
where
3(k1 +k2 )
+10
B3 = 2j− 2 . (3.22)

When ξ1 ξ2 ≥ 0, we have
r
3α p
ξ 2 (ξ 2 − ξ1 ξ2 − ) ± B2 + ξ 2(ξ 2 − ξ1 ξ2 ) ≥ ξ 2 . (3.23)
5
By using a direct computation, since
p p B1
B1 ≤ 3ξ 4 , ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 ξ 2 (ξ 2 − ξ1 ξ2 ) ≥ , (3.24)
3
we have
 
 B1 B1 
−
 
 2 p p
 ξ1 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 


B1 B2
=h p ih p i
ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 )
1
× hp p i
ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 + ξ 2 (ξ 2 − ξ1 ξ2 )
B3
≤ . (3.25)
10

16
By using a direct computation, we have
B2 B3
p ≤ . (3.26)
ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 10

Combining (3.25) with (3.26), we have (3.21) is valid.

When ξ1 ξ2 ≤ 0, we have
p p 1
ξ 2(ξ 2 − ξ1 ξ2 ) ± B2 + ξ 2 (ξ 2 − ξ1 ξ2 ) ≥ |ξ||ξ1ξ2 | 2 . (3.27)

Since
h ih i B
1
p p
ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ≥ , (3.28)
4
by a direct computation we have
 
 B1 B1 
−
 
 2 p p
 ξ1 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 


B1 B2
=h p ih p i
ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 )
1
× hp p i
ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 + ξ 2 (ξ 2 − ξ1 ξ2 )
B3
≤ . (3.29)
10
By a direct computation, we have
B2 B3
p ≤ . (3.30)
ξ12 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 ) ± B2 10

By (3.23)-(3.30), we see that (3.21) is valid.

To obtain (3.18), we make the change of variable β1 = β2 + β. Thus, (3.11)-(3.12)
can be rewritten as follows:
√
β ≥ − 5(ξ12 + ξ22),

 2 √ 
β + 2 5β(ξ1 + ξ2 ) + 5(4ξ1 ξ2 − 3ξ ξ1 ξ2 ) ≤ 2j+k−k1−k2 +3 .
2 2 2 2 2
(3.31)


Since |ξj | ∈ [2kj −1 , 2kj +1 ](j = 1, 2), we can assume that ξj = aj 2kj (j = 1, 2), where
1
2
≤ |aj | ≤ 2. Consequently, (3.21) can be rewritten as follows:
 √ 
β − 5f1 (a1 , a2 , k1 , k2) ≤ B3 , (3.32)
 

17
where

f1 (a1 , a2 , k1 , k2 )
a1 a2 2k1 +k2 (3a21 4k1 + a1 a2 2k1 +k2 +1 + 3a22 4k2 )
= p
a21 4k1 + a22 4k2 + (a1 2k1 + a2 2k2 )2 (a21 4k1 + a1 a2 2k1 +k2 + a22 4k2 )
B1
= 2 p . (3.33)
ξ1 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 )
Thus, the left hand side of (3.18) can be bounded by
 √ 
β − 5f1 (a1 , a2 , k1 , k2 )
Z
k1 +k2
2 2 h1 (ξ1, β + β2 )h2 (ξ2 , β2 )χ[−1,1) −m
S̃ B3
× h(ξ1 + ξ2 , A(ξ1 , ξ2 , β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2, β))dξ1dξ2 dβdβ2 , (3.34)

where

S̃ = (ξ1 , ξ2, β, β2 ) ∈ R4 , ξ1 + ξ2 ≥ 0, β satisfies (3.21) ,


(3.35)
√ 
A(ξ1 , ξ2, β) = 5 ξ13 − ξ23 + βξ1 ,

(3.36)
ξ1 ξ2 h 2 √ i
B(ξ1 , ξ2, β) = β + 2 5β(ξ12 + ξ22 ) + 5ξ1 ξ2 (4ξ1ξ2 − 3ξ 2 ) . (3.37)
ξ
3(k1 +k2 )
Let j ′ = j − 2
+ 10. Decompose
√
β′ −
 
′ ′
X
′ ′ 5f1 (a1 , a2 , k1 , k2 )
hi (ξ , β ) = hi (ξ , β )χ[0,1) −m
m∈Z
2j ′
X
′ ′
= hm
i (ξ , β ), i = 1, 2 (3.38)
m∈Z

for all aj ∈ R, 21 ≤ |aj | ≤ 2(j = 1, 2). Obviously, if m1 , m2 ∈ Z, m1 6= m2 , then

2  ′ √ 
Y β − 5f1 (a1 , a2 , k1 , k2 )
χ[0,1) − mi = 0.
i=1
2j ′
2
hm ′ ′
Q
Thus, for m1 , m2 ∈ Z, m1 6= m2 , we have i (ξ , β ) = 0. Consequently, we have
i

i=1
X  21
′ ′
X m ′ ′ 2
khi (ξ , β )kL2′ =
hi (ξ , β ) = khm
i kL2′ . (3.39)
ξ β′ ξ β′
m∈Z L2ξ′ β ′ m∈Z

Thus, (3.34) is controlled by

k1 +k2 X Z
m′
2 2 hm1 (ξ1 , β + β2 )h2 (ξ2 , β2 )
|m−m′ |≤4 S̃

× h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))dξ1dξ2 dβdβ2. (3.40)

18
To prove (3.18), it suffices to prove that
√ !
β2 − 5f1 (a1 , a2 , k1 , k2)
Z
k1 +k2
m′
2 2 hm
1 (ξ1 , β + β2 )h2 (ξ2 , β2 )χ[m′ ,m′ +1)
S̃ 2j ′
× h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))dξ1dξ2 dβdβ2
j 7 k3 ′
≤C2 2 2− 4 (k1 +k2 )+ 2 khkL2 khm m
1 kL2 kh2 kL2 . (3.41)

If (3.41) is valid, by using the Cauchy-Schwartz inequality, we have

k1 +k2 X Z
m′
2 2 hm1 (ξ1 , β + β2 )h2 (ξ2 , β2 )
|m−m′ |≤4 S̃

× h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))dξ1dξ2dβdβ2

j 7 k3
m′
X
≤C2 2 2− 4 (k1 +k2 )+ 2 khkL2 khm
1 kL2 kh2 kL2
|m−m′ |≤4
" !#
j k
− 47 (k1 +k2 )+ 23 m′
X X
=C2 2 2 khkL2 khm
1 k L2 kh2 kL2
m∈Z m−4≤m′ ≤m+4
" # 21  !2  12
j 7 k3 X X X ′
≤C2 2 2− 4 (k1 +k2 )+ 2 khkL2 khm k
1 L2
2  khm
2 k L2

m∈Z m∈Z m−4≤m′ ≤m+4
! 12
j k
− 47 (k1 +k2 )+ 23 m′
X X
≤C2 2 2 khkL2 kh1 kL2 kh2 k2L2
m∈Z m−4≤m′ ≤m+4
2
j k3
− 47 (k1 +k2 )+
Y
≤C2 2 2 2 khkL2 khi kL2 . (3.42)
i=1

To prove (3.41), without loss of generality, we assume that |ξ1 | ≤ |ξ2 |. To prove (3.41),
by using the Minkowski inequality with respect to variables (ξ1 , ξ2 , β) with

S ′ = (ξ1 , ξ2, β) ∈ R3 , ξ1 + ξ2 ≥ 0,

β satisfies (3.21) , (3.43)

the left-hand side of (3.40) is controlled by

√ !
β2 − 5f1 (a1 , a2 , k1 , k2 )
Z
k1 +k2
2 2 χ[m,m+1)
R 2j ′
Z  21
′ 2
× |hm
1 (ξ1 , β + β2 )hm
2 (ξ2 , β2 )| dξ1 dξ2 dβ
S′
Z  21
× |h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))|2dξ1 dξ2 dβ dβ2 . (3.44)
S′

19
From (3.44), it suffices to prove that
Z  12
|h(ξ1 + ξ2 , A(ξ1 , ξ2 , β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2, β))|2 dξ1dξ2 dβ
S′
3(k1 +k2 ) k
+ 23
≤C2− 2 khkL2 . (3.45)

If (3.45) is valid, by using the Cauchy-Schwartz inequality with respect to β2 , we have

(3.44) is controlled by
 √ 
β2 − 5f1 (a1 , a2 , k1 , k2)
Z
k3
−(k1 +k2 )+
C2 2 χ[m′ ,m′ +1)
R 2j ′
′
× khm m
1 kL2 kh2 (·, β2 )kL2ξ khkL2 dβ2
2
′ √
k3
Z (m′ +1)2j + 5f1 (a1 ,a2 ,k1 ,k2 )
′
=C2−(k1 +k2 )+ 2 √
khm m
1 kL2 kh2 (·, β2 )kL2ξ khkL2 dβ2
m′ 2j ′ + 5f1 (a1 ,a2 ,k1 ,k2 ) 2

k3 j′ ′
≤C2−(k1 +k2 )+ 2 2 2 khkL2 khm m
1 kL2 kh2 kL2
j 7 k3 ′
≤C2 2 2− 4 (k1 +k2 )+ 2 khkL2 khm m
1 kL2 kh2 kL2 . (3.46)

To prove (3.45), we may assume that β2 = 0 and make the change of variable β =
√
5ξ1 ξ2 ν. From (3.21), we have
 
2 2
 3ξ 1 + 2ξ ξ
1 2 + 3ξ 2

ν −  ≤ 2−20 . (3.47)
 
p
2 2
 2 2
 ξ1 + ξ2 + ξ (ξ − ξ1 ξ2 ) 

The left hand side of (3.44) is controlled by

Z  21
k1 +k2
C2 2 |h(ξ1 + ξ2 , H1 (ξ1 , ξ2, ν), H2 (ξ1 , ξ2 , ν))dξ1 dξ2 dν , (3.48)
S ′′

where
n o
′′ ˜ 3
S = (ξ1 , ξ2, ν) ∈ R : ξi ∈ Iki , ν satisfies (3.45) , (3.49)
√
H1 (ξ1 , ξ2 , ν) = 5(ξ13 − ξ23 + νξ12 ξ2 ), (3.50)
5(ξ1 ξ2 )2
ξ1 ξ2 ν 2 + 2ν(ξ12 + ξ22 ) + 4ξ1 ξ2 − 3ξ 2 .


H2 (ξ1 , ξ2 , ν) = (3.51)
ξ

We consider ξ1 ξ2 ≥ 0, ξ1ξ2 ≤ 0, respectively.

Firstly, we consider ξ1 ξ2 ≥ 0. We define the function
√  3  5(ξ1 ξ2 )2   2
  
−k3 +2k1 +2k2  3 2

G(ξ, x, y) = 2 h ξ, 5 ξ1 − ξ2 + x , y + 4ξ1 ξ2 − 3ξ  , (3.52)
ξ 

20
where

x = ξ12 ξ2 ν, y = ξ1 ξ2 ν 2 + 2(ξ12 + ξ22)ν. (3.53)

Obviously, kGkL1 = khk2L2 . From (3.52), we have (3.48) can be bounded by

Z  12
k1 +k2 −k3
−
C2 2 |G(ξ1 + ξ2 , ξ12 ξ2 ν, ξ1 ξ2 ν 2 + 2(ξ12 + ξ22 )ν)|dξ1 dξ2 dν . (3.54)
S ′′

We make the change of variables (ξ1 , ξ2 , ν) −→ (ξ1 + ξ2 , ξ12ξ2 ν, ξ1 ξ2 ν 2 + 2(ξ12 + ξ22)ν), thus
the absolute value of the Jacobi determinant equals

|νξ1 | νξ1 ξ2 (ξ1 − 3ξ2 ) + 2 ξ13 − ξ1 ξ22 − 2ξ23  .



(3.55)

By using a direct computation, we have (3.55) equals

3 2 3 
 
2
 2(ξ 1 − ξ 1 ξ 2 − 2ξ 2 )
|νξ1 ξ2 (ξ1 − 3ξ2 )| ν + , (3.56)
ξ1 ξ2 (ξ1 − 3ξ2 ) 

where ν satisfies (3.47).

By using a direct computation, we have
 3
 2(ξ1 − ξ1 ξ22 − 2ξ23) 

 ξ1 ξ2 (ξ1 − 3ξ2 )  ≥ 2. (3.57)
 

From (3.47), we have

 
2 2
 3ξ + 2ξ ξ + 3ξ  3
1 − 2−20 ≤ |ν| ≤  2 1
p1 2 2
+ 2 −20
≤ + 2−20 . (3.58)
 
 ξ1 + ξ22 + ξ 2 (ξ 2 − ξ1 ξ2 )  2


Combining (3.57) with (3.58), we have

2(ξ13 − ξ1 ξ22 − 2ξ23 )  1

 

|ν| ν +
 ≥ . (3.59)
ξ1 ξ2 (ξ1 − 3ξ2 )  4

Combining (3.56) with (3.59), we have

|νξ1 | νξ1 ξ2 (ξ1 − 3ξ2 ) + 2 ξ13 − ξ1 ξ22 − 2ξ23  ≥ Cξ12 ξ22 .



(3.60)

Combining (3.54) with (3.60), we have (3.45) can be bounded by

3(k1 +k2 ) k 3(k1 +k2 ) k

+ 23 + 23
C2− 2 kGkL1 ≤ C2− 2 khk2L2 . (3.61)

21
 Now we consider ξ1 ξ2 ≤ 0. We define
√  3  5(ξ1ξ2 )2   2
  
−k3 +7k1  3 2 2

G(ξ, x, y) = 2 h ξ, 5 ξ1 − ξ2 + ξ1 ξ2 x , y + 4ξ1 ξ2 − 3ξ  , (3.62)
ξ 

where

x = ν, y = ξ1 ξ2 ν 2 + 2(ξ12 + ξ22 )ν. (3.63)

Obviously, kGkL1 ≈ khk2L2 . From (3.62), we have (3.48) can be bounded by

Z  21
5k1 −k3
− 2
C2 2 |G(ξ1 + ξ2 , ν, ξ1 ξ2 ν + 2(ξ12 + ξ22 )ν)|dξ1dξ2 dν . (3.64)
S ′′

We make the change of variables (ξ1 , ξ2 , ν) −→ (ξ1 + ξ2 , ν, ξ1 ξ2 ν 2 + 2(ξ12 + ξ22 )ν), thus the
absolute value of the Jacobi determinant equals

|ξ1 − ξ2 ||ν(4 − ν)|. (3.65)

From (3.47), we have

3ξ12 + 2ξ1 ξ2 + 3ξ22

1 − 2−20 ≤ 2 2
p − 2−20
2 2
ξ1 + ξ2 + ξ (ξ − ξ1 ξ2 )
3ξ12 + 2ξ1 ξ2 + 3ξ22
≤ |ν| ≤ 2 2
p + 2−20 ≤ 3 + 2−20 . (3.66)
2 2
ξ1 + ξ2 + ξ (ξ − ξ1 ξ2 )

Combining (3.65) with (3.66), we have

|ξ1 − ξ2 ||ν(4 − ν)| ∼ 2k1 . (3.67)

Combining (3.64) with (3.67), we have (3.44) can be bounded by

3(k1 +k2 ) k 3(k1 +k2 ) k
+ 23 + 23
C2− 2 kGkL1 ≤ C2− 2 khk2L2 . (3.68)

Therefore the proof of (3.3) is completed.

Now we prove (3.4). From (3.5)-(3.17), we know that it suffices to prove
Z
k1 +k2
2 2 h1 (ξ1 , β1 )h2 (ξ2 , β2 )
S̃
√
× h(ξ1 + ξ2 , 5(ξ13 − ξ23 ) + β1 ξ1 + β2 ξ2 , R̃1 (ξ1 , β1 , ξ2 , β2 )dξ1 dξ2dβ1 dβ2
2
j k1 5k
− 22
Y
−
≤C2 2 2 2 khkL2 khj kL2 . (3.69)
j=1

22
where hi (i = 1, 2) are defined as in (3.16)-(3.17) and S is defined as in (3.14) and
R̃1 (ξ1 , β1 , ξ2, β2 ) is defined as in (3.15) and β1 − β2 satisfies (3.19)-(3.20). To obtain
(3.69), we make the change of variable β1 = β2 + β. Now we claim that the following
inequality is valid
 √ 
 5B1 
β − 2  ≤ B4 , (3.70)
 
p
2
 ξ1 + ξ2 + ξ 2(ξ 2 − ξ1 ξ2 ) 

where B1 is defined as in (3.21) and

B4 = 2j−3k2+10 . (3.71)

(3.70) can be proved similarly to (3.21). Since |ξj | ∈ [2kj −1 , 2kj +1 ](j = 1, 2), we can
1
assume that ξj = aj 2kj (j = 1, 2), where 2
≤ |aj | ≤ 2. Consequently, (3.70) can be
rewritten as follows:
 √ 
β − 5f1 (a1 , a2 , k1 , k2) ≤ B4 , (3.72)
 

where f1 (a1 , a2 , k1 , k2 ) is defined as in (3.33).

Thus, the left hand side of (3.68) can be bounded by
 √ 
β2 − 5f1 (a1 , a2 , k1 , k2)
Z
k1 +k2
2 2 h1 (ξ1 , β + β2 )h2 (ξ2 , β2 )χ[−1,1) −m
S̃ B4
× h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))dξ1dξ2 dβdβ2, (3.73)

where S̃ satisfies (3.35) and A(ξ1 , ξ2 , β) satisfies (3.36) and B(ξ1 , ξ2, β) satisfies (3.37).
Let j ′′ = j − 3k2 + 10. Decompose
√
β′ −
 
′ ′
X
′ ′ 5f1 (a1 , a2 , k1 , k2 )
hi (ξ , β ) = hi (ξ , β )χ[0,1) −m
m∈Z
2j ′′
X
′ ′
= hm
i (ξ , β ), (3.74)
m∈Z

  21
2
khm
P
i = 1, 2. Obviously, khi kL2 = i k L2 . Thus, (3.73) is controlled by
m∈Z
Z
k1 +k2 X ′
2 2 hm m
1 (ξ1 , β + β2 )h2 (ξ2 , β2 )
|m−m′ |≤4 S̃

× h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))dξ1dξ2 dβdβ2. (3.75)

23
To prove (3.69), it suffices to prove that
√ !
β2 − 5f1 (a1 , a2 , k1 , k2)
Z
k1 +k2
m′
2 2 hm
1 (ξ1 , β + β2 )h2 (ξ2 , β2 )χ[m′ ,m′ +1)
S̃ 2j ′′
× h(ξ1 + ξ2 , A(ξ1 , ξ2, β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2 , β))dξ1dξ2 dβdβ2
j k1 5k2 ′
≤C2 2 2− 2 − 2 khkL2 khm m
1 kL2 kh2 kL2 . (3.76)

From (3.76), it suffices to prove that

Z  12
2
|h(ξ1 + ξ2 , A(ξ1 , ξ2 , β) + β2 (ξ1 + ξ2 ), B(ξ1 , ξ2, β))| dξ1dξ2 dβ
S′
3k2
≤C2− 2
−k1
khkL2 . (3.77)

If (3.77) is valid, by using the Cauchy-Schwartz inequality with respect to β2 , we have

(3.75) is controlled by
√ !
β2 − 5f1 (a1 , a2 , k1 , k2 )
Z
k
− 21 −k2
C2 χ[m′ ,m′ +1)
R 2j ′′
′
× khm m
1 kL2 kh2 (·, β2 )kL2ξ khkL2 dβ2
2
′′ √
k1
Z (m′ +1)2j − 5f1 (a1 ,a2 ,k1 ,k2 )
′
= C2− 2 −k2 √
khm m
1 kL2 kh2 (·, β2 )kL2ξ khkL2 dβ2
m′ 2j ′′ − 5f1 (a1 ,a2 ,k1 ,k2 ) 2

k1 j ′′ ′
≤ C2− 2 −k2 2 2 khkL2 khm m
1 kL2 kh2 kL2
5k2 k j ′
− 21
≤ C2− 2 2 2 khkL2 khm m
1 kL2 kh2 kL2 . (3.78)

To prove (3.77), we may assume that β2 = 0 and make the change of variable β =
√
5ξ1 ξ2 ν. The left hand side of (3.77) is controlled by
Z  21
k1 +k2
2
C2 2 |h(ξ1 + ξ2 , H1 (ξ1 , ξ2 , ν), H2 (ξ1 , ξ2 , ν))| dξ1 dξ2dν , (3.79)
S ′′

where S ′′ is defined as in (3.49) and H1 (ξ1 , ξ2 , ν) is defined as in (3.50) and H2 (ξ1 , ξ2 , ν)

is defined as in (3.51).
We define the function G(ξ, x, y) as in (3.52) and x, y as in (3.53). Obviously, kGkL1 ≈
khk2L2 . Thus, we have (3.79) can be bounded by
Z  12
k
− 21
C2 |G(ξ1 + ξ2 , ξ12 ξ2 ν, ξ1ξ2 ν 2 + 2(ξ12 + ξ22 )ν)|dξ1 dξ2 dν . (3.80)
S ′′

24
We make the change of variables (ξ1 , ξ2 , ν) −→ (ξ1 + ξ2 , ξ12ξ2 ν, ξ1 ξ2 ν 2 + 2(ξ12 + ξ22)ν), thus
the absolute value of the Jacobi determinant equals
 
3 2 3

 ξ1 ξ2 (ξ1 − 3ξ2 )ν 
2|νξ1 ξ1 − ξ1 ξ2 − 2ξ2 |  3 2 3
+ 1 . (3.81)
2 (ξ1 − ξ1 ξ2 − 2ξ2 )

In this case, by using a direct computation, we have

 
 ξ1 ξ2 (ξ1 − 3ξ2)ν  1
 2 (ξ 3 − ξ1 ξ 2 − 2ξ 3)  ≤ 32 . (3.82)
 
1 2 2

From (3.43), we have

 
2 2
−20
 3ξ1 + 2ξ1 ξ2 + 3ξ2 
1−2 ≤ 2  − 2−20
 
p
2 2 2
 ξ1 + ξ2 + ξ (ξ − ξ1 ξ2 ) 
 
2 2
 3ξ1 + 2ξ1 ξ2 + 3ξ2 
≤ |ν| ≤  2  + 2−20 ≤ 3 + 2−20 . (3.83)
 
p
2 2 2
 ξ1 + ξ2 + ξ (ξ − ξ1 ξ2 ) 

Combining (3.82) with (3.83), we have

 

 ξ1 ξ2 (ξ1 − 3ξ2)ν
ξ13 ξ1 ξ22 2ξ23

|νξ1 − − |

3 2 3
+ 1
2 (ξ1 − ξ1 ξ2 − 2ξ2 )
∼ |ξ1 ξ1 − ξ1 ξ2 − 2ξ2 | ∼ 2k1 +3k2 .
3 2 3


(3.84)

Combining (3.84) with the fact that kGkL1 ≈ khk2L2 , we have (3.78) can be bounded by

3k2 3k2
C2− 2
−k1
kGkL1 ≤ C2− 2
−k1
khk2L2 . (3.85)

This completes the proof of Lemma 3.1.

Inspired by the idea of Lemma 5.2 of [18], we give the proof of Lemma 3.2.

Lemma 3.2. Assume α ∈ R and k1 , k2 , k3 ∈ Z, kmax = max {k1 , k2, k3 } ≥ 20 and

kmin = min {k1 , k2 , k3 } and j1 , j2 , j3 ∈ Z+ , jmax = max {j1 , j2 , j3 } and fi : R3 → R are
L2 functions supported in Dki ,∞,ji (i = 1, 2, 3). Then, kmax ≥ 20, we have
Z 3
j1 +j2 +j3 kmin
− jmax − kmax
Y
−
(f1 ∗ f2 )f3 dξdµdτ ≤ C2 2 2 2 4 4 kfj kL2 . (3.86)
R3 j=1

Proof. From (3.79), we assume that j3 = max {j1 , j2 , j3 } . Then, we have

Z 2
Y
(f1 ∗ f2 )f3 dξdµdτ ≤ Ckf3 kL2 kf1 ∗ f2 kL2 ≤ Ckf3 kL2 kF −1 (fj )kL4 . (3.87)
3
R j=1

25
From Theorem 3.1 of [15], we have
Z
1
ixξ+iyµ itφ(ξ,µ)
|ξ| 4 f (ξ, µ)e
j e dξdµ ≤ Ckfj kL2ξµ (j = 1, 2). (3.88)
R2

L4xyt

From (3.88), by using the Cauchy-Schwartz inequality with respect to θ, we have

Z
1
ixξ+iyµ itτ

3 |ξ j | 4 f (ξ, µ, τ )e
j e dξdµdτ
4
R Lxyt
Z
1 #
ixξ+iyµ itθ itφ(ξ,µ)

= |ξj | 4 fj (ξ, µ, θ)e e e dξdµdθ
R3

L4xyt
Z
ji
≤C kfj# (ξ, µ, θ)kL2ξµ dθ ≤ C2 2 kfj# (ξ, µ, θ)kL2ξµθ
R
ji
=C2 kfj (ξ, µ, τ )kL2ξµτ .
2 (3.89)

Here fj# (j = 1, 2) are defined as in Lemma 3.1. From (3.89), we have

ki ji
kF −1 (fj )kL4 ≤ C2− 4 2 2 kfj (ξ, µ, τ )kL2ξµτ . (3.90)

Inserting (3.90) into (3.87) yields

Z
(f1 ∗ f2 )f3 ≤ Ckf3 kL2 kf1 ∗ f2 kL2
R3
2
j1 +j2 kmax +kmin Y
−
≤ C2 2 2 4 kf kL2 kF −1 (fj )kL4 . (3.91)
j=1

Combining the fact with j3 = max {j1 , j2 , j3 } with (3.91), we have (3.86) is valid.
This completes the proof of Lemma 3.2.

4. Bilinear estimates
This section is devoted to establishing Lemmas 4.1-4.3. Lemma 4.1 is used to establish
Theorems 1.1. Lemma 4.2 is used to establish the almost conservation. Lemma 4.3 is
used to establish Lemma 6.1 which is the variant of Theorem 1.1.

Lemma 4.1. Let − 89 + 16ǫ ≤ s1 < 0, s2 ≥ 0 and uj ∈ X s11+ǫ

,s2
(j = 1, 2). Then, we have
2

2
Y
k∂x (u1 u2 )kX s11,s2 ≤ C kuj kX s11 ,s2 . (4.1)
− 2 +2ǫ 2 +ǫ
j=1

26
Proof. To derive (4.1), by duality, it suffices to show that
Z  2
Y
 
 ū∂x (u1 u2 )dxdydt ≤ CkukX −s1 ,−s2
 kuj kX s11 ,s2 . (4.2)
1 −2ǫ
R3 2 +ǫ

2 j=1

for u ∈ X −s
1
1 ,−s2
−2ǫ
. Let
2

1
F (ξ, µ, τ ) = hξi−s1 hµi−s2 hσi 2 −2ǫ F u(ξ, µ, τ ),
1
Fj (ξj , µj , τj ) = hξj is1 hµj is2 hσj i 2 +ǫ F uj (ξj , µ, τj )(j = 1, 2), (4.3)

and
( 2 2 2
)
X X X
D := (ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ R6 , ξ = ξj , µ = µj , τ = τj .
j=1 j=1 j=1

To derive (4.2), from (4.3), it suffices to show

2
Z |ξ|hξis1 hµis2 F (ξ, µ, τ )
Q
Fj (ξj , µj , τj ) 2
j=1 Y
2
dV ≤ CkF kL2τ ξµ kFj kL2τ ξµ , (4.4)
D 1 1
−2ǫ +ǫ
Q
hσj i 2 hξj is1 hµj is2 hσj i 2 j=1
j=1

where dV := dξ1 dµ1 dτ1 dξdµdτ . Without loss of generality, by using the symmetry, we
assume that |ξ1 | ≥ |ξ2| and F (ξ, µ, τ ) ≥ 0, Fj (ξj , µj , τj ) ≥ 0(j = 1, 2) and

D ∗ := {(ξ1, µ1 , τ1 , ξ, µ, τ ) ∈ D, |ξ2| ≥ |ξ1 |} .

Let

Ω1 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ2| ≤ |ξ1 | ≤ 80} ,

Ω2 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1| ≥ 80, |ξ1| ≫ |ξ2|, |ξ2| ≤ 20},

Ω3 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1| ≥ 80, |ξ1| ≫ |ξ2|, |ξ2| > 20},

Ω4 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1| ≥ 80, 4|ξ| ≤ |ξ1| ∼ |ξ2 |, |ξ| ≤ 20, ξ1ξ2 < 0},

Ω5 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1| ≥ 80, 4|ξ| ≤ |ξ1| ∼ |ξ2 |, |ξ| > 20, ξ1ξ2 < 0},
|ξ2 |
Ω6 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1| ≥ 80, |ξ2| ∼ |ξ1 |, ξ1 ξ2 < 0, |ξ| ≥ },
4
Ω7 = {(ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1| ≥ 80, |ξ2| ∼ |ξ1 |, ξ1 ξ2 > 0} .
7
Obviously, D ∗ ⊂
S
Ωj . We define
j=1

|ξ|hµis2 hξis1
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) := 2
(4.5)
1 1
−2ǫ
hξj is1 hµj is2 hσj i 2 +ǫ
Q
hσj i 2
j=1

27
and
Z 2
Y
Intj := K1 (ξ1 , µ1 , τ1 , ξ, µ, τ )F (ξ, µ, τ ) Fj (ξj , µj , τj )dξ1 dµ1 dτ1 dξdµdτ,
Ωj j=1

2 2
µj , we have hµis2 ≤ hµj is2 , thus, we have
P Q
1 ≤ j ≤ 7, j ∈ N. Since s2 ≥ 0 and µ =
j=1 j=1

|ξ|hξis1
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
. (4.6)
1 1
−2ǫ +ǫ
Q s
hσi 2 hξj i hσj i
1 2
j=1

Now we estimate the integrals Intj over the above seven regions one by one.
(I) Region Ω1 . In this region |ξ| ≤ |ξ1 | + |ξ2| ≤ 160, this case can be proved similarly
to case low + low −→ low of Pages 344–345 of Theorem 3.1 in [38].
(II) Region Ω2 . In this region, we have |ξ| ∼ |ξ1|.
By using the Cauchy-Schwartz inequality with respect to ξ1 , µ1 , τ1 , from (4.6), we
have
Z  12
|ξ|
Z
−(1+2ǫ) −(1+2ǫ)
Int2 ≤ C 1 hσ1 i hσ2 i dξ1 dµ1 dτ1
R3 hσi 2 −2ǫ R3
Z Y 2
! 21
× |Fj (ξj , µj , τj )|2 dξ1dµ1 dτ1 F (ξ, µ, τ )dξdµdτ. (4.7)
R3 j=1

By using (2.2), we have

Z  12
|ξ| −(1+2ǫ) −(1+2ǫ)
1 hσ1 i hσ2 i dξ1 dµ1 dτ1
hσi 2 −2ǫ R3
Z  21
|ξ| dξ1 dµ1
≤C 1 . (4.8)
hσi 2 −2ǫ R2 hτ + φ(ξ1 , µ1 ) + φ(ξ2 , µ2)i1+2ǫ

Let ν = τ + φ(ξ1, µ1 ) + φ(ξ2 , µ2 ) and ∆ = ξξ1ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 ), since |ξ1 | ≫ |ξ2 |, then
we have the absolute value of Jacobian determinant equals
   
 ∂(∆, ν) 
 =2  µ1 − µ2  5(ξ14 − ξ24 ) − 3α(ξ12 − ξ22 )
  

 ∂(ξ1 , µ1 )   ξ1 ξ2 
 1
1  ξ 2 
 5(ξ 4 − ξ 4 ) − 3α(ξ 2 − ξ 2 )

=2 |σ − ν − ∆| 2  
1 2 1 2
ξ1 ξ2 
 1
1  ξ 2
∼ |σ − ν + δ| 2   |ξ1 |4 . (4.9)
ξ1 ξ2 

28
Inserting (4.9) into (4.8), by using (2.3), we have
Z  21
|ξ| −(1+2ǫ) −(1+2ǫ)
1 hσ1 i hσ2 i dξ1 dµ1 dτ1
hσi 2 −2ǫ R3
Z  21
|ξ| dξ1 dµ1
≤C 1 1+2ǫ
hσi 2 −2ǫ R2 hτ + φ(ξ1 , µ1 ) + φ(ξ2 , µ2 )i
! 21
C dνd∆
Z
≤ 1 1
|ξ|hσi 2 −2ǫ R2 |σ − ν − ∆| 2 hνi1+2ǫ
! 21
C d∆
Z
≤ 1 1 . (4.10)
|ξ|hσi 2 −2ǫ |∆|<20|ξ|4 h∆ − σi 2

When |σ| < 20|ξ|4, combining (4.10) with (2.1), we have

! 12
C d∆ |ξ|
Z
1
−2ǫ 1 ≤C 1 ≤ C. (4.11)
|ξ|hσi 2 |∆|<20|ξ|4 h∆ − σi 2 |ξ|hσi 2 −2ǫ
When |σ| ≥ 20|ξ|4, from (4.10), we have
! 12
C d∆ |ξ|2 |ξ|
Z
1
−2ǫ 1 ≤C 1
−2ǫ
≤C 1 ≤ C. (4.12)
|ξ|hσi 2 |∆|<20|ξ|4 h∆ − σi 2 |ξ|hσi 2 hσi 2 −2ǫ
Combining (4.8) with (4.9)-(4.12), we have
Z  12
|ξ| −(1+2ǫ) −(1+2ǫ)
1 hσ1 i hσ2 i dξ1dµ1 dτ1 ≤ C. (4.13)
hσi 2 −2ǫ R3
Inserting (4.13) into (4.7), by using the Cauchy-Schwartz inequality with respect to
ξ, µ, τ , we have
Z  21
|ξ|
Z
−(1+2ǫ) −(1+2ǫ)
Int2 ≤ C 1 hσ1 i hσ2 i dξ1 dµ1dτ1
R3 hσi 2 −2ǫ R3
Z Y 2
! 21
× |Fj (ξj , µj , τj )|2 dξ1 dµ1 dτ1 F (ξ, µ, τ )dξdµdτ
R3 j=1

Z Z 2
! 12
Y
≤C |Fj (ξj , µj , τj )|2 dξ1 dµ1 dτ1 F (ξ, µ, τ )dξdµdτ
3 3
R R j=1

≤ CkF kL2τ ξµ prod2j=1kFj kL2τ ξµ . (4.14)

(III) Region Ω3 . In this region, we have |ξ| ∼ |ξ1|. In this region, we consider
  2 
 ξ 1 ξ 2 µ 1 µ 2
|σ − σ1 − σ2 | = ξξ1ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 ) −
 
  −  
 ξ  ξ1 ξ2  
|ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 )|
≥ (4.15)
270

29
and
  2 
 ξ 1 ξ 2 µ 1 µ 2
|σ − σ1 − σ2 | = ξξ1ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 ) −
 
  −  
 ξ  ξ1 ξ2  
|ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12)|
< , (4.16)
270
respectively.
When (4.15) is valid, we have one of the following three cases must occur:

|σ| := max {|σ|, |σ1|, |σ2 |} ≥ C ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12) ,

 
(4.17)

|σ1 | := max {|σ|, |σ1|, |σ2 |} ≥ C ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 ) ,

 
(4.18)

|σ2 | := max {|σ|, |σ1|, |σ2 |} ≥ C ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 ) .

 
(4.19)

When (4.17) is valid, since − 98 + 16ǫ ≤ s1 < 0, we have

1
|ξ|hξis1 |ξ|−1+8ǫ |ξ2|−s1 − 2 +2ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
≤ C 2
1 1 1
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ hσj i 2 +ǫ
Q Q
j=1 j=1
5
−14ǫ − 12
|ξ|−1+8ǫ |ξ2| 8 |ξ1 ||ξ2|
≤C 2
≤C 2
. (4.20)
1 1
hσj i 2 +ǫ +ǫ
Q Q
hσj i 2
j=1 j=1

Thus, combining (2.17) with (4.20), we have

2
Y
|Int3 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

When (4.18) is valid, since − 98 + 16ǫ ≤ s1 < 0 and

1 1 1 1
hσi− 2 +2ǫ hσ1 i− 2 −ǫ ≤ hσi− 2 −ǫ hσ1 i− 2 +2ǫ ,

we have
|ξ|hξis1
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
1 1
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ
Q
j=1
1 1
|ξ|−1+8ǫ|ξ2 |−s1 − 2 +2ǫ |ξ|− 2 |ξ2 |
≤C 1 1 ≤C 1 1 . (4.21)
hσi 2 +ǫ hσ2 i 2 +ǫ hσi 2 +ǫ hσ2 i 2 +ǫ
Thus, combining (2.19) with (4.21), we have
2
Y
|Int3 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

30
 When (4.19) is valid, since − 89 + 16ǫ ≤ s1 < 0 and

1 1 1 1
hσi− 2 +2ǫ hσ2 i− 2 −ǫ ≤ hσi− 2 −ǫ hσ2 i− 2 +2ǫ ,

we have
1
|ξ|hξis1 |ξ|−1+8ǫ |ξ1|−s1 − 2 +2ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
≤ C 1 1
1 1 hσ1 i 2 +ǫ hσi 2 +ǫ
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ
Q
j=1
1 1
−14ǫ
|ξ1 | 4 |ξ|−1+8ǫ |ξ1 |− 2 |ξ|
≤C 1 1 ≤C 1 1 . (4.22)
hσ1 i 2 +ǫ hσi 2 +ǫ hσ1 i 2 +ǫ hσi 2 +ǫ

Thus, combining (2.20) with (4.22), we have

2
Y
|Int3 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

When (4.16) is valid, we consider the following two cases respectively.

|ξξ1 ξ2 (5ξ 2 − 5ξξ1 + 5ξ12)|

max {|σ|, |σ1 |, |σ2 |} ≥ , (4.23)
280
|ξξ1ξ2 (5ξ 2 − 5ξξ1 + 5ξ12 )|
max {|σ|, |σ1 |, |σ2 |} < . (4.24)
280

We dyadically decompose the spectra as

hσi ∼ 2j , hσ1 i ∼ 2j1 , hσ2 i ∼ 2j2 , |ξ| ∼ 2k , |ξ1 | ∼ 2k1 , |ξ2| ∼ 2k2 .

We define

fkm ,jm := |ηkm (ξm )ηjm (σm )Fl (ξm , µm , τm )| , (m = 1, 2),

fk,j := |ηk (ξ)ηj (σ)F (ξ, µ, τ )| .

When (4.23) is valid, by using Lemma 3.2, since − 89 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 21 +ǫ)−k2 s1 +k
X X Y
Int3 ≤ C 2 fk,j fkm ,jm dV
k,k1 ,k2 >0 j1 ,j2 ,j≥0 R6 m=1
2
jmax
−k2 (s1 + 14 )+ 3k
X X Y
≤C 22jǫ−(j1+j2 )ǫ− 2 4 kfk,j kL2 kfkm ,jm kL2 ; (4.25)
k,k1 ,k2 >0 j1 ,j2 ,j≥0 m=1

31
when j = jmax , from (4.25), since − 89 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )ǫ−k2 (s1 + 14 )+ 3k
X X Y
Int3 ≤ C 2 4 fk,j fkm ,jm dV
k,k1 ,k2 >0 j1 ,j2 ,j≥0 R6 m=1
2
3 5
X Y
≤C 2−k2 (s1 + 4 −2ǫ)−k( 4 −8ǫ) kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
−k2 (s1 + 89 )−k( 87 −10ǫ)
X Y
≤C 2 kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
Y
≤ Ckf kL2 kfm kL2 ; (4.26)
m=1

when j1 = jmax , from (4.25), since − 98 + 16ǫ ≤ s1 < 0, we have

2
2jǫ−(j1 +j2 )ǫ− jmax −k2 (s1 + 41 )+ 3k
X X Y
Int3 ≤ C 2 2 4 kfk,j kL2 kfkm ,jm kL2
k,k1 ,k2 >0 j1 ,j2 ,j≥0 m=1
2
3 5
X Y
≤C 2−k2 (s1 + 4 −ǫ)−k( 4 −ǫ) kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
−k2 (s1 + 98 )−k( 87 −5ǫ)
X Y
≤C 2 kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
Y
≤ Ckf kL2 kfm kL2 ; (4.27)
m=1

when j2 = jmax , this case can be proved similarly to case j1 = jmax .

When (4.24) is valid, by using (2) of Lemma 3.1, since − 89 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)−k2 s1 +k
X X Y
Int3 ≤ C 2 fk,j fkm ,jm dV
k,k1 ,k2 >0 j1 ,j2 ,j≥0 R6 m=1
2
1 3
X X Y
≤C 22jǫ−(j1+j2 )ǫ−k2 (s1 + 2 )− 2 k kfk,j kL2 kfkm ,jm kL2
k,k1 ,k2 >0 j1 ,j2 ,j≥0 m=1
2
9 3k
X Y
≤C 2−k2 (s1 + 8 )− 8 kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
Y
≤ Ckf kL2 kfm kL2 . (4.28)
m=1

(IV) Region Ω4 . In this case, we consider (4.15), (4.16), respectively.

When (4.15) is valid, one of (4.17)-(4.19) must occur.

32
 When (4.17) is valid, since − 89 + 16ǫ ≤ s1 < 0, we have
1
|ξ|hξis1 |ξ| 2 +2ǫ |ξ2|−2s1 −2+8ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
≤ C 2
1 1 1
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ hσj i 2 +ǫ
Q Q
j=1 j=1
1
+2ǫ − 12 −24ǫ − 21
|ξ| 2 |ξ2| |ξ1 | |ξ2 |
≤C 2
≤C 2
. (4.29)
1 1
+ǫ +ǫ
Q Q
hσj i 2 hσj i 2
j=1 j=1

Thus, combining (2.17) with (4.29), we have

2
Y
|Int4 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

When (4.18) is valid, since − 98 + 16ǫ ≤ s1 < 0 and

1 1 1 1
hσi− 2 +2ǫ hσ1 i− 2 −ǫ ≤ hσi− 2 −ǫ hσ1 i− 2 +2ǫ ,

we have
1
|ξ|hξis1 |ξ| 2 +2ǫ |ξ2|−2s1 −2+8ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
≤ C 1
+ǫ 1
+ǫ
1 1 hσi 2 hσ i 2
hσi 2 −2ǫ +ǫ 2
Q s
hξj i hσj i 2
1

j=1
1 1 1
+2ǫ
|ξ| 2 |ξ2|− 2 −24ǫ |ξ|− 2 |ξ2 |
≤C 1 1 ≤C 1 1 . (4.30)
hσi 2 +ǫ hσ2 i 2 +ǫ hσi 2 +ǫ hσ2 i 2 +ǫ

Thus, combining (2.19) with (4.30), we have

2
Y
|Int4 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

When (4.19) is valid, this case can be proved similarly to (4.18).

When (4.16) is valid, we consider (4.23) and (4.24), respectively.
We dyadically decompose the spectra as

hσi ∼ 2j , hσ1 i ∼ 2j1 , hσ2 i ∼ 2j2 , |ξ| ∼ 2k , |ξ1 | ∼ 2k1 , |ξ2| ∼ 2k2 .

We define

fkm ,jm := |ηkm (ξm )ηjm (σm )Fl (ξm , µm , τm )| , (m = 1, 2),

fk,j := |ηk (ξ)ηj (σ)F (ξ, µ, τ )| , dV = dξ1 dµ1dτ1 dξdµdτ.

33
When (4.23) is valid, by Lemma 3.2, since − 98 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)−2k2 s1 +k
X X Y
Int4 ≤ C 2 fk,j fkm ,jm dV
k1 ,k2 >0, k j1 ,j2 ,j≥0 R6 m=1
2
jmax
−k2 (2s1 + 41 )+ 3k
X X Y
≤C 22jǫ−(j1 +j2 )ǫ− 2 4 kfk,j kL2 kfkm ,jm kL2 ; (4.31)
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1

9
when j = jmax , from (4.31), since 8
+ 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)−2k2 s1 +k
X X Y
Int4 ≤ C 2 fk,j fkm ,jm dV
k1 ,k2 >0, k j1 ,j2 ,j≥0 R6 m=1
2
9 1
X Y
≤C 2−k2 (2s1 + 4 −8ǫ)+k( 4 +2ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ Ckf kL2 kfm kL2 ; (4.32)
m=1

when j1 = jmax , from (4.31), since − 98 + 16ǫ ≤ s1 < 0, we have

2
2jǫ−(j1 +j2 )ǫ− jmax −k2 (2s1 + 41 )+ 3k
X X Y
Int4 ≤ C 2 2 4 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
9 1
X Y
≤C 2−k2 (2s1 + 4 −4ǫ)+k( 4 +ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ Ckf kL2 kfm kL2 ; (4.33)
m=1

when j2 = jmax , this case can be proved similarly to case j1 = jmax .

When (4.24) is valid, by using (1) of Lemma 3.1, since − 89 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)−2k2 s1 +k
X X Y
Int4 ≤ C 2 fk,j fkm ,jm dV
k1 ,k2 >0, k j1 ,j2 ,j≥0 R6 m=1
2
2jǫ−(j1 +j2 )ǫ−k2 (2s1 + 27 )+ 3k
X X Y
≤C 2 2 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
7 3
X Y
≤C 2−k2 (2s1 + 2 −8ǫ)+k( 2 +ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ Ckf kL2 kfm kL2 . (4.34)
m=1

(V) Region Ω5 . In this region, we consider (4.15), (4.16), respectively.

34
 When (4.15) is valid, one of (4.17)-(4.19) must occur.
When (4.17) is valid, since − 89 + 16ǫ ≤ s1 < 0, we have

|ξ|hξis1
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
1 1
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ
Q
j=1
s1 + 12 −2ǫ 5 1 1
|ξ| |ξ2 | −2s1 −2+8ǫ
|ξ|− 8 +14ǫ |ξ2 | 4 −24ǫ |ξ1 |− 2 |ξ2 |
≤C 2
≤C 2
≤ C 2
. (4.35)
1 1 1
+ǫ +ǫ +ǫ
Q Q Q
hσj i 2 hσj i 2 hσj i 2
j=1 j=1 j=1

Thus, combining (2.17) with (4.35), we have

2
Y
|Int5 | ≤ CkF k L2τ ξµ kFj kL2τ ξµ .
j=1

When (4.18) is valid, since − 98 + 16ǫ ≤ s1 < 0 and

1 1 1 1
hσi− 2 +2ǫ hσ1 i− 2 −ǫ ≤ hσi− 2 −ǫ hσ1 i− 2 +2ǫ ,

we have
1
|ξ|hξis1 |ξ|s1+ 2 −2ǫ |ξ2|−2s1 −2+8ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ 2
≤ C 1 1
1 1 hσ2 i 2 +ǫ hσi 2 +ǫ
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ
Q
j=1
− 85 +14ǫ 1 1
|ξ| |ξ2 | 4 −24ǫ |ξ|− 2 |ξ2 |
≤C 1 1 ≤C 1 1 . (4.36)
hσ2 i 2 +ǫ hσi 2 +ǫ hσi 2 +ǫ hσ2 i 2 +ǫ
Thus, combining (2.19) with (4.36), we have
2
Y
|Int5 | ≤ CkF k L2τ ξµ kFj kL2τ ξµ .
j=1

When (4.19) is valid, this case can be proved similarly to (4.18) with the aid of (2.20).
When (4.16) is valid, consider (4.23), (4.24), respectively.
We dyadically decompose the spectra as

hσi ∼ 2j , hσ1 i ∼ 2j1 , hσ2 i ∼ 2j2 , |ξ| ∼ 2k , |ξ1| ∼ 2k1 , |ξ2 | ∼ 2k2 .

We define

fkm ,jm := |ηkm (ξm )ηjm (σm )Fl (ξm , µm , τm )| (m = 1, 2),

fk,j := |ηk (ξ)ηj (σ)F (ξ, µ, τ )| , dV = dξ1 dµ1 dτ1 dξdµdτ.

35
 When (4.23) is valid, we use Lemma 3.2 to deal with this case. Thus, by Lemma 3.2,
since − 98 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)−2k2 s1 +k(1+s1 )
X X Y
Int5 ≤ C 2 fk,j fkm ,jm dV
k1 ,k2 >0, k j1 ,j2 ,j≥0 R6 m=1
2
2jǫ−(j1 +j2 )ǫ− jmax −k2 (2s1 + 41 )+k ( 43 +s1 )
X X Y
≤C 2 2 kfk,j kL2 kfkm ,jm kL2 .
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
(4.37)
9
When j = jmax , from (4.37), if 8
+ 16ǫ ≤ s1 < − 43 , we have
2
jmax
−k2 (2s1 + 41 )+k ( 34 +s1 ) kf k 2
X X Y
Int5 ≤ C 22jǫ−(j1 +j2 )ǫ− 2
k,j L kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
9 1
X Y
≤C 2−k2 (2s1 + 4 −8ǫ)+k( 4 +2ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ Ckf k L2 kfm kL2 . (4.38)
m=1

When j = jmax , from (4.37), if − 43 ≤ s1 < 0, we have

2
2jǫ−(j1 +j2 )ǫ− jmax −k2 (2s1 + 41 )+k ( 34 +s1 )
X X Y
Int5 ≤ C 2 2 kfk,j k L2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
3
X Y
≤C 2−k2 (s1 + 2 −8ǫ) kf kL2 kfm kL2
k1 ,k2 >0 m=1
2
Y
≤ Ckf kL2 kfm kL2 . (4.39)
m=1

When j1 = jmax , from (4.38), if − 89 + 16ǫ ≤ s1 < − 34 , we have

2
2jǫ−(j1 +j2 )ǫ− jmax −k2 (2s1 + 41 )+k( 34 +s1 )
X X Y
Int5 ≤ C 2 2 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
−k2 (2s1 + 49 −4ǫ)
X Y
≤C 2 kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ Ckf kL2 kfm kL2 . (4.40)
m=1

36
When j2 = jmax , from (4.38), if − 34 ≤ s1 < 0, we have
2
2jǫ−(j1 +j2 )ǫ− jmax −k2 (2s1 + 41 )+k ( 34 +s1 )
X X Y
Int5 ≤ C 2 2 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
3
X Y
≤C 2−k2 (s1 + 2 −4ǫ) kf kL2 kfm kL2
k1 ,k2 >0 m=1
2
Y
≤ Ckf kL2 kfm kL2 . (4.41)
m=1

When j2 = jmax , this case can be proved similarly to case j1 = jmax .

When (4.24) is valid, by using (1) of Lemma 3.1, since − 89 + 16ǫ ≤ s1 < 0, we have
Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)−2k2 s1 +k(1+s1 )
X X Y
Int5 ≤ C 2 fk,j fkm ,jm dV
k,k1 ,k2 >0 j1 ,j2 ,j≥0 R6 m=1
2
7 3
X X Y
≤C 22jǫ−(j1+j2 )ǫ−k2 (2s1 + 2 )+k( 2 +s1 ) kfk,j kL2 kfkm ,jm kL2
k,k1 ,k2 >0 j1 ,j2 ,j≥0 m=1
2
7 3
X Y
≤C 2−k2 (s1 + 2 −8ǫ)+k( 2 +2ǫ) kf kL2 kfm kL2
k,k1 ,k2 >0 m=1

X 2
Y
≤C 2−k2 (s1 +2−10ǫ) kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
Y
≤ Ckf kL2 kfm kL2 . (4.42)
m=1

(VI) Region Ω6 . In this region, we consider (4.15), (4.16), respectively.

When (4.15) is valid, one of (4.17)-(4.19) must occur.
When (4.17) is valid, since − 98 + 16ǫ ≤ s1 < 0, we have
3
|ξ|hξis1 |ξ2|− 2 −s1 +10ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ C 2
≤ C 2
1 1 1
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ hσj i 2 +ǫ
Q Q
j=1 j=1
− 83 − 21
|ξ2 | |ξ1 | |ξ2 |
≤C 2
≤ C 2
. (4.43)
1 1
+ǫ +ǫ
Q Q
hσj i 2 hσj i 2
j=1 j=1

Thus, combining (2.17) with (4.43), we have

2
Y
|Int6 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

37
 When (4.18) is valid, since − 89 + 16ǫ ≤ s1 < 0 and

1 1 1 1
hσi− 2 +2ǫ hσ1 i− 2 −ǫ ≤ hσi− 2 −ǫ hσ1 i− 2 +2ǫ ,

we have
3
|ξ|hξis1 |ξ2|− 2 −s1 +10ǫ
K1 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ C 2
≤ C 1 1
1 1 hσi 2 +ǫ hσ2 i 2 +ǫ
hσi 2 −2ǫ hξj is1 hσj i 2 +ǫ
Q
j=1
− 83 1
|ξ2 | |ξ|− 2 |ξ2 |
≤C 2 ≤C 1
+ǫ 1
+ǫ
. (4.44)
Q 1
+ǫ hσi 2 hσ2 i 2
hσj i 2
j=1

Thus, combining (2.19) with (4.44), we have

2
Y
|Int6 | ≤ CkF kL2τ ξµ kFj kL2τ ξµ .
j=1

When (4.19) is valid, this case can be proved similarly to (4.18) with the aid of (4.20).
When (4.16) is valid, this case can be proved similarly to case (4.16) of Region Ω5 .
(VII) Region Ω7 . This case can be proved similarly to Region Ω6 .
This completes the proof of Lemma 4.1.
Remark 3. In the case (4.23) of Region Ω4 it leads to the requirement − 89 < s1 < 0.

Lemma 4.2. Let −1 + 10ǫ ≤ s < 0. Then, we have

2
Y
−2+10ǫ
k∂x [IN (u1 u2 ) − IN u1 IN u2 ] kX 0,0 ≤ CN kIN uj kX 0,0 . (4.45)
−1
2 +2ǫ
1
2 +ǫ
j=1

Proof. To prove (4.45), by duality, it suffices to prove that

Z 
 
 3 h̄∂x [IN (u1 u2 ) − IN u1 IN u2 ] dxdydt
 
R
2
Y
−2+10ǫ
≤CN khkX 0,0 kIN uj kX 0,0 . (4.46)
1 −2ǫ 1
2 j=1 2 +ǫ

for h ∈ X 0,0
1
−2ǫ
. Let
2

1
F (ξ, µ, τ ) = hσi 2 −2ǫ M(ξ)F h(ξ, µ, τ ),
1
Fj (ξj , µj , τj ) = M(ξj )hσj i 2 +ǫ F uj (ξj , µ, τj ) (j = 1, 2). (4.47)

38
To obtain (4.46), from (4.47), it suffices to prove
2
Q
Z Z |ξ|G(ξ1, ξ2 )F (ξ, µ, τ ) Fj (ξj , µj , τj )
j=1
dξ1dµ1 dτ1 dξdµdτ
R2
ξ = ξ1 + ξ2 , 1
2 1
−2ǫ +ǫ
Q
µ = µ1 + µ2 , hσj i 2 hσj i 2
j=1
τ = τ1 + τ2
2
Y
−1+10ǫ
≤CN kF kL2ξµτ kFj kL2ξµτ , (4.48)
j=1

where
M(ξ1 )M(ξ2 ) − M(ξ)
G(ξ1 , ξ2 ) = .
M(ξ1 )M(ξ2 )
Without loss of generality, we assume that F (ξ, µ, τ ) ≥ 0, Fj (ξj , µj , τj ) ≥ 0(j = 1, 2). By
symmetry, we can assume that |ξ1| ≥ |ξ2|.
We define
 
∗ N
A1 = (ξ1, µ1 , τ1 , ξ, µ, τ ) ∈ D , |ξ2 | ≤ |ξ1 | ≤ ,
2
N
A2 = {(ξ1 , µ1, τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1 | > , |ξ1 | ≥ |ξ2 |, |ξ2| ≤ 2A},
2
N
A3 = {(ξ1 , µ1, τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1 | > , |ξ1 | ≥ |ξ2 |, 2A < |ξ2 | ≤ N},
2
N
A4 = {(ξ1 , µ1, τ1 , ξ, µ, τ ) ∈ D ∗ , |ξ1 | > , |ξ1 | ≥ |ξ2 |, |ξ2| > N}.
2
4
Here D ∗ is defined as in Lemma 3.1. Obviously, D ∗ ⊂
S
Aj . We define
j=1

|ξ|G(ξ1, ξ2 )
K2 (ξ1 , µ1, τ1 , ξ, µ, τ ) := 2
(4.49)
1 1
hσj i 2 −2ǫ hσj i 2 +ǫ
Q
j=1

and
Z 2
Y
Jk := K2 (ξ1 , µ1, τ1 , ξ, µ, τ )F (ξ, µ, τ ) Fj (ξj , µj , τj )dξ1 dµ1dτ1 dξdµdτ,
Aj j=1

1 ≤ k ≤ 4, k ∈ N.
We consider (4.15) and (4.16), respectively.
When (4.15) is valid, one of (4.17)-(4.19) must occur, from [20, Lemma 1.4], we have
4
X 2
Y
−(2−10ǫ)
Jk ≤ CN kF kL2τ ξµ kFj kL2τ ξµ .
k=1 j=1

39
Thus, we only consider the case (4.16).
Now we consider the integrals over the above four regions one by one.
(I) Region A1 . In this case, since M(ξ1 , ξ2 ) = 0, thus we have J1 = 0.
(II) Region A2 . From [20, Page 902], we have

|ξ2 |
G(ξ1 , ξ2) ≤ C . (4.50)
|ξ1 |

Inserting (4.46) into (4.47) yields

|ξ|G(ξ1, ξ2 ) C|ξ2 |
K2 (ξ1 , µ1, τ1 , ξ, µ, τ ) ≤ C 2
≤ 2
. (4.51)
1 1 1 1
−2ǫ +ǫ −2ǫ +ǫ
Q Q
hσi 2 hσj i 2 hσi 2 hσj i 2
j=1 j=1

We dyadically decompose the spectra as

hσi ∼ 2j , hσ1 i ∼ 2j1 , hσ2 i ∼ 2j2 , |ξ| ∼ 2k , |ξ1 | ∼ 2k1 , |ξ2| ∼ 2k2 .

We define

fkm ,jm := ηkm (ξm )ηjm (σm )Fj (ξm , µm , τm )(m = 1, 2),

fk,j := ηk (ξ)ηj (σ) |F (ξ, µ, τ ))| .

Thus, by using (2.4), we have

Z 2
−j( 21 −2ǫ)−(j1 +j2 )( 12 +ǫ)+k2
X X Y
J2 ≤ C 2 fk,j fkm ,jm dV. (4.52)
k1 ,k2 >0, k j1 ,j2 ,j≥0 R6 m=1

In this case, we consider (4.23), (4.24), respectively.

When (4.23) is valid, we consider j = jmax , j1 = jmax , j2 = jmax , respectively.
When j = jmax is valid, from (4.52), we have
2
9 1
X X Y
J2 ≤ C 2−k( 4 −8ǫ)−(j1 +j2 )ǫ+k2 ( 4 +2ǫ) kf kL2 kfj kL2
k1 ,k2 >0, k j1 ,j2 j=1
2
9
Y
≤ CN −( 4 −8ǫ) kf kL2 kfj kL2 . (4.53)
j=1

40
When j1 = jmax is valid, from (4.52), we have
2
j1
+ 34 k2
X X Y
2jǫ−(j1 +j2 )ǫ−
J2 ≤ C 2 2 kf kL2 kfj kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 j=1
2
k1
−j2 ǫ−j1 ( 21 −ǫ)− + 43 k2
X X Y
≤C 2 4 kf kL2 kfj kL2
k1 ,k2 >0, k j1 ,j2 ≥0 j=1
2
9 1
X Y
≤C 2−( 4 −4ǫ)k1 +k2 ( 2 +ǫ) kf kL2 kfj kL2
k1 ,k2 >0, k j=1
2
−( 94 −4ǫ)
Y
≤ CN kf kL2 kfj kL2 . (4.54)
j=1

When j2 = jmax is valid, this case can be proved similarly to j1 = jmax of Region A2 .
When (4.24) is valid, by using (2) of Lemma 3.1, we have

X X 2
Y
2jǫ−(j1 +j2 )ǫ−2k1 +k2 ǫ
J2 ≤ C 2 kf kL2 kfj kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 j=1

X X 2
Y
≤C 2−(j1 +j2 )ǫ−(2−8ǫ)k1 +3k2 ǫ kf kL2 kfj kL2
k1 ,k2 >0, k j1 ,j2 ≥0 j=1
2
7 1
X Y
≤C 2−( 2 −8ǫ)k1 +k2 ( 2 +2ǫ) kf kL2 kfj kL2
k1 ,k2 >0, k j=1
2
−( 27 −8ǫ)
Y
≤ CN kf k L2 kfj kL2 . (4.55)
j=1

(III) Region A3 . From of [20, Page 902], we know that (4.50) is valid. Combining
(4.50) with (4.49), we have

min {|ξ|, |ξ1|, |ξ2|}

K2 (ξ1 , µ1 , τ1 , ξ, µ, τ ) ≤ C 2
. (4.56)
1 1
−2ǫ +ǫ
Q
hσi 2 hσj i 2
j=1

We dyadically decompose the spectra as

hσi ∼ 2j , hσ1 i ∼ 2j1 , hσ2 i ∼ 2j2 , |ξ| ∼ 2k , |ξ1 | ∼ 2k1 , |ξ2 | ∼ 2k2 .

We define

fkm ,jm := ηkm (ξm )ηjm (σk )Fm (ξm , µm , τm ) (m = 1, 2),

fk,j := ηk (ξ)ηj (σ) |F (ξ, µ, τ ))| .

41
Thus, we have

1 1
X X
J3 ≤ C 2−j( 2 −2ǫ)−(j1 +j2 )( 2 +ǫ)+min{k,k1 ,k2 +k2 }
m1 ,m2 >0, m j1 ,j2 ,j≥0
Z 2
Y
× fm,j fkm ,jm dV. (4.57)
R6 m=1

In this case, we consider (4.23), (4.24), respectively.

When (4.23) is valid, by using Lemma 3.2, we have

1 j jmax k1
+ 43 min{k,k1 ,k2 }
X X
J3 ≤ C 2−j( 2 −2ǫ)−(j1 +j2 )ǫ+ 2 − 2
− 4

k,k1 ,k2 >0 j1 ,j2 ,j≥0

Z 2
Y
× fk,j fkm ,jm dV. (4.58)
R6 m=1

When j = jmax , from (4.58), we have

2
k1
−j( 21 −2ǫ)−(j1 +j2 )ǫ− + 43 min{k,k1 ,k2 }
X X Y
J3 ≤ C 2 4 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
−k2 ( 94 −8ǫ)+min{k,k1 ,k2 }( 14 +2ǫ)
X Y
≤C 2 kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ CN −2+10ǫ kf kL2 kfm kL2 . (4.59)
m=1

When j1 = jmax , from (4.59), we have

2
X X j1 k1 3
Y
J3 ≤ C 22jǫ−(j1 +j2 )ǫ− 2 + 4 + 4 min{k,k1,k2 } kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
9 1
X Y
≤C 2−k2 ( 4 −4ǫ)+min{k,k1 ,k2 }( 4 +ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1
2
Y
≤ CN −2+5ǫ kf kL2 kfm kL2 . (4.60)
m=1

When j2 = jmax , this case can be proved similarly to the case j1 = jmax .

42
 When (4.24) is valid, combining (2) of Lemma 3.1 with (4.60), we have
Z 2
−j( 12 −2ǫ)−(j1 +j2 )( 21 +ǫ)+min{k,k1 ,k2 }
X X Y
J3 ≤ C 2 fk,j fkm ,jm dV
k,k1 ,k2 >0 j1 ,j2 ,j≥0 R6 m=1

X X 2
Y
≤C 22jǫ−(j1 +j2 )ǫ−2k1 kfk,j kL2 kfkm ,jm kL2
k,k1 ,k2 >0 j1 ,j2 ,j≥0 m=1

X 2
Y
≤C 2−k1 (2−10ǫ) kf kL2 kfm kL2
k,k1 ,k2 >0 m=1
2
Y
≤ CN −2+11ǫ kf kL2 kfm kL2 . (4.61)
m=1

(IV) Region A4 . In this case, we have

M(ξ1 , ξ2 ) ≤ CN 2s1 (|ξ1| |ξ2 |)−s1 . (4.62)

We dyadically decompose the spectra as

hσi ∼ 2j , hσ1 i ∼ 2j1 , hσ2 i ∼ 2j2 , |ξ| ∼ 2k , |ξ1 | ∼ 2k1 , |ξ2| ∼ 2k2 .

We define

fkm ,jm := ηkm (ξm )ηjm (σm )Fm (ξm , µm , τm ), (m = 1, 2),

fk,j := ηk (ξ)ηj (σ) |F (ξ, µ, τ ))| .

Thus, we have

1 1
X X
J4 ≤ CN 2s 2−j( 2 −2ǫ)−(j1 +j2 )( 2 +ǫ)+(k1 +k2 )|s|+k
m1 ,m2 >0, m j1 ,j2 ,j≥0
Z 2
Y
× fk,j fkm ,jm dV. (4.63)
R6 m=1

In this case, we consider (4.23), (4.24), respectively.

When (4.23) is valid, by using Lemma 3.2, from (4.63), we have

jmax
+(m1 +m2 )|s|− 41 k1 − 14 min{k,k1 ,k2 }+k
X X
J4 ≤ CN 2s 22jǫ−(j1 +j2 )ǫ− 2

m1 ,m2 >0, m j1 ,j2 ,j≥0

2
Y
× kf kL2 kfm kL2 . (4.64)
m=1

43
In this case, we consider k = kmin , k2 = kmin, respectively.
When k = kmin is valid, we consider j = jmax , j1 = jmax , j2 = jmax , respectively.
When j = jmax , from (4.64), since s1 ≥ −1 + 6ǫ, we have
2
−j( 21 −2ǫ)−(j1 +j2 )ǫ−k1 (2s1 + 14 )+ 34 k
X X Y
2s1
J4 ≤ CN 2 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ,j≥0 m=1
2
9 1
X Y
≤C 2−k2 (2s1 + 4 −8ǫ)+k( 4 +2ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1

X 2
Y
≤C 2−k2 (2s1 +2−10ǫ) kf kL2 kfm kL2
k1 >0 m=1
2
Y
≤ CN −2+10ǫ kf kL2 kfm kL2 . (4.65)
m=1

When j1 = jmax is valid, from (4.64), since s1 ≥ −1 + 6ǫ, we have

2
k1
−( 12 −ǫ)j1 −j2 ǫ−(k1 +k2 )s1 − + 43 k
X X Y
J4 ≤ C 2 4 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ≥0 m=1
2
9 1
X Y
≤C 2−k2 (2s1 + 4 −4ǫ)+k( 4 +ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1

X 2
Y
≤C 2−k2 (2s1 +2−5ǫ) kf kL2 kfm kL2
k1 >0 m=1
2
Y
≤ CN −2+11ǫ kf kL2 kfm kL2 . (4.66)
m=1

When j2 = jmax is valid, this case can be proved similarly to j1 = jmax .

When k2 = kmin is valid, we consider j = jmax , j1 = jmax , j2 = jmax , respectively.

44
 When j = jmax , from (4.64), since s1 ≥ −1 + 10ǫ, we have
1 3 1
X X
J4 ≤ CN 2s1 2−j( 2 −2ǫ)−(j1 +j2 )ǫ−k1 (s1 − 4 )−k2 (s1 + 4 )
k1 ,k2 >0, k j1 ,j2 ,j≥0
2
Y
× kfk,j kL2 kfkm ,jm kL2
m=1
2
−k1 (s1 + 54 −8ǫ)−k2 (s1 + 34 −2ǫ)
X Y
2s1
≤ CN 2 kf kL2 kfm kL2
k1 ,k2 >0, k m=1

X 2
Y
≤ CN 2s1 2−k1 (s1 +1−8ǫ)−k2 (s1 +1−2ǫ) kf kL2 kfm kL2
k1 ,k2 >0 m=1
2
Y
−2+10ǫ
≤ CN kf kL2 kfm kL2 . (4.67)
m=1

When j1 = jmax is valid, from (4.64), since s1 ≥ −1 + 10ǫ, we have

2
−( 21 −ǫ)j1 −j2 ǫ−k1 (s1 − 43 )−k2 (s1 + 14 )
X X Y
2s1
J4 ≤ CN 2 kfk,j kL2 kfkm ,jm kL2
k1 ,k2 >0, k j1 ,j2 ≥0 m=1
2
5 3
X Y
≤ CN 2s1 2−k1 (s1 + 4 −4ǫ)−k2 (s1 + 4 +ǫ) kf kL2 kfm kL2
k1 ,k2 >0, k m=1

X 2
Y
≤ CN 2s1 2−k2 (s1 +1−4ǫ)−k2 (s1 +1+ǫ) kf kL2 kfm kL2
k1 ,k2 >0 m=1
2
Y
≤ CN −2+5ǫ kf kL2 kfm kL2 . (4.68)
m=1

When j2 = jmax is valid, this case can be proved similarly to j1 = jmax .

This completes the proof of Lemma 4.2.

Lemma 4.3. Let s ≥ − 89 + 16ǫ, s2 ≥ 0 and uj ∈ X s11+ǫ

,s2
(j = 1, 2). Then, we have
2

2
Y
k∂x I(u1 u2 )kX 0,0 ≤C kIuj kX 0,0 . (4.69)
1 +2ǫ
−2 1
j=1 2 +ǫ

Proof. To prove (4.69), by duality, it suffices to prove that

Z  Y2
 
 ū∂x I(u1 u2 )dxdydt ≤ CkukX 0,0
 kIuj kX 0,0 . (4.70)
1 −2ǫ 1
R3

2 j=1 2 +ǫ

for u ∈ X 0,0
1
−2ǫ
. Let
2
1
F (ξ, µ, τ ) = hσi 2 −2ǫ F u(ξ, µ, τ ),
1
Fj (ξj , µj , τj ) = M(ξj )hσj i 2 +ǫ F uj (ξj , µ, τj )(j = 1, 2), (4.71)

45
and
( 2 2 2
)
X X X
D := (ξ1 , µ1 , τ1 , ξ, µ, τ ) ∈ R6 , ξ = ξj , µ = µj , τ = τj .
j=1 j=1 j=1

To obtain (4.70), from (4.71), it suffices to prove that

|ξ|M(ξ)F (ξ, µ, τ )Fj (ξj , µj , τj )
Z
2
dξ1 dµ1 dτ1 dξdµdτ
D 1 1
−2ǫ +ǫ
Q
hσj i 2 M(ξj )hσj i 2
j=1
2
Y
≤CkF kL2τ ξµ kFj kL2τ ξµ . (4.72)
j=1

From (2.4) of [26], we have

M(ξ) hξis
2
≤ C 2
. (4.73)
Q Q s
M(ξj ) hξj i
j=1 j=1

Inserting (4.73) into the left hand side of (4.72), we have

|ξ|hξisF (ξ, µ, τ )Fj (ξj , µj , τj )
Z
2
dξ1dµ1 dτ1 dξdµdτ. (4.74)
D 1 1
−2ǫ +ǫ
Q s
hσj i 2 hξj i hσj i 2
j=1

2
Q
By using (4.4), we have (4.74) can be bounded by CkF kL2τ ξµ kFj kL2τ ξµ .
j=1
This completes the proof of Lemma 4.3.

5. Proof of Theorem 1.1

This section is devoted to proving Theorem 1.1.
We define
 Z t
1 t
Φ1 (u) := ψ(t)W (t)u0 + ψ W (t − τ )∂x (u2 )dτ, (5.1)
2 τ 0
 
B1 (0, 2Cku0kH s1 ,s2 ) := u : kukX s11 ,s2 ≤ 2Cku0 kH s1 ,s2 . (5.2)
2 +ǫ

Combining Lemmas 2.2, 4.1 with (5.1)-(5.2), we derive that

 Z t
1 t 2

kΦ1 (u)kX s1 ,s2 ≤ kη(t)W (t)u0kX s1 ,s2 +
2 η W (t − τ )∂x (u )dτ
b 1 +ǫ
2 τ 0
s1 ,s2
Xb
2
ǫ

≤Cku0 kH s1 ,s2 + CT ∂x (u ) s1 ,s2 X 1 +2ǫ
−2

≤Cku0 kH s1 ,s2 + CT ǫ kuk2X s1 ,s2

b

≤Cku0 kH s1 ,s2 + 4C 3 T ǫ ku0 k2H s1 ,s2 . (5.3)

46
We define
−1
T ǫ := 16C 2 (ku0 kH s1 ,s2 + 1)

. (5.4)

From (5.3)-(5.4), we have

kΦ1 (u)kX s1 ,s2 ≤ Cku0kH s1 ,s2 + C ku0 kH s1 ,s2 = 2C ku0 kH s1 ,s2 . (5.5)
1 +ǫ
2

Thus, Φ1 maps B1 (0, 2Cku0kH s1 ,s2 ) into B1 (0, 2Cku0kH s1 ,s2 ). Combining Lemmas 2.2,
4.1 with (5.4)-(5.5), we have
 Z t
1 t 2 2

kΦ1 (u1 ) − Φ1 (u2 )kX s1 ,s2 ≤C
2η τ W (t − τ )∂x (u1 − u2 )dτ
1 +ǫ s1 ,s2
2 0 X1
2 +ǫ
 
ǫ
≤ CT ku1 − u2 kX s1 ,s2 ku1 kX s1 ,s2 + ku2 kX s1 ,s2
1 +ǫ 1 1 +ǫ
2 2 +ǫ 2

≤ 4C 2 T ǫ ku0 kH s1 ,s2 ku1 − u2 kX s1 ,s2

1
2 +ǫ
1
≤ ku1 − u2 kX s1 ,s2 . (5.6)
2 1
2 +ǫ

Thus, Φ1 is a contraction mapping in the closed ball B1 (0, 2Cku0kH s1 ,s2 ). Consequently,
u is the fixed point of Φ in the closed ball B1 (0, 2Cku0kH s1 ,s2 ). Then v := u|[0,T ] ∈
X s11+ǫ
,s2
([0, T ]) is a solution to the Cauchy problem for (1.1) with the initial data u0 in
2

the interval [0, T ]. For the facts that uniqueness of the solution and the solution to the
Cauchy problem for (1.1) is continuous with respect to the initial data, we refer the
readers to Theorems II, III of [24].
This completes the proof of Theorem 1.1.

6.Proof of Theorem 1.2

We firstly prove Lemma 6.1 which is a variant of Theorem 1.1, then we apply Lemma
6.1 to prove Theorem 1.2.

Lemma 6.1. Let s1 > − 89 and R := 1

8(C+1)3
, where C is the largest of those constants
which appear in (2.7)-(2.8), (4.42), (4.66). Then, the Cauchy problem for (1.1) is locally
well-posed for data satisfying

kIN u0 kL2 ≤ R. (6.1)

Moreover, the solution to the Cauchy problem for (1.1) exists on a time interval [0, 1].

47
Proof. We define v := IN u. Let u be the solution to the Cauchy problem for (1.1), then
v is the solution to the following equations
1
vt + ∂x5 v + ∂x−1 ∂y2 v + IN ∂x (IN−1 v)2 = 0. (6.2)
2
Thus, v satisfies the following equations
1 t
Z
v = W (t)v0 + W (t − τ )IN ∂x (IN−1 v)2 . (6.3)
2 0
We define
t
1
Z
Φ2 (v) := ψ(t)W (t)IN u0 + ψ(t) W (t − τ )IN ∂x (IN−1 v)2 . (6.4)
2 0

Combining Lemma 2.2 with 4.3, we have

Z t
−1 2

kΦ2 (v)kX 0,0 ≤ kψ(t)W (t)IN u0 kX 0,0 + C ψ(t)
W (t − τ )IN ∂x (IN v)
1 1 +ǫ
2 +ǫ 2 0 X 0,0
1
2 +ǫ

+ C IN ∂x (IN−1 v)2 X 0,0

≤C kIN u0 kL2
−1 2 +2ǫ
−1 2

≤C kIN u0 kL2 + C IN ∂x (IN v) X 0,0

−1
2 +2ǫ

≤C kIN u0 kL2 + Ckvk2X 0,0

1
2 +ǫ

≤CR + Ckvk2X 0,0 . (6.5)

1
2 +ǫ

We define
 
B2 (0, 2CR) := v : kvkX 0,0 ≤ 2CR . (6.6)
1 +ǫ
2

Combining (6.5)-(6.6) with the definition of R, we have

kΦ2 (v)kX 0,0 ≤ CR + 4C 3 R2 = 2CR. (6.7)

1
2 +ǫ

Thus, Φ2 is a map from B2 (0, 2CR) to B2 (0, 2CR). We define

vj := IN uj (j = 1, 2), w1 = IN−1 v1 − IN−1 v2 , w2 := IN−1 v1 + IN−1 v2 . (6.8)

Combining Lemmas 2.2, 3.1, 3.2, (6.5)-(6.6) with the definition of R, we have
Z t
 −1 2 −1 2

kΦ2 (v1 ) − Φ2 (v2 )kX 0,0 ≤C ψ(t)
W (t − τ )∂x IN (IN v1 ) − (IN v2 ) dτ
1 +ǫ
X 0,0

2 0 1
2 +ǫ

≤ C k∂x IN (w1 w2 )kX 0,0

− 1 +2ǫ
 2 
≤ Ckv1 − v2 kX 0,0 kv1 kX 0,0 + kv2 kX 0,0
1 +ǫ 1 1 +ǫ
2 2 +ǫ 2
1
≤ 4C 2 Rkv1 − v2 kX 0,0 ≤ kv1 − v2 kX 0,0 . (6.9)
1
2 +ǫ
2 1 +ǫ
2

48
Thus, Φ2 is a contraction mapping from B2 (0, 2CR) to B2 (0, 2CR). Consequently, u is
the fixed point of Φ2 in the closed ball B2 (0, 2CR). Then v := Iu|[0,1] ∈ X 0,0
1
+ǫ
([0, 1])
2

is a solution to the Cauchy problem for (5.3) with the initial data IN u0 on the interval
[0, 1]. For the uniqueness of the solution and the fact that the solution is continuous
with respect to the initial data, we refer the readers to Theorem II, III of [24].
This completes the proof of Lemma 6.1.
Now we apply the idea of [20] and Lemmas 2.7, 4.2, 6.1 to prove Theorem 1.2.
For λ > 0, we define
4
 1 3
 4
 1 3

uλ (x, y, t) := λ u λ x, λ y, λt , u0λ (x, y) := λ u λ x, λ y .
5 5 5 5 5 5 (6.10)

Thus, uλ (x, y, t) ∈ X s11+ǫ

,0
([0, Tλ ]) is the solution to
2

∂t uλ + ∂x5 uλ + ∂x−1 ∂y2 uλ + uλ ∂x uλ = 0, (6.11)

uλ (x, y, 0) = u0λ (x, y), (6.12)

if and only if u(x, y, t) ∈ X s,0

1
+ǫ
([0, T ]) is the solution to the Cauchy problem for (1.1) in
2

[0, T ] with the initial data u0 . By using a direct computation, for λ ∈ (0, 1), we have
2 s
kIN u0λ kL2 ≤ CN −s λ 5 + 5 ku0 kH s,0 . (6.13)

For u0 6= 0 and u0 ∈ H s,0 (R2 ), we choose λ, N such that

2 s R
kIN u0λ kL2 ≤ CN −s λ 5 + 5 ku0 kH s,0 := . (6.14)
4
Then there exists w3 which satisfies that kw3 kX s,0 ≤ 2CR such that v := w3 |[0,1] is a
1 +ǫ
2
solution to the Cauchy problem for (6.11) with u0λ . Multiplying (6.11) by 2IN uλ and
integrating with respect to x, y yield
d
Z Z
2
IN u∂x IN (u)2 dxdy = 0.
 
(IN u) dxdy + (6.15)
dt R2 R2

Inserting
Z
IN u∂x (IN u)2 dxdy = 0
 
R2

into (6.15) yields

d
Z Z
2
IN u∂x IN (u)2 − (IN u)2 dxdy.


(IN u) dxdy = − (6.16)
dt R2 R2

49
Combining (6.16) with Lemmas 2.6, 4.2, we have
Z Z
2
(IN u(x, y, 1)) dxdy − (IN u0λ )2 dxdy
2 2
R R
Z 1Z
IN uλ ∂x IN (uλ )2 − (IN uλ)2 dxdydt


=−
2
Z0 Z R
χ[0,1] (t)IN uλ χ[0,1] (t)∂x IN (uλ )2 − (IN uλ )2 dxdydt




=−
2
R R
≤ C χ[0,1] (t)IN uλ X 0,0 χ[0,1] (t)∂x IN (uλ)2 − (IN uλ )2 X 0,0



1 −ǫ 1 +ǫ
−2
2
≤ C kIN uλ kX 0,0 ∂x IN (uλ )2 − (IN uλ )2 X 0,0


1−ǫ 1 +2ǫ
2 −2
−2+10ǫ
≤ CN kIN uλ k3X 0,0 . (6.17)
1
2 +ǫ

From (6.14) and (6.15) and the definition of R, we have

R2
Z
(IN u(x, y, 1))2dxdy ≤ + CN −2+10ǫ kIN uλ k3X 0,0
R 2 16 1 +ǫ
2

R2 R 2
≤ + 8C 4 N −2+10ǫ R3 ≤ + CN −2+10ǫ . (6.18)
16 16

Let N be sufficiently large such that such that 8C 4 N −2+10ǫ R3 ≤ 43 R2 , then

Z  12
2
(IN u(x, y, 1)) dxdy ≤ R. (6.19)
R2

We consider IN u(x, y, 1) as the initial data and repeat the above argument, from Lemma
6.1, we obtain that (6.11)-(6.12) possess a solution in R2 × [1, 2]. In this way, we can
extend the solution to (6.11)-(6.12) to the time interval [0, 2]. The above argument can
be repeated L steps, where L is the maximal positive integer such that

3
CN −2+10ǫ L ≤ R2 . (6.20)
4

More precisely, the solution to (6.11)-(6.12) can be extended to the time interval [0, L].
Thus, we can prove that (6.11)-(6.12) are globally well-posed in [0, Tλ ] if

T
L≥ . (6.21)
λ

From (6.20), we know that

L ∼ N 2−10ǫ . (6.22)

50
We know that (6.21) is valid provided that the following inequality is valid

T −5s
CN 2−10ǫ ≥ ∼ CT N 2+s . (6.23)
λ

In fact, (6.23) is valid if

−5s
N 2 > N 2+s (6.24)

which is equivalent to − 74 < s < 0.

This completes the proof of Theorem 1.2.

Acknowledgments

This work is supported by the Natural Science Foundation of China under grant
numbers 11571118 and 11771127. The first author is also supported by the Fundamental
Research Funds for the Central Universities of China under the grant number 2017ZD094,
while second author is also supported by the Young core Teachers program of Henan
Normal University under grant number 15A110033.

References

References

[1] L. A. Abramyan, Y. A. Stepanyants, The structure of two-dimensional solitons in

media with anomalously small dispersion, Sov. Phys. JETP. 61(1985), 963–966.

[2] M. Beals, Self-spreading and strength of singularities for solutions to semilinear

wave equations, Ann. of Math. 118(1983), 187–214.

[3] M. Ben-Artzi, J. C. Saut, Uniform decay estimates for a class of oscillatory integrals
and applications, Diff. Int. Eqns. 12(1999), 137–145.

[4] J. Bourgain, On the Cauchy problem for the Kadomtsev-Petviashvili equation,

Geom. Funct. Anal. 3 (1993), no. 4, 315–341.

51
 [5] J. Bourgain, Fourier transform restriction phenomena for certain lattice subsets and
applications to nonlinear evolution equations, part II: The KdV equation, Geom.
Funct. Anal. 3(1993), 209–262.

[6] W. G. Chen, J. F. Li, C. X. Miao, On the low regularity of the fifth order Kadomtsev-
Petviashvili I equation, J. Diff. Eqns. 245(2008), 3433–3469.

[7] J. Colliander, C. E. Kenig, G. Staffilani, Low regularity solutions for the Kadomtsev-
Petviashvili I equation, Geom. Funct. Anal. 13(2003), 737–794.

[8] J. Colliander, A. D. Ionescu, C. E. Kenig, G. Staffilani, Weighted low-regularity

solutions of the KP-I initial-value problem, Discrete Contin. Dyn. Syst. 20(2008),
219–258.

[9] J. Colliander, M. Keel, G. Staffilani, H. Takaoka, T. Tao, Global well-posedness for

KdV in Sobolev Spaces of negative index, Electr. J. Diff. Eqns. 2001(2001), 1–7.

[10] J. Colliander, M. Keel, G. Staffilani, H. Takaoka, T. Tao, Sharp global well-

posedness for KdV and modified KdV on R and T , J. Amer. Math. Soc. 16(2003),
705–749.

[11] J. Ginibre, Le problème de Cauchy pour des EDP semi-linéaires périodiques en

variables d’espace (d’après Bourgain), Astérisque (1996), no. 237, Exp. No. 796, 4,
163–187, séminaire Bourbaki, Vol. 1994/95.

[12] B. L. Guo, Z. H. Huo, S. M. Fang, Low regularity for the fifth order Kadomtsev-
Petviashvili-I type equation, J. Diff. Eqns. 263(2017), 5696–5726.

[13] Z. H. Guo, L. Z. Peng, B. X. Wang, On the local regularity of the KP-I equation in
anisotropic Sobolev space, J. Math. Pures Appl. 94(2010), 414–432.

[14] A. Grünrock, New applications of the Fourier restriction norm method to wellposed-
ness problems for nonlinear Evolution Equations, Ph.D. Universität Wuppertal,
2002, Germany, Dissertation.

[15] M. Hadac, Well-posedness for the Kadomtsev-Petviashvili II equation and general-

izations, Trans. Amer. Math. Soc. 360(2008), 6555–6572.

52
[16] M. Hadac, S. Herr, H. Koch, Well-posedness and scattering for the KP-II equation
in a critical space, Ann. Inst. H. Poincaré-AN, (2009), 917–941.

[17] N. Hayashi, P. l. Naumkin, Large time asymptotics for the Kadomtsev-Petviashvili

equation, Comm. Math. Phys. 332(2014), 505–533.

[18] A. D. Ionescu, C. E. Kenig, D. Tataru, Global well-posedness of the KP-I initial-

value problem in the energy space, Invent. Math. 173 (2008), 265–304.

[19] P. Isaza, F. Linares, G. Ponce, On the propagation of regularity of solutions of the

Kadomtsev-Petviashvili equation, SIAM J. Math. Anal. 48(2016), 1006–1024.

[20] P. Isaza, J. López, J. Mejı́a, Cauchy problem for the fifth order Kadomtsev-
Petviashvili (KPII) equation, Commun. Pure Appl. Anal. 5(2006), 887–905.

[21] P. Isaza, J. Mejı́a, V. Stallbohm, Local solution for the Kadomtsev-Petviashvili

equation with periodic conditions, Manuscr. Math. 75(1992), 383–393.

[22] P. Isaza, J. Mejı́a, V. Stallbohm, A regularity theorem for the Kadomtsev-

Petviashvili equation with periodic boundary conditions, Nonlinear Anal. 23(1994),
683–687.

[23] P. Isaza, J. Mejı́a, V. Stallbohm, Local solution for the Kadomtsev-Petviashvili

equation in R2 , J. Math. Anal. Appl. 196(1995), 566–587.

[24] P. Isaza, J. Mejı́a V. Stallbohm, The Cauchy problem for the Kadomtsev-
Petviashvili equations (KPII) in Sobolev spaces H s , s > 0, Diff. Int. Eqns. 14(2001),
529–557.

[25] P. Isaza, J. Mejı́a, Local and global Cauchy problems for the Kadomtsev-Petviashvili
(KP-II) equation in Sobolev spaces of negative indices, Commun. Partial Diff. Eqns.
26(2001), 1027–1054.

[26] P. Isaza, J. Mejı́a, Global solution for the Kadomtsev-Petviashvili equation (KP-II)
in anisotropic Sobolev spaces of negative indices, Electr. J. Diff. Eqns. 2003, 68, 12
pp.

53
[27] P. Isaza, J. Mejı́a, N. Tzvetkov, A smoothing effect and polynomial growth of the
Sobolev norms for the KP-II equation, J. Diff. Eqns. 220(2006), 1–17.

[28] P. Isaza, J. López, J. Mejı́a, The Cauchy problem for the Kadomtsev-Petviashvili
(KPII) equation in three space dimensions, Commun. Partial Diff. Eqns. 32(2007),
611–641.

[29] P. Isaza, J. López, J. Mejı́a, On the support of solutions to the Kadomtsev-

Petviashvili (KP-II) equation, Commun. Pure Appl. Anal. 10(2011), 1239–1255.

[30] B. B. Kadomtsev, V. I. Petviashvili, On the stability of solitary waves in weakly

dispersive media, Soviet. Phys. Dokl. 15(1970), 539–541.

[31] V. I. Karpman, V. Yu. Belashov, Dynamics of two dimensional solitons in weakly

dispersive media, Phys. Lett. A. 154(1991), 131–139.

[32] V. I. Karpman, V. Yu. Belashov, Evolution of three-dimensional nonlinear pulses

weakly dispersive media, Phys. Lett. A. 154(1991), 1401–144.

[33] C. E. Kenig, On the local and global well-posedness theory for the KP-I equation,
Ann. Inst. H. Poincaré-AN, 21(2004), 827–838.

[34] C. E. Kenig, G. Ponce, L. Vega, The Cauchy problem for the Korteweg-de Vries
equation in Sobolev spaces of negative indices, Duke Math. J. 71(1993), 1–21.

[35] C. E. Kenig, G. Ponce, L. Vega, A bilinear estimate with applications to the KdV
equation, J. Amer. Math. Soc. 9(1996), 573–603.

[36] S. Klainerman, M. Machedon, Smoothing estimates for null forms and applications,
Int. Math. Res. Not. 1994(9), 1994, 383–389.

[37] H. Koch, J. F. Li, Global well-posedness and scattering for small data for the
three-dimensional Kadomtsev-Petviashvili II equation, Commun. Partial Diff. Eqns.
42(2017), 950–976.

54
[38] J. F. Li, J. Xiao, Well-posedness of the fifth order Kadomtsev-Petviashvili I equa-
tion in anisotropic Sobolev spaces with nonnegative indices, J. Math. Pures Appl.
90(2008), 338-352.

[39] Y. Liu, Strong instability of solitary-wave solutions to a Kadomtsev-Petviashvili

equation in three dimensions, J. Diff. Eqns. 180(2002), 153–170.

[40] L. Molinet, J. C. Saut, N. Tzvetkov, Well-posedness and ill-posedness results for

the Kadomtsev-Petviashvili-I equation, Duke Math. J. 115(2002), 353–384.

[41] L. Molinet, J. C. Saut, N. Tzvetkov, Global well-posedness for the KP-I equation,
Math. Ann. 324(2002), 255–275.

[42] L. Molinet, J. C. Saut, N. Tzvetkov, Correction: Global well-posedness for the

KP-I equation [Math. Ann.324(2002), 255-275;MR1933858] Math. Ann. 328(2004),
707–710.

[43] L. Molinet, J. C. Saut, N. Tzvetkov, Global well-posedness for the KP-I equation
on the background of a non localized solution, Commun. Math. Phys. 272(2007),
775–810.

[44] L. C. Molinet, J. C. Saut, N. Tzvetkov, Global well-posedness for the KP-II equa-
tion on the background of a non-localized solution, Ann. Inst. H. Poincaré-AN, 28
(2011), 653–676.

[45] J. Rauch, M. Reed, Nonlinear microlocal analysis of semilinear hyperbolic systems

in one space dimension, Duke Math. J. 49(1982), 397–475.

[46] J. C. Saut, N. Tzvetkov, The Cauchy problem for higher-order KP equations, J.

Diff. Eqns. 153(1999), 196–222.

[47] J. C. Saut, N. Tzvetkov, The Cauchy problem for the fifth order KP equations, J.
Math. Pures Appl. 79(2000), 307–338.

[48] J. C. Saut, N. Tzvetkov, On periodic KP-I type equations, Commun. Math. Phys.
221(2001), 451–476.

55
[49] H. Takaoka, Global well-posedness for the Kadomtsev-Petviashvili II equation, Dis-
crete Contin. Dyn. Syst. 6(2000), 483–499.

[50] H. Takaoka, Well-posedness for the Kadomtsev-Petviashvili II equation, Adv. Diff.

Eqns. 5(2000), 1421-1443.

[51] H. Takaoka, N. Tzvetkov, On the local regularity of the Kadomtsev-Petviashvili-II

equation, Int. Math. Res. Not. 2001(2001), 77–114.

[52] N. Tzvetkov, On the Cauchy problem for Kadomtsev-Petviashvili equation, Com-

mun. Partial Diff. Eqns. 24(1999), 1367–1397.

[53] N. Tzvetkov, Global low-regularity solutions for Kadomtsev-Petviashvili equation,

Diff. Int. Eqns. 13(2000), 1289-1320.

[54] N. Tzvetkov, Long time bounds for the periodic KP-II equation, Int. Math. Res.
Not. 2004(2004), 2485–2496.

[55] W. Yan, Y. S. Li, J. H. Huang, J. Q. Duan, The Cauchy problem for two dimen-
sional generalized Kadomtsev-Petviashvili-I equation in anisotropic Sobolev spaces,
arXiv:1709.01983.

[56] Y. Zhang, Local well-posedness of KP-I initial value problem on torus in the Besov
space, Commun. Partial Diff. Eqns. 41(2016), 256–281.

56