Escolar Documentos
Profissional Documentos
Cultura Documentos
u +∇
u ·∇
u+
-Δ
ut
iut + Δu = F(u)
Baoxiang Wang
Peking University, China
Zhaohui Huo
Chinese Academy of Sciences, China
Chengchun Hao
Chinese Academy of Sciences, China
Zihua Guo
Peking University, China
World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I • CHENNAI
For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.
ISBN-13 978-981-4360-73-9
ISBN-10 981-4360-73-2
Printed in Singapore.
To our parents
v
This page intentionally left blank
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Preface
During the past thirty years, the nonlinear evolution equations (NLE)
have made a great progress by using the harmonic analysis techniques. This
book is devoted to introduce the harmonic analysis method for NLE, which
is based on lectures given by B. X. Wang at the Peking University during
the past seven years.
In order to solve an NLE, one needs to choose suitable function spaces as
working spaces. It is well known that most function spaces are established
by using the Lebesgue integrals as basic tools. However, the Lebesgue inte-
gration was not recognized to be of importance in its early stage. As long as
Lebesgue attempted to attend a conference, some mathematicians working
in analysis said to him: “This is not interested for you, we are discussing
the differentiable functions.”, and some experts engaged in geometry told
him: “We are talking the surface which has a tangent plane.” One can
imagine how Lebesgue was hurt inside. New ideas are usually not so easy
to grow up and mathematics brings mathematicians too much sadness and
blessedness, which is hard to express by languages.
However, the authors believe that, at least from the local history point
of view, many important progresses for nonlinear evolution equations have
brought us a series of perpetual surprises over the recent thirty years. The
harmonic analysis techniques of NLE can be gone back to the pioneering
work of R. S. Strichartz in 1977 who discovered the time-space decay of the
solutions of the linear wave equation, so called the Strichartz inequality,
which is now a fundamental tool in the study of nonlinear dispersive equa-
tions. Since 1980s, the NLE, especially the nonlinear dispersive equations
have gained a great development by using the harmonic analysis techniques.
vii
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
viii Preface
Resonance to a Book1
—Xi Zhu
A square of pool likes an opening mirror
Blue sky and white cloud map in it, freely
wander
Why is the pond crystal and bottomed out
It is from the source of alive water
More precisely, we will study a class of nonlinear dispersive equations,
such as nonlinear Schrödinger equations, nonlinear Klein-Gordon eqautions,
KdV equations, as well as the Navier-Stokes equations and the Boltzmann
equation. As the first book of this series we will mainly study the local
and global wellposedness to the Cauchy problem for those equations. In
Chapter 1 we briefly introduce the theories of various function spaces, say
Besov and Triebel-Lizorkin spaces. In Chapter 2 we study the Navier-
Stokes equation by using the Littlewood-Paley decomposition to establish
some time-space estimates for the linear heat equation. Strichartz type
estimates for a class of linear dispersive equations will be systematically
set up in Chapter 3. Applying those Strichartz’ inequalities, in Chapter 4
we will consider the wellposedness of the solutions of nonlinear Schrödinger
and Klein-Gordon equations. In Chapter 5 we introduce the X s,b -method
and study the KdV and derivative Schrödinger equations. In Chapter 6 we
introduce the frequency-uniform decomposition method. It is known that
1A pool indicates a book.
June 29, 2011 9:51 World Scientific Book - 9in x 6in booken-main
Preface ix
Morawetz’ estimates are basic tools for the scattering theory of nonlinear
dispersive equations, which will be summarized in Chapter 7 for nonlinear
Schrödinger equations. Finally, in Chapter 8 we introduce some fundamen-
tal results for the Boltzmann equation. Chapters 1, 2, 3, 4 and 6 are written
by B. X. Wang, Chapter 5 is written by Z. H. Guo and Z. H. Huo, Chapter
7 is written by C. C. Hao and Chapter 8 is written by Z. H. Huo. Many
results in Chapters 2, 3, 4, 5 and 6 have been reproved or simplified by the
authors.
The authors would like to genuinely thank Professors Yulin Zhou, Hes-
heng Sun, Boling Guo, Ling Hsiao, Lizhong Peng and Carlos E. Kenig for
their constant supports and they are grateful to Professors Kong Ching
Chang, Weiyue Ding and Gang Tian for their helps. B. X. Wang thanks
Professors Zhouping Xin and Jiecheng Chen for their invitation to give a
series of lectures based on the book at Hong Kong and Hangzhou, respec-
tively. B. X. Wang deeply cherishes the memory of Professor Tingfu Wang
for his rudimental guidance in mathematics. Thanks are also due to Ms Ji
Zhang for her excellent editorial work.
Contents
Preface vii
s
1. Fourier multiplier, function space Xp,q 1
1.1 Schwartz space, tempered distribution, Fourier transform 1
1.2 Fourier multiplier on Lp . . . . . . . . . . . . . . . . . . . 4
1.3 Dyadic decomposition, Besov and Triebel spaces . . . . . 8
s
1.4 Embeddings on Xp,q . . . . . . . . . . . . . . . . . . . . . 13
s
1.5 Differential-difference norm on Xp,q . . . . . . . . . . . . . 16
s
1.6 Homogeneous space Ẋp,q . . . . . . . . . . . . . . . . . . . 19
1.7 Bessel (Riesz) potential spaces Hps (Ḣps ) . . . . . . . . . . 22
1.8 Fractional Gagliardo-Nirenberg inequalities . . . . . . . . 25
s
1.8.1 GN inequality in Ḃp,q . . . . . . . . . . . . . . . . 25
s
1.8.2 GN inequality in Ḟp,q . . . . . . . . . . . . . . . . 29
2. Navier-Stokes equation 33
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.1 Model, energy structure . . . . . . . . . . . . . . . 34
2.1.2 Equivalent form of NS . . . . . . . . . . . . . . . 34
2.1.3 Critical spaces . . . . . . . . . . . . . . . . . . . . 35
2.2 Time-space estimates for the heat semi-group . . . . . . . 35
2.2.1 Lr → Lp estimate for the heat semi-group . . . . 35
2.2.2 Time-space estimates for the heat semi-group . . 36
2.3 Global well-posedness in L2 of NS in 2D . . . . . . . . . . 39
2.4 Well-posedness in Ln of NS in higher dimensions . . . . . 42
2.5 Regularity of solutions for NS . . . . . . . . . . . . . . . 44
s
2.5.1 Gevrey class and function space E2,1 . . . . . . . 44
xi
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
xii Contents
s
2.5.2 Estimates of heat semi-group in E2,1 . . . . . . . 46
s
2.5.3 Bilinear estimates in E2,1 . . . . . . . . . . . . . . 47
2.5.4 Gevrey regularity of NS equation . . . . . . . . . 48
Contents xiii
xiv Contents
Bibliography 269
Index 281
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Chapter 1
s
Fourier multiplier, function space Xp,q
The intensive study to the nature is the most plentiful source for the
mathematical discovery. —— J. Fourier1
1 Joseph Fourier (1768–1830), a French mathematician and physicist, is best known for
initiating the investigation of Fourier series and their application to problems of the heat
transfer. He is also generally credited with the discovery of the greenhouse effect. He is
one of the authority in French analysis school.
1
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
2 s
Fourier multiplier, function space Xp,q
F −1 (F φ) = φ.
Moreover, F (F −1 ) : S → S is a continuously linear bijection, i.e., an
isomorphism.
σd
λ φ = |λ|
−n b
(σ1/λ φ).
For the proofs of the above propositions, except that the proof of
F −1 φb = φ needs a special technique, say
e\
2
−|x|2 /2 = e−|ξ| /2 ,
4 s
Fourier multiplier, function space Xp,q
2
It is easy to see that u = F −1 e−t|ξ| F u0 =: H(t)u0 is the solution of (1.3).
For any u0 ∈ Lp , we want to know if H(t)u0 also belongs to Lp . More
precisely, does
2
kF −1 e−t|ξ| F u0 kp . ku0 kp , 16p6∞ (1.4)
hold for all u0 ∈ Lp ? The answer is affirmative (see Chapter 2). One can
generalize (1.4) to the following notion.
kF −1 ρF f kp 6 Ckf kp , ∀ f ∈ S, (1.5)
Noticing that (F −1 ρF f, g) = (F −1 ρ ∗ f ∗ e
g )(0), we see that (1.7) implies
0
F −1 ρF g ∈ Lp and ρ ∈ Mp0 with kρkMp0 6 kρkMp . It follows that (1)
holds true.
2 Notice that F −1 ρF f = (F −1 ρ) ∗ f , in view of Proposition 1.6, we see that F −1 ρF f
is a smoothRfunction for any f ∈ S .
3 (f, g) = f (x)g(x)dx.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
6 s
Fourier multiplier, function space Xp,q
8 s
Fourier multiplier, function space Xp,q
some decay at the infinity, and one usually has all the convergence properties of the
summation that one needs. In many applications, one can make the a priori assumption
that f is Schwartz, in which case the convergence is uniform. However, if the function
does not decay, this formula fails. For instance, if f ≡ 1, then all 4k f ’s vanish because
4k 1 = ϕk (0) = ϕ(0) = 0.
P q 1/q
P
7 k(a )k q p :=
k ` (L ) k kak (x)kp , k(ak )kLp (`q ) :=
( k |ak (x)|q )1/q
p .
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
10 s
Fourier multiplier, function space Xp,q
We define
s
Bp,q = f ∈ S 0 (Rn ) : kf kBp,q
s <∞ , (1.35)
X
∞ 1/q
kf kBp,q
s := 2ksq k4k f kqp , (1.36)
k=0
s
Bp,q is said to be the Besov space8 . Assume that
s
Fp,q is said to be the Triebel-Lizorkin space9 . Note that we need replace
the `q -norm by the `∞ -norm in the above definition if q = ∞.
Besov and Triebel-Lizorkin spaces had been formulated during 1960s–
1980s, which have been widely applied in recent years. Roughly speaking,
s is the spatial regularity index; 4k f is the frequency localization of f at
the frequency |ξ| ∼ 2k ; the norms of the function sequence {2sk |4k f |} in
`q (Lp ) and Lp (`q ) generate Besov and Triebel norms of f , respectively.
The dyadic decomposition in the frequency space goes back to the
equivalent norm on Lp (Rn )
kf kp ∼ kf kFp,2
0 , 1 < p < ∞. (1.40)
Proof. Since `p ⊂ `p+a , a > 0, we can easily get the result of (1). Let
us observe that
∞
!1/q2
X
2skq2 |ak |q2 . sup 2(s+ε)k |ak |. (1.44)
k>0
k=0
Taking ak = k4k f kp or ak = |4k f |, we can show that (2) holds with the
help of (1). Finally, we prove (3). Denote bk = 2sk 4k f . We divide the
proof into the following two cases.
Case 1. q 6 p. In view of `q ⊂ `p and Minkowski’s inequality,11 we have
kbk k`p (Lp ) 6 kbk kLp (`q ) 6 kbk k`q (Lp ) , (1.45)
which implies the result.
Case 2. q > p. Analogous to Case 1, we can use Minkowski’s inequality
and `p ⊂ `q to get the result, as desired.
s s s
Proposition 1.14. Let Xp,q be Bp,q or Fp,q . Then
s
(1) Xp,q is a Banach space;
n s
(2) S (R ) ⊂ Xp,q ⊂ S 0 (Rn );
(3) If 1 6 p, q < ∞, then S (Rn ) is dense in Xp,q
s
.
Proof. Since `q (Lp ) and Lp (`q ) are normed spaces, one can conclude that
s
Xp,q is a normed space. In order to show the result of (1), it suffices to
s
prove that Xp,q is complete, whose proof will be left to the end. The proof
of (2) is separated into the following four steps.
s
Step 1. We show that S ⊂ Bp,∞ . In fact, for sufficiently large
L, M, N ∈ N,
kf kBp,∞
s = sup 2sk k4k f kp
k>0
11 Minkowski’s inequalities
P P∞ read
i) kP ∞
j=0 fj kp 6 Pj=0 kfj kp , for any p ∈ [1, ∞];
ii) ∞j=0 kfj kp 6 k
∞
j=0 fj kp , for any p ∈ (0, 1) and fj > 0.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
12 s
Fourier multiplier, function space Xp,q
. k(1 + | · |2 )M (I − 4)L F f k∞
. pN (F f ). (1.46)
Since F : S → S is a continuous map, from (1.46) it follows the result,
as desired.
s s+ε
Step 2. We prove that S ⊂ Xp,q . Since S ⊂ Bp,∞ , by Proposition
s+ε s s s
1.13, we see that Bp,∞ ⊂ Bp,p∧q ⊂ Bp,q ∩ Fp,q . Thus, we have the desired
result.
s
Step 3. We show that Bp,∞ ⊂ S 0 . For convenience, we denote ϕ−1 ≡ 0.
From the construction of ϕk , one has that ϕk ϕk+` ≡ 0 for ` 6= −1, 0, 1. For
s
any f ∈ Bp,∞ , ψ ∈ S , we can take N ∈ N sufficiently large,
∞ X
X 1
|hf, ψi| 6 |h4k f, F ϕk+` F −1 ψi|
k=0 `=−1
∞ X
X 1
. k4k f kp k4k+` ψkp0
k=0 `=−1
∞
X
. kf kBp,∞
s 2−sk k4k ψkp0
k=0
. kf kBp,∞
s kψkB −s+ε
0 p ,∞
. kf kBp,∞
s pN (F ψ). (1.47)
For any bounded set B in S , we have pN (F ψ) . 1 for ψ ∈ B. The result
follows.
s
Step 4. Similarly as in Step 2, we can show that Xp,q ⊂ S 0 and the
details of the proof are omitted.
s
Finally, we prove the completeness of Bp,q . In an analogous way we can
get that Fp,q is also complete. Assume that {f` }∞
s
`=1 is a Cauchy sequence
in Bp,q . By (2) we see that it is also a Cauchy sequence in S 0 . Since
s
s
1.4. Embeddings on Xp,q 13
s
1.4 Embeddings on Xp,q
14 s
Fourier multiplier, function space Xp,q
Proof. Due to
kf kq = λ−n/q kf (·/λ)kq , (1.59)
n
F (f (·/λ)) = λ (F f )(λ·), (1.60)
we see, for any f ∈ LpBλ , that f (·/λ) ∈ LpB1 . In view of (1.59), (1.60) and
Proposition 1.16 we obtain the result, as desired.
Let ϕk be defined by (1.31). Since supp ϕk ⊂ B2k+1 , for any 1 6 p1 6
p2 6 ∞, we have, from (1.58), that Bernstein’s inequality holds
k4k f kp2 . 2n(1/p1 −1/p2 )k k4k f kp1 . (1.61)
Assume that s1 and s2 satisfy
n n
s1 − = s2 − . (1.62)
p1 p2
By (1.61) and (1.62), we immediately have
2s2 k k4k f kp2 . 2s1 k k4k f kp1 . (1.63)
Taking the `r -norm in both sides of (1.63), one has that
kf kBps2,r . kf kBps1 ,r . (1.64)
2 1
s
1.4. Embeddings on Xp,q 15
. 1. (1.71)
Now we estimate II. By Corollary 1.1,
k4k f k∞ . 2kn/p1 k4k f kp1 . 2k(n/p1 −s1 ) kf kFps1,∞ . (1.72)
1
16 s
Fourier multiplier, function space Xp,q
. 1. (1.75)
The estimates of I and II imply the conclusion.
s
Proposition 1.17. Let 1 6 p < ∞, s > n/p and 1 6 q 6 ∞. Let Xp,q be
s s
Bp,q or Fp,q . Then
s 0
Xp,q ⊂ B∞,1 ⊂ L∞ . (1.76)
Proof. Using the dyadic decomposition and (1.61), we have
∞
X X∞
kuk∞ 6 k4k uk∞ . 2kn/p k4k ukp
k=0 k=0
∞
!
X
. 2k(n/p−s) kukBp,∞
s 6 kukXp,q
s , (1.77)
k=0
which is the result, as desired.
s
1.5 Differential-difference norm on Xp,q
s
1.5. Differential-difference norm on Xp,q 17
kF −1 ρm N N
h ϕk F ∂xj f kp . k4k ∂xj f kp . (1.85)
Observing the following identity
m
ρm h ξ
ρm
h ϕk =
h
, (|h||ξ|)m ϕk (1.86)
hh, ξim |h| |ξ|
and noticing that kρh /hh, ξikMp (Rn ) = k(eiξ − 1)/ξkMp (R) < ∞, it follows
that for all k > 1,
kρm
h ϕk kMp . 2
km
|h|m . (1.87)
Obviously, (1.87) also holds for k = 0. Hence, in view of (1.84), we have
kF −1 ρm N
h ϕk F ∂xj f kp . 2
km
|h|m k4k ∂xNj f kp . (1.88)
Applying the technique as in (1.84), we have from (1.15) that
k∂xNj 4k f kp . 2N k k4k f kp . (1.89)
Collecting (1.85), (1.88) and (1.89), we get (1.83). Inserting (1.83) into
(1.82), one has that
∞
X
2i(s−N ) ωpm (2−i , ∂xNj f ) . 2(i−k)(s−N ) (1 ∧ 2(k−i)m )2sk k4k f kp . (1.90)
k=0
12 Notice that Mh and 4k are different operators.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
18 s
Fourier multiplier, function space Xp,q
Clearly, kf kp . kf kBp,q
s . Thus, the right hand side of (1.80) can be con-
kF −1 ρm N m −k
jk F ∂xj f kp . ωp (2 , ∂xNj f ), k > 1. (1.93)
∞
!1/q
n
X X q
kf kBp,q
s . kf kp + 2k(s−N ) ωpm (2−k , ∂xNj f ) , (1.94)
j=1 k=1
which implies that the left hand side of (1.80) can be controlled by its right
hand side.
Finally, we prove (1.92). We need the following lemma
s
1.6. Homogeneous space Ẋp,q 19
Now we can finish the proof of (1.92). By Lemma 1.1, for any k > 1,
n
X
k4k f kp 6 kF −1 ρ−m
jk ϕk χj (2
−k
·)ξj−N F (F −1 ρm N
jk F ∂xj f )kp . (1.98)
j=1
By Proposition 1.11,
(eiξj − 1)−m ϕχj ξj−N ∈ Mp , (1.99)
where ϕ is as in (1.30). It follows from Propositions 1.10 and (1.98) that
(1.92) holds.
As a consequence of Proposition 1.18, we have
where Mh :=M1h .
s
1.6 Homogeneous space Ẋp,q
20 s
Fourier multiplier, function space Xp,q
X
∞ 1/q
kf kḂ s := 2ksq k4k f kqp , (1.107)
p,q
k=−∞
s
Ḃp,q is said to be a homogeneous Besov space. Let
−∞ < s < ∞, 1 6 p < ∞, 1 6 q 6 ∞. (1.108)
Define
s
Ḟp,q ˙ 0 n
= f ∈ S (R ) : kf kḞ s < ∞ , (1.109)
p,q
∞ 1/q
X ksq
kf kḞ s :=
2 |4 k f | q
,
(1.110)
p,q
k=−∞ p
s
Ḟp,q is said to be a homogeneous Triebel-Lizorkin space.
s s s
If there is no confusion, we will write Ẋp,q to stand for Ḃp,q or Ḟp,q .
Using the dilation, one has that
kf (2` ·)kẊp,q
s = 2`(n−s/p) kf kẊp,q
s , (1.111)
s
from which one can easily understand why we call Ẋp,q as a homogeneous
s s
space. Ẋp,q and Xp,q are quite similar. In the following we only state some
s
results on Ẋp,q , whose proofs follow an analogous way as the nonhomoge-
neous case. If there is no explanation, we will always assume that conditions
s s
(1.105) and (1.108) are satisfied for the spaces Ḃp,q and Ḟp,q , respectively.
s s s s
Proposition 1.19. Let Ẋp,q = Ḃp,q (or Ẋp,q = Ḟp,q ). Then
s
(1) Ẋp,q is a Banach space;
(2) ˙
S ⊂ Ẋp,qs
⊂ S˙ 0 ;
(3) If 1 6 p, q < ∞, then S˙ (Rn ) is dense in Ẋp,qs
;
s s
(4) If q1 6 q2 , then Ẋp,q1 ⊂ Ẋp,q2 ;
s s s
(5) Let 1 6 p < ∞. Then Ḃp,p∧q ⊂ Ḟp,q ⊂ Ḃp,p∨q .
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
s
1.6. Homogeneous space Ẋp,q 21
In the sequel, we will frequently use Theorem 1.3, where the first inclu-
sion can be shown in the same way as in Besov space, the second one is
slightly different from the Triebel-Lizorkin space, see Appendix.
22 s
Fourier multiplier, function space Xp,q
N
!1/q
X Y
σkq
6 2 k4k vkθpiiq
k∈Z i=1
N
!1/q
XY
kσi θi
= (2 k4k vkθpii )q
k∈Z i=1
N
Y
6 kvkθḂiσi .
pi ,qi
i=1
σ
The case Ḟp,q is similar.
Recall that Js = (I − ∆)s/2 and Is = (−∆)s/2 are said to be the Bessel and
the Riesz potentials, respectively. Assume that
1 < p < ∞, −∞ < s < ∞. (1.112)
Denote
n o
Hps = f ∈ S 0 : kf kHps := kJs f kp < ∞ ; (1.113)
n o
Ḣps = f ∈ S˙ 0 : kf kḢps := kIs f kp < ∞ . (1.114)
Hps and Ḣps are said to be Bessel and Riesz potential spaces, respectively.
In the following we give the equivalent norms on Hps and Ḣps :
Remark 1.1. The results in (1)–(3) of Proposition 1.22 also hold for Riesz
potential spaces. The proof of Proposition 1.22 follows from Theorem 1.5
s
and the corresponding results of Fp,2 .
Proposition 1.23. Let s > 0 and 1 < p < ∞. We have Hps = Lp ∩ Ḣps .
Moreover, if s is an integer, then
X
kf kḢ s ∼ kDα f kp .
p
|α|=s
NY
+1
N
Y +1
|Dαi ui |ρi
6 C kui kρḢisi . (1.118)
p pi
i=0 i=0
Noticing that 1/ρi qi − |αi |/n = 1/pi − si /n, and using Theorem 1.6, we get
|α |
Ḣpsii ⊂ Ḣρi qii . By (1.119), we have the desired result.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
24 s
Fourier multiplier, function space Xp,q
s
1.8.1 GN inequality in Ḃp,q
kukḂ s . kuk1−θ
s
Ḃ 0
kukθḂ s1 (1.129)
p,q p0 ,q0 p1 ,q1
∗
Using the inclusion Ḃps∗ ,q ⊂ Ḃp,q
s
, we get the conclusion.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
26 s
Fourier multiplier, function space Xp,q
s
The result below clarify how the third indices q in Ḃp,q contribute the
validity of the GN inequalities.
kukḂ s . kuk1−θ
s
Ḃ 0
kukθḂ s1 (1.132)
p,q p0 ,∞ p1 ,∞
X q2
kukḂp,q
s 6 (N1s ...NKs
k4N1 ukp ...k4NK ukp )
N1 >...>NK
q(1−q)
× (N1s . . . NK
s
k4N1 ukp . . . k4NK ukp ) . (1.140)
13 Here 4N is different from the notation as in Section 1.6, which is identical with 4logN
2
as in (1.103).
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
where
− n + n −s1 +s −n n
a(− n + n −s1 +s)
p + p1 −s1 +s
Λ(N1 , . . . NK ) = N1 p p1 . . . Nk−1 Nk p p1
(1−a)(− n n
−n n
− n + n +s q(1−q)
p + p0 +s) p + p0 +s
× Nk Nk+1 . . . NK p p 0 .
(1.144)
By (1.143), we have
X K
X
kukḂp,q
s . Λ(N1 , . . . NK ) (Nis k∆Ni ukp )q (1.145)
N1 >...>NK i=1
(1−q)θ (1−θ)(1−q)
× kukḂ s1 kukḂ 0 . (1.146)
p1 ,∞ p0 ,∞
Nk−1 >Nk
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
28 s
Fourier multiplier, function space Xp,q
. Nk k∆Nk ukqp
Nk
Nk−1 >Nk
. kukqḂ s . (1.148)
p,q
1−θ
+ θ
kf kḂp,∞
s+ε . kf k s0
Ḃ
kf kḂ+s1 . (1.152)
p0 ,∞ p1 ,∞
We easily see that θ = (θ+ + θ− )/2. Inserting (1.151) and (1.152) into
(1.150), we have the result, as desired.
We omit the proof of the necessity, one can refer to [99] for details.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
s
1.8.2 GN inequality in Ḟp,q
s
In homogeneous Triebel-Lizorkin spaces Ḟp,q , we have the following (cf.
[99]):
30 s
Fourier multiplier, function space Xp,q
Remark 1.2.
(1) In [5], the Qα space was introduced for which the norm is given by
Z Z 1/2
|f (x) − f (y)|2
kf kQα (Rn ) = sup n+2α
dxdy < ∞,
I I I |x − y|
where I ranges over all cubes in Rn . Xiao [253] considered the well
posedness of the Navier-Stokes equation in Q type spaces.
(2) Yang and Yuan [254] gave a unified way to handle the Qα and
Triebel-Lizorkin spaces. Indeed, for any p ∈ (1, ∞), q ∈ (1, ∞],
s,τ
τ, s ∈ R, they introduced the space Ḟp,q in the following way
p/q 1/p
Z
1 X
∞
sj q
kf kḞp,q = sup
s,τ
τ (2 |4j f (x)|) dx ,
P dyadic |P |
P j=− logl(P
2
)
where l(P ) denotes the side length of the cube P . They showed
α,1/2−α/n s,τ
that Qα = Ḟ2,2 . In a similar way as Ḟp,q , Yang and Yuan
s,τ
[255] introduced Besov type space Ḃp,q . Some recent progress on
s,τ s,τ
Bp,q and Fp,q spaces can be found in [257].
s
(3) Function space like Ḟ∞,q are not considered in this chapter. It is
of importance in the case s = 0 and q = 2, which is equivalent
to BMO space introduced by F. John and L. Nirenberg in 1961.
An interesting generalization of BMO type on homogeneous-type
spaces or on measurable subsets of such spaces was given by X.
T. Duong and L. X. Yan [69]. Their spaces are defined by certain
maximal functions associated with generalized approximation of
the identity, which coincide with the classical BMO spaces of F.
John and L. Nirenberg in some special cases.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Chapter 2
Navier-Stokes equation
33
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
34 Navier-Stokes equation
2.1 Introduction
36 Navier-Stokes equation
2
Proof. Noticed that e−t|ξ| is exponentially decaying, in order to exten-
sively use this fact, the frequency localization techniques will be applied.
We have
2
k4j H(t)f kr 6 kϕj e−t|ξ| kMr kf kr . (2.17)
−j
Using the multiplier criteria and noticing that ϕj = ϕ(2 ξ), suppϕ ⊂ {ξ :
1/2 6 |ξ| 6 2}, one has that
2 2j
|ξ|2 2j
|ξ|2
kϕj e−t|ξ| kMr = kϕe−t2 kMr 6 kϕe−t2 kH L , (2.18)
where L > n/2. Due to supx>0 xk /ex 6 C, it is easy to see that
2j
|ξ|2 2j
kϕe−t2 kH L . e−ct2 . (2.19)
So, by (2.17)–(2.19) we get
2j
k4j H(t)f kr . e−ct2 kf kr . (2.20)
P
Since 4j = 4j ( `=0,±1 4j+` ), (2.20) implies that
2j
k4j H(t)f kr . e−ct2 k4j f kr . (2.21)
Taking the `λ -norm in (2.21), we obtain that
1/λ
X 2j
kH(t)f kḂ 0 . e−ct2 k4j f kλr . (2.22)
r,λ
j
In what follows the proof is separated into two cases. The first case is
that γ > λ. Taking the norm in Lγt (R+ ) on (2.22) and using Minkowski’s
inequality, we have
∞
1/λ
X −ctλ22k
kH(t)f kLγ (R+ ,Ḃ 0 ) .
e k4 k f k λ
r
γ/λ
r,λ
k=−∞ Lt
X
∞
1/λ
−ctλ22k
.
e
γ/λ
k4k f kλr . (2.23)
Lt
k=−∞
Noticing that
−ctλ22k
e
γ/λ
. 2−λ2k/γ , (2.24)
Lt
from (2.23) and (2.24) we get the consequence in the case p = r and γ > λ.
−2/γ+n(1/r−1/p) −2/γ
As r < p, applying the inclusion Ḃr,λ ⊂ Ḃp,λ , we can obtain
the result.
Secondly, we consider the case γ < λ. By (2.22),
Z X
kH(t)f kγḂ 0 dt . 2−2j k4j f kγp . (2.25)
R+ p,λ
j
We get the result in the case p = r and γ < λ. If r < p, in an analogous
way to the first case one can prove the conclusion.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
38 Navier-Stokes equation
In fact, the above estimates (2.15) and (2.16) can be developed to the
case 0 < r 6 p 6 ∞ and we have no restriction on λ, γ > 0. We have
a little bit of surprise about this fact, see [240] for details. Some earlier
results related to (2.16) were due to Weissler [250] and Giga [77], where
their proof is based on Marcinkiewicz’ interpolation theorem.
Corollary 2.1. Let 2/γ(p) = n(1/2 − 1/p). For any 2 6 p < ∞, we have
k∇H(t)f kL2 (R+ ;L2 ) . kf kL2 , (2.26)
kH(t)f kLγ(p) (R+ ;Lp ) . kf kL2 . (2.27)
0
Proof. Taking λ = r = 2 in Proposition 2.2 and noticing that Ḃp,2 ⊂ Lp ,
one immediately has the consequence.
For simplicity, we write
Z t
(A f )(t, x) := H(t − τ )f (τ, x)dτ. (2.28)
0
Proposition 2.3 can not handle the case γ = ∞, but we have (see [240])
Corollary 2.2. Let 2/γ(p) = n(1/2 − 1/p). For any 2 6 p < ∞, 2/γ(p) <
1, we have
k∇A f kL2 (R+ ;L2 ) . kf kLγ(p)0 (R+ ,Lp0 ) , (2.35)
kA f kL∞ (R+ ,L2 ) ∩ Lγ(p) (R+ ;Lp ) . kf kLγ(p)0 (R+ ,Lp0 ) . (2.36)
40 Navier-Stokes equation
Theorem 2.1. Let u0 ∈ [L2 ]20 . Then the NS equation (2.5) has a unique
solution u satisfying
and
Z t
1 1
ku(t)k22 + k∇u(s)k22 ds = ku0 k22 , 0 < t < ∞. (2.56)
2 0 2
The contraction mapping argument implies that the solution map is
analytic and of course, is continuous and the persistence of the regularity
of solutions is also easy to obtain by using the integral equation. So the NS
equation is globally well-posed in L2 .
Finally, we point out that the method used here can be applied to some
other kinds of nonlinear parabolic equations. For examples,
42 Navier-Stokes equation
For convenience, we still write Lpx,t∈[0,T ] = Lp (0, T ; Lp (Rn )) and use the
same notations as in the previous section. By Corollaries 2.1 and 2.2, we
have for n > 3,
k∇H(t)u0 kL2 (0,T ; L2 ) . ku0 kL2 , (2.57)
kH(t)u0 kL2+4/n . ku0 kL2 , (2.58)
x,t∈[0,T ]
One may expect that the techniques in the previous section also work very
well. Unfortunately,
k(u · ∇)ukL(2+4/n)0 6 k∇ukL2x,t∈[0,T ] kukLn+2 . (2.61)
x,t∈[0,T ] x,t∈[0,T ]
Ln+2
x,t∈[0,T ] is out of the control of the above linear estimates. So, we need
the following
2/(2+n) 0 2/(2+n)
Using the inclusions Ḃn,n = Ḟn,n ⊂ Ḟn+2,2 = Ln+2 , we see that
(2.66) implies that (2.62) holds. Obviously, we have (2.63).
(2.64) is a straightforward consequence of Proposition 2.3. Taking r =
λ = (n + 2)/2 and q 0 = (n + 2)/2 in Proposition 2.4, we get
kA f kL∞ (R+ ; Ḃ 2n/(2+n) )
. kf kL(n+2)/2 (R+ ; Ḃ 0 ). (2.67)
(n+2)/2, (n+2)/2 (n+2)/2, (n+2)/2
This implies
k∇A f kL∞ (R+ ; Ḃ (n−2)/(2+n) )
. kf kL(n+2)/2 . (2.68)
(n+2)/2, (n+2)/2 x,t∈R+
(n−2)/(2+n) 0(n−2)/(2+n)
Using the embedding Ḃ(n+2)/2, (n+2)/2 = Ḟ(n+2)/2, (n+2)/2 ⊂ Ḟn,2 = Ln , we
see that (2.68) implies that (2.65) holds.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
The above linear estimates can derive the local well-posedness and the
global well-posedness with small data in Ln for n > 3. We will use the
same notations as in the case n = 2. We have
Theorem 2.2. Let u0 ∈ [Ln ]n0 . Then there exists a Tm > 0 such that the
NS equation (2.5) has a unique solution u satisfying
u ∈ C([0, Tm ); [Ln ]n0 ) ∩ L2+n
loc (0, Tm ; [L
2+n n
]0 ). (2.69)
If Tm < ∞, then we have kukL2+n(0,Tm ; L2+n ) = ∞. If ku0 kn is sufficiently
small, then Tm = ∞. Moreover, if u0 ∈ [L2 ]n0 , then u ∈ C(0, Tm ; [L2 ]n0 ),
∂xi u ∈ L2 (0, Tm ; [L2 ]n0 ), and
Z t
1 1
ku(t)k22 + k∇u(s)k22 ds = ku0 k22 , 0 < t < Tm . (2.70)
2 0 2
Proof. Put
n o
D = u : kukL2+n 6 δ, kukL∞ ([0,T ];Ln ) 6 2Cku0 kn , (2.71)
x,t∈[0,T ]
. ku0 kn + δ 2 . (2.75)
If Cδ 6 1/4, we can show that M is a contraction mapping from D into
itself. So, there exists a u satisfying
u(t) = H(t)u0 + A P∇ · (u ⊗ u). (2.76)
By a standard argument, we see that u is unique in L2+n (0, T ; [L2+n ]n0 ).
Moreover, one can extend the solution step by step and find a maximal Tm
such that u ∈ C([0, Tm ); [Ln ]n0 ) ∩ L2+n
loc (0, Tm ; [L
2+n n
]0 ).
If u0 ∈ [L ]0 , then we can use (2.57)–(2.61) to obtain u ∈ C(0, T ; [L2 ]n0 )
2 n
and ∂xi u ∈ L2 (0, T ; [L2 ]n0 ) in a similar way as in the 2D case. According
to the regularity theory (see [67] for instance), we see that u is infinitely
smooth in the domain (0, T ) × Rn . So, a straightforward computation will
lead to (2.2). Moreover, if ku0 kn is small enough, we can take T = ∞ in
(2.71) and (2.72).
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
44 Navier-Stokes equation
The result of Theorem 2.2 has a long history. Weissler [251] gives a de-
tailed Lp -theory in half-space for local solutions. Kato [119] used a different
way to obtain the well posedness result in Ln by constructing the resolution
norm like supt tθ ku(t)kp . The idea only taking Ln+2 x,t as a resolution space
was hidden in some references, see for instance [67]. Here the proof is to
take Ln+2 n
x,t∈[0,T ] ∩ C([0, T ]; L ) as a resolution space which enables us to get
the local well posedness of solutions in Ln , where (2.65) plays a crucial role.
Similar bilinear estimate to (2.65) in 3D was shown in [70] in a different
way:
kA P div (u ⊗ v)kL∞ ([0,T ];L3 ) . kukL5x,t kvkL5x,t . (2.77)
In 3D case, Kenig and Koch [123] obtained the result of Theorem 2.2 by
applying (2.77).
We will show that, for any t > 0, the solution of the NS equation is infinitely
differentiable and in fact, is really analytic.
s
2.5.1 Gevrey class and function space E2,1
First, we consider the Gevrey classes. Let s > 0, denote
s
n ∞ n ∃ m!
Gs (R ) = f ∈ C (R ) : ρ, M > 0 s.t. kf kḢ m 6 M ∀m ∈ Z+ .
ρm
(2.78)
Gs (Rn ) is said to be a Gevrey s-class.
Our aim is to show that the solution of the NS equation belongs to the
Gevrey 1-class. In order to show this fact, we will use Wiener’s decomposi-
tion of Rn . In Chapter 6, we will continuously use Wiener’s decomposition
to study the nonlinear dispersive equations. Let Qα be the unit cube with
the center at α ∈ Zn , i.e., Qα = α + Q0 , Q0 = {x = (x1 , ..., xn ) : −1/2 6
xi < 1/2}. It is easy to see that
Rn = ∪α∈Zn Qα , Qα ∩ Qβ = ∅, α 6= β.
n
So, {Qα } constitutes a decomposition of R , which is said to be the uniform
decomposition (or Wiener’s decomposition) of Rn . Comparing it with the
dyadic decomposition, we see that it is more delicate than the dyadic de-
composition of Rn . Putting the uniform decomposition into the frequency
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
space and combining it with the `1 (L2 ) space, we have the following
( )
X
s 0 n s|k|
−1 b
E2,1 = f ∈ S (R ) : kf kE2,1
s = 2
F χQk f
< ∞ . (2.79)
2
k∈Zn
s
Such a kind of space and its generalized version Ep,q were studied in [247],
which is a kind of generalized modulation spaces [86]. It is easy to see
s
that E2,1 is a Banach space, which has an exponential regularity weight.
s
From PDE’s point of view, E2,1 has infinite smoothness, which is different
from any Besov spaces and any classical modulation spaces. The classical
s
modulation space M2,1 takes the following norm:
( )
X
s 0 n
M = f ∈ S (R ) : kf kM s =
2,1 hki
F χQ fb
< ∞ , (2.80)
2,1
s −1
k
2
k∈Zn
s s
where hki = 1 + |k|. M2,1 and its generalizations Mp,q have finite smooth-
s
ness, which are similar to Besov spaces (see Chapter 6). E2,1 is of impor-
tance in our regularity argument for the parabolic equation, whose infinite
regularity can be seen in the following (cf. [247]):
X
. kχQk |ξ|m fb k2
k∈Zn
X p
. (|k| + n/2)m kχQk fb k2
k∈Zn
m! X (cs)m (|k| + C)m s|k|
. 2 kχQk fb k2
(cs)m n
m! · 2s|k|
k∈Z
m!
. kf kE2,1
s .
(2s)m
On the other hand, let f ∈ G1 and L > n/2. Using Taylor’s expansion and
Hölder’s inequality,
X
kf kE2,1
s = 2s|k| kχQk fb k2
k∈Zn
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
46 Navier-Stokes equation
X∞ X
(s ln 2)m
6 kf k2 + |k|m kχQk fb k2
m=0
m!
k∈Zn , |k|1
∞
X m X
(s ln 2)
. kf k2 + C m |k|−L kχQk ∇\
m+L f k
2
m=0
m!
k∈Zn , |k|1
∞
X m
(Cs)
. kf k2 + kf kḢ L+m
m=0
m!
X ∞
(Cs)m (m + L)!
. kf k2 + · .
m=0
m! ρm
P∞
Choosing 0 < s 1, we see that m=0 (Cs/ρ)m (m + L)L is a convergent
s
series. Thus, f ∈ E2,1 .
s
2.5.2 Estimates of heat semi-group in E2,1
Using Proposition 2.5, we want to get the solution of the NS equation
s
u ∈ G1 . It suffices to show that u belongs to E2,1 (s > 0). Hence, one
needs to estimate kH(t)u0 kE2,1
s and kA f kE2,1
s .
Proposition 2.6. Let H(t) = et∆ . There exists a c > 0 such that for
j = 0, 1 and T 6 1,
k∇j H(t)u0 kE2,1ct . t
−j/2
ku0 kE2,1
0 , t 6 T, (2.82)
Z t
j
sup
∇ H(t − τ )f (τ )dτ
. (T + T 1−j/2 ) sup kf (τ )kE2,1
cτ .
t∈[0,T ] 0 ct
E2,1 τ ∈[0,T ]
(2.83)
Proof. For convenience, we will use the notation
k = F −1 χQk F . (2.84)
By Plancherel’s identity, for |k| > 1,
2 2
kk H(t)u0 k2 = kχQk (ξ)e−t|ξ| F u0 k2 6 e−ct|k| kk u0 k2 . (2.85)
If k = 0, we see that (2.85) also holds for t 6 1. Taking the summation
over all k ∈ Zn , we can get (2.82) hold for j = 0.
We now prove (2.82) for j = 1. If k = 0 and t 6 1,
2
k∇k H(t)u0 k2 . kχQk (ξ)e−t|ξ| F u0 k2 6 kk u0 k2 . (2.86)
For |k| > 1,
2
k∇k H(t)u0 k2 . k|ξ|χQk (ξ)e−t|ξ| F u0 k2 6 t−1/2 e−ct|k| kk u0 k2 . (2.87)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
s
2.5.3 Bilinear estimates in E2,1
By Proposition 2.6, our goal is to show that the solution of the NS equation
ct
belongs to E2,1 , it is necessary to estimate ku ⊗ ukE2,1
ct . We have
λ
Proposition 2.7. E2,1 is a Banach algebra. More precisely, we hace
kuvkE2,1
λ 6 C2Cλ kukE2,1
λ kvkE λ ,
2,1
(2.90)
λ
where C is independent of λ > 0 and u, v ∈ E2,1 .
Proof. Noticing that for k = (k1 , ..., kn ) ∈ Zn with |k| = |k1 | + ... + |kn |,
we have
X
kuvkE2,1
λ = 2λ|k| k k (uv)k2 . (2.91)
k∈Zn
48 Navier-Stokes equation
Remark 2.1.
(1) By Theorem 2.3, we can get that the solution of the NS equation
belongs to C ∞ ((0, Tm ) × Rn ).
(2) Theorem 2.3 implies that the solution of the NS equation belongs
to the Gevrey 1-class and so, is really analytic.
(3) Theorem 2.3 also describes the disappearing process of the regu-
larity of the solutions to the NS equation.
Remark 2.2. The Gevrey regularity for the evolution equations is of im-
portance for its own sake. The Gevrey regularity of the weak solutions for
a class of linear and semi-linear Fokker-Planck equations
(∂t + v · ∇x − ∆v )u = F (t, x, v, u, ∇v u)
was recently studied by Chen, Li and Xu [35], see also [37] and references
therein for a class of the linear model of spatially inhomogeneous Boltzmann
equations without an angular cutoff.
Remark 2.3. As the end of this chapter, we state some recent progress on
NS equation without proofs.
(1) Let us mention the result by Koch and Tartaru [151] where the
global solutions for NS in 3D are obtained with the small data in
the space BM O−1 with the norm:
Z !1/2
−3/2 t∆
kukḂ∞,∞
−1 + sup R |e u(y)|dydt ,
x∈R3 , R>0 [0,R2 ]×{y:|x−y|6R}
see also Chemin and Gallagher’s generalizations in [30; 31; 32] for
a class of large data.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
50 Navier-Stokes equation
(2) Recently, Escauriaza, Seregin and Šverák [70] showed that any
“Leray-Hopf” weak solution in 3D which remains bounded in
L3 (R3 ) cannot develop a singularity in finite time. Their proof
used a blow-up procedure and reduction to a backwards unique-
ness question for the heat equation, and was then completed using
Carleman-type inequalities and the theory of unique continuation.
Kenig and Koch [123] gave an alternative proof by substituting L3
with Ḣ 1/2 . Dong and Du [68] generalized their results in higher
spatial dimensions n > 3.
(3) Noticing that L3 ⊂ B∞,∞ −1
in 3D is a sharp embedding, for any
solution u of the NS equation in C([0, T ∗ ); L3 ), we see that u ∈
C([0, T ∗ ); B∞,∞
−1
). May [163] prove that if T ∗ < ∞, then there
exists a constant c > 0 independent of the solution of NS equation
such that
lim sup ku(t) − ωkB∞,∞
−1 >c
t→T ∗
Chapter 3
∂t u − iP (D)u = F, (3.1)
√
where u(t, x) is an unknown function, i = −1, P\ (D)u = P (ξ)b u, and
P (ξ) is a real-valued function, which is said to be the dispersion relation of
(3.1)1 . If F is a nonlinear function of u, then (3.1) is said to be a nonlinear
dispersive equation.
In this chapter we study the time-space estimates for the solutions of
the linear dispersive equation in mixed Lebesgue spaces Lqt Lpx (or more
general spaces Lq (0, T ; Bp,r
s
)), so called the Strichartz inequalities, which is
a starting point to the study of nonlinear dispersive equations. For instant,
we consider the Schrödinger equation2
ib b = fb,
ut − |ξ|2 u b(0) = u
u b0 .
1 We emphasize that P (·) must be a real function.
2 Erwin Schrödinger (1887-1961) was an Austrian theoretical physicist who achieved
fame for his fundamental contributions to quantum mechanics, especially the Schrödinger
equation, for which he received the Nobel Prize in 1933.
51
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
0
3.1 Lp → Lp estimates for the dispersive semi-group
2
Let us consider the Schrödinger semi-group eit∆ := F −1 eit|ξ| F . Since
Z
|x−y|2
eit∆ u0 = ct−n/2 e 4it u0 (y)dy, (3.4)
Rn
1 ∞
we immediately have the L → L decay
keit∆ u0 k∞ . t−n/2 ku0 k1 .
Taken noticing of keit∆u0 k2 = ku0 k2 , an interpolation yields
keit∆u0 kp . t−n/2 ku0 kp0 , p > 2, 1/p + 1/p0 = 1.
0
This is the fundamental Lp → Lp estimate, or the time decay estimate for
the Schrödinger semi-group. For the general semi-group, it is not expected
to have an analytic expression as in (3.4), we need to look for other ways
0
to get the Lp → Lp estimates.
0
In this section we study the Lp → Lp estimates for a class of disper-
sive semi-groups U (t) := F −1 exp(itP (ξ))F and use two different ways to
0
consider its Lp → Lp decay. The first method is very effective for homo-
geneous functions P (·), the second method can deal with nonhomogeneous
radial functions P (·).
First, we consider the decay estimates for the semi-group Um (t) :=
F −1 exp(it|ξ|m )F . Some earlier decay estimates on Um (t) were obtained
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
0
3.1. Lp → Lp estimates for the dispersive semi-group 53
0
by Pecher [191]. Following Pecher’s proof, we can get an Lp → Lp decay
estimate with smooth effects in the case m > 2. Some nontrivial generalized
dispersive Lq − Lp estimates can be founded in Cui [52; 53], Sugimoto [207;
208] (see also [195]).
We need the following lemma, which is due to Littman [161].
P1
Substituting f by `=−1 4k+` f in (3.11), we get
1
X
2−2σ(m,p)k kUm (t)4k f kp . t−ρ(1/2−1/p) k4k+` f kp0 . (3.13)
`=−1
s s
Using the inclusion Ḃp,2 ⊂ Ḟp,2 = Ḣps and Ḃps0 ,2 ⊃ Ḟps0 ,2 = Ḣps0 , we have
kUm (t)f kḢ −2σ(m,p) . t−ρ(1/2−1/p) kf kp0 , 2 6 p < ∞. (3.15)
p
2
Proposition 3.2. Let n > 1, S(t) = F −1 eit|ξ| F , 2 6 p < ∞, 1 6 q 6 ∞,
and 1/p + 1/p0 = 1. Then we have
kS(t)f kḂ 0 . t−n(1/2−1/p) kf kḂ 00 , (3.18)
p,q p ,q
Noticed that for m > 2, 2σ(m, p) = n(2 − m)(1/2 − 1/p) < 0, Um (t) gains
some regularities in Proposition 3.3.
Below, we consider another decay estimate of Um (t) without the smooth-
ness in the case m > 2. We have
0
3.1. Lp → Lp estimates for the dispersive semi-group 55
We write 4−1 = 0. It follows that (3.23) holds for all k > 0. Taking the `2
in (3.23), one has that
kUm (1)f kBp,2
0 . kf kB 00 . (3.25)
p ,2
The above method is also valid for some other homogeneous non-radial
functions, say P (ξ) = ξ14 ± ... ± ξk4 , k 6 n. Noticing that, if P (ξ) is not
a homogeneous function, (3.7) can not be obtained by scaling, we need to
look for another way to handle the nonhomogeneous case.
Our idea is to simplify P (·) as a radial function P (ξ) := P (|ξ|), which
is essentially reduced to one dimensional case. P (|ξ|) can be separated into
two parts, lower and higher frequency parts, which correspond to |ξ| . 1
and |ξ| 1, respectively. We further assume that P (ξ) has a different
growth as |ξ| . 1 and |ξ| 1, which is sufficient for many semi-groups.
In what follows, we always assume that P : (0, ∞) → R is a smooth
radial function satisfying
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Remark 3.1. The following are somep examples for which conditions (H1)–
(H4) are satisfied: (1) P (ξ) = 1 + |ξ|2 (relevant to the Klein-Gordon
4 2
semi-group); (2) P (ξ) = |ξ|
p + |ξ| (relevant to the fourth order Schrödinger
semi-group); (3) P (ξ) = 1 + |ξ|4 (relevant
p to the beam semi-group).
Comparing P (|ξ|) = |ξ| with P (|ξ|) = 1 + |ξ|2 , althoughp
they have the
2
same growth as t → ∞, we easily see that the rank of (∂ij 1 + |ξ|2 )n×n
p
2
is n and 1 + |ξ| is better than |ξ| at ξ = 0. So, we can expect that the
0
Klein-Gordon semi-group has a better Lp → Lp decay.
We first consider 1D case, which is easier to find the ideas. The following
result is due to [94].
0
3.1. Lp → Lp estimates for the dispersive semi-group 57
Lemma 3.2. Let ϕ ∈ C0∞ (R), and P ∈ C 2 (R) satisfy |P 00 (ξ)| > λ > 0 for
all ξ ∈ supp ϕ. Then
Z
eiP (ξ) ϕ(ξ)dξ . λ−1/2 (kϕk∞ + kϕ0 k1 ).
We first consider the case m2 < 2. Since min(1/m2 , 1/2) = 1/2, from (b)
it follows that
X
kU (t)40 u0 k∞ . |t|−1/2 2k(1−m2 /2) ku0 k1 . |t|−1/2 ku0 k1 . (3.32)
k<0
Proposition 3.7. Let n > 2. Denote U (t) = F −1 eitP (|ξ|) F . We have the
following decay estimates.
(a) Let {4k }k∈Z be as in Sec. 1.6, k > 0, and P satisfy (H1). Then
n−1
kU (t)4k u0 k∞ . |t|−θ 2k(n−m1 θ) ku0 k1 , 0 6 θ 6 . (3.34)
2
Moreover, if P satisfies (H3), then
m1 n α1 −m1
kU (t)4k u0 k∞ . |t|−n/2 2k(n− 2 − 2 )
ku0 k1 . (3.35)
(b) Let {4k }k∈Z be as in Sec. 1.6, k < 0, and P satisfy (H2). Then
we have
n−1
kU (t)4k u0 k∞ . |t|−θ 2k(n−m2 θ) ku0 k1 , 0 6 θ 6 . (3.36)
2
Moreover, if P satisfies (H4), then
m2 n α2 −m2
kU (t)4k u0 k∞ . |t|−n/2 2k(n− 2 − 2 )
ku0 k1 . (3.37)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
0
3.1. Lp → Lp estimates for the dispersive semi-group 59
Proof. The idea is to use the Bessel function. After making the polar
coordinate transform, the estimate is reduced to an oscillating integration
in one spatial dimension. Next, using the decay and cycle properties of the
Bessel function, analogous to the 1D case, we can get the conclusion. Let
Jm (r) be the Bessel function
Z 1
(r/2)m
Jm (r) = eirt (1 − t2 )m−1/2 dt, m > −1/2.
Γ(m + 1/2)π 1/2 −1
Let us state some properties on the Bessel functions, see [82] and [203].
If m = − n−2 [ ]
2 , Jm (r) has the following property (see 117, Chapter 1, (1.5) )
n−2
r− 2 J n−2 (r) = cn R(eir h(r)), (3.42)
2
where h satisfies
n−1
|∂rk h(r)| 6 ck (1 + r)− 2 −k . (3.43)
So, for any s > 2 and k > 0,
n−1
|∂rk ϕ(r)rn−1 h(rs) | 6 ck s− 2 . (3.44)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
q
2kn X X
= k q
Cq,m
(it2 ) m=0 q
l1 ,...lq ∈Λm
Z ∞ q
Y
itP (2k r) 1 n−2
× e ∂rlj ∂rq−m ϕ̃(r)(rs)− 2 J n−2 (rs) dr,
0 j=1
P 0 (2k r) 2
(3.47)
where Λqm +
= {l1 , . . . , lq ∈ Z : 0 6 l1 < . . . < lq 6 q, l1 + . . . lq = m}. By
(3.41), (3.46) and (3.47), we get that for any q ∈ Z+ ,
|Ik (s)| . |t|−q 2k(n−m1 q) . (3.48)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
0
3.1. Lp → Lp estimates for the dispersive semi-group 61
Interpolating (3.48) with (3.45), we get for any θ > 0, |Ik (s)| .
|t|−θ 2k(n−m1 θ) .
Case 2. s > 2. By (3.42),
Z ∞
k
Ik (s) = cn 2kn eitP (2 r) ϕ̃(r)(eirs h(rs) + e−irs h̄(rs))dr
Z0 ∞
k
kn
= cn 2 ei(tP (2 r)+rs) ϕ̃(r)h(rs)dr
0
Z ∞
k
+ cn 2kn ei(tP (2 r)−rs) ϕ̃(r)h̄(rs)dr
0
=: B1 + B2 .
We can assume, without loss of generality that t > 0 and P 0 (r) > 0. We
consider the estimate of B1 . Put P1 (r) = tP (2k r) + rs. Noticing that
P10 (r) = t2k P 0 (2k r) + s > ct2km1 , we see that (3.46) also holds if one
replaces P by P1 . In view of (3.44), analogous to Case 1, we can get that
for any θ > 0,
|B1 | . |t|−θ 2k(n−m1 θ) .
We now consider the estimate of B2 . Put P2 (r) = tP (2k r) − rs. Notice
that if s = t2k P 0 (2k r), then P20 (r) = 0. We divide Case 2 into two subcases.
Case 2a. s > 2 supr∈[1/2,2] t2k P 0 (2k r), or s < 21 inf r∈[1/2,2] t2k P 0 (2k r).
It is easy to see that |P20 (r)| > ct2km1 if r ∼ 1 and (3.46) still holds if one
substitutes P by P2 . Using (3.44), we have for any θ > 0,
|B2 | . |t|−θ 2k(n−m1 θ) .
1
Case 2b. 2 inf r∈[1/2,2] t2k P 0 (2k r) 6 s 6 2 supr∈[1/2,2] t2k P 0 (2k r). Us-
ing (3.44),
n−1 n−1 (n−1)m1
|B2 | . 2kn s− 2 . t− 2 2k(n− 2 )
. (3.49)
Making an interpolation between (3.49) and (3.45), we get that for any
0 6 θ 6 n−1
2 ,
The proof of (b) is similar to that of (a) and we omit the details. Finally,
we prove (c). Let 0 6 θ 6 min( mn2 , n−1 n
2 ). If θ < m2 , then n − m2 θ > 0. In
view of (b), we immediately have
X2
kU (t)40 u0 k∞ . |t|−θ 2k(n−m2 θ) k40 u0 k1
k=−∞
. |t|−θ k40 u0 k1 .
Assume that n−1 n n
2 > m2 and θ = m2 . From the proof of (b), we see that
k0 m2
for k0 < 0 and s ∼ t2 > 2, there holds
n−1 (n−1)m2 n
|Ik0 (s)| . t− 2 2k0 (n− 2 ) . t− m2 .
If |k − k0 | > C 1, then
|Ik (s)| . t−α 2k(n−m2 α) , ∀ α > 0.
So, choosing α large enough, we have
X X
|I60 (s)| . |Ik (s)| + |Ik (s)|
|k−k0 |6C |k−k0 |>C
− mn
X X
.t 2 + 2kn + t−α 2k(n−m2 α)
− 1 − 1
2k <t m2 2k >t m2
n
−
. t m2 ,
which implies the result, as desired. If (H4) holds and m2 = α2 , the proof
is analogous and the details are omitted.
Using the above estimates to the Klein-Gordon equation
utt + u − ∆u = f (t, x), u(0, x) = u0 (x), ut (0, x) = u1 (x),
1/2
which corresponds to the semi-group G(t) = eit(I−∆) , we have the fol-
lowing
1/2
Proposition 3.8. Let G(t) = eit(I−∆) , θ ∈ [0, 1], 2σ(θ, ·) = (n + 1 +
θ)(1/2 − 1/·), and 2/β(θ, ·) = (n − 1 + θ)(1/2 − 1/·). Then we have
kG(t)f kB s−2σ(θ,p) . |t|−2/β(θ,p) kf kBps0 ,q .
p,q
X = Lp , 0
or X = Bp,2 , 2 6 p < ∞. (3.53)
Lemma 3.4. Assume that (3.53) and (3.54) are satisfied. For any T > 0
and s ∈ R, we have
where (2/θ)0 denotes the conjugate number of 2/θ, i.e., θ/2 + 1/(2/θ)0 = 1.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Lemma 3.5. Assume that (3.53) and (3.54) are satisfied. Then
kU (t)f kL2/θ (R,X s+α/2 ) . kf kH s . (3.57)
In fact,
Z
Z
T
T
(U (t)ϕ, ψ(t))dt . kϕk2
U (−t)ψ(t)dt
. (3.59)
−T
−T
2
By Lemma 3.4,
Z
2
T
U (−t)ψ(t)dt
−T
2
Z Z T !
T
= ψ(t), U (t − τ )ψ(τ )dτ dt
−T −T
Z
T
. kψkL(2/θ)0 (I,(X ∗ )−α/2 )
U (t − τ )ψ(τ )dτ
−T
L2/θ (I,X α/2 )
Lemma 3.6. Assume that (3.53) and (3.54) are satisfied. Then for any
T > 0, I = (−T, T ),
kAf kL∞ (I,H s+α/2 ) . kf kL(2/θ)0 (I,(X ∗ )s ) . (3.62)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Rt
where I = (−T, T ) ⊂ R is arbitrary, AG := 0 G(t − τ ) · dτ . In (3.83) and
(3.84), replacing Besov spaces by corresponding Bessel potential spaces, the
results also hold.
Rt
where I ⊂ R is arbitrary, AUm := 0 Um (t−τ )·dτ . In (3.87) and (3.88), re-
placing homogeneous Besov spaces by corresponding Riesz potential spaces,
the conclusions also hold.
Proof. The proofs of Theorems 3.1–3.4 are analogous and we only prove
Theorem 3.4. By Proposition 3.3,
0
Putting Ẋ = Ḃp,2 and α = −2σ(m, p), in view of Lemma 3.5 we can get
(3.87). Now we show that (3.88) holds. By Lemmas 3.4, 3.6, 3.7 and
Remark 3.3,
kAUm f kLγ(p) (I,Ḃ s−σ(m,p) ) . kf kLγ(p)0 (I,Ḃ s+σ(m,p) ) , (3.90)
p,2 p0 ,2
Case I. p ∈ [2, r]. Taking θ ∈ [0, 1] such that 1/p = (1 − θ)/2 + θ/r, we
deduce that
1 θ 1−θ
= + , σ(m, p) = θσ(m, r) + (1 − θ)σ(m, 2). (3.93)
γ(p) γ(r) ∞
In view of the convexity Hölder inequality, (3.90) and (3.91) imply that
kAUm f kLγ(p) (I,Ḃ s−σ(m,p) ) 6 kAUm f k1−θ
L∞ (I,Ḃ )
θ
s−σ(m,2) kAUm f k γ(r)
L (I,Ḃ
s−σ(m,r)
)
p,2 2,2 r,2
Case II. p > r > 2. Take θ ∈ (0, 1) such that 1/r0 = (1 − θ)/2 + θ/p0 . It
follows that
1 θ 1−θ
0
= 0
+ , σ(m, r) = θσ(m, p) + (1 − θ)σ(m, 2). (3.95)
γ(r) γ(p) 1
Noticing the complex interpolation,
0 s+σ(m,p) 0 s+σ(m,r)
(Lγ(p) (I, Ḃp0 ,2 ), L1 (I, Ḃ2,2
s
))[θ] = Lγ(r) (I, Ḃr0 ,2 ), (3.96)
(3.90) and (3.92) imply that
0 s+σ(m,r) s−σ(m,p)
AUm : Lγ(r) (I, Ḃr0 ,2 ) → Lγ(p) (I, Ḃp,2 ) (3.97)
is a bounded operator. Theorem 3.4 follows.
Let us recall that the starting point in previous section is the estimate
(3.54), where we always assume θ ∈ (0, 1). We consider in this section the
case θ = 1. Assume that U (t) is defined in (3.52), and for any 2 < p < ∞,
there exist α(p) ∈ R and θ(p) > 0 such that
kU (t)f kH α(p) . t−θ(p) kf kp0 , ∀ 2 6 p < ∞. (3.98)
p
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Suppose that there exists a p1 > 2 satisfying θ(p1 ) > 1, then there exists a
r ∈ (2, p1 ) such that θ(r) = 1 (see (3.118)), i.e.,
kU (t)f kH α(r) . t−1 kf kr0 . (3.99)
r
This estimate was essentially obtained by Keel and Tao [121],3 where they
used the techniques of the interpolation on bilinear operators.
According to the proof of (3.57), it suffices to show
Z
T
(U (t)ϕ, ψ(t))dt . kϕk2 kψkL2 (I,B −α(r)/2 ) . (3.101)
−T r0 ,2
In order to prove (3.101), analogous to (3.59) and (3.60), one needs to show
that
Z
2
T
U (−t)ψ(t)dt
. kψk2L2 (I,B −α(r)/2 ) . (3.102)
−T
r0 ,2
2
where
D := {(s, t) : s, t ∈ [−T, T ], s 6 t}. (3.105)
Our goal is to prove that
|L(F, G)| . kF kL2 (I,B −α(r)/2 ) kGkL2 (I,B −α(r)/2 ) . (3.106)
r0 ,2 r0 ,2
3 Keel and Tao did not consider the case α(r) 6= 0, here our proof is a modification of
Lemma 3.8. Assume that (3.98) and (3.99) are satisfied, and P =
(1/r, 1/r). Let (1/a, 1/b) ∈ B(P, ε) and ε > 0 is sufficiently small. Then
we have
|Lj (F, G)| . (2j T )−β(a,b) kF kL2 (I,H −α(a)/2 ) kGkL2 (I,H −α(b)/2 ) , (3.110)
a0 b0
where
1
β(a, b) = (θ(a) + θ(b)) − 1. (3.111)
2
Proof. We can assume that F and G are Schwartz functions which have
compact support contained in [−T, T ] on the time variable. We will consider
the following three cases:
(1) a = b = p ∈ (2, ∞);
(2) 2 6 a < r, b = 2;
(3) a = 2, 2 6 b < r.
First, we consider case (1). Using Young’s and Hölder’s inequalities,
Z Z
|Lj (F, G)| . (t − s)−θ(p) kF (s)kH −α(p)/2 kG(t)kH −α(p)/2 dsdt
I t−s∼2j T p0 p0
j 1−θ(p)
. (2 T ) kF kL2(I, H −α(p)/2 ) kGkL2 (I, H −α(p)/2 ) . (3.112)
p0 p0
Z Z 2j T Z 2j T
. |s − σ|−θ(a) ×
I 0 0
kF (σ + t − 2j+1 T )kH −α(a)/2 kF (s + t − 2j+1 T )kH −α(a)/2 dσdsdt
a0 a0
j 2−θ(a)
. (2 T ) kF k2L2 (I, H −α(a)/2 ) . (3.114)
a0
Theorem 3.5. Let U (t) and A be as in (3.52) and satisfy (3.98) and
(3.99). Assume that 2 6 p, q 6 r. Then
kU (t)φkL2/θ(p) (I,Bp,2
s ) . kφkH s−α(p)/2 , (3.123)
kAf kL2/θ(p) (I,B s−α(p)/2 ) . kf kL(2/θ(q))0 (I,Ḃ s+α(q) ) . (3.124)
p,2 q0 ,2
Proof. Now we can use Lemmas 3.8 and 3.9 to show (3.100). Take p =
q = 2, η = 2/3 and η0 = η1 = 1/3 in Lemma 3.9. Choose a0 = b0 and
a1 = b1 satisfying
θ(a0 ) = 1 + , θ(a1 ) = 1 − 2, (3.125)
where > 0 is sufficiently small. Put
−α(ai )/2
Ai = Bi = L2 (I, Ha0 ), Ci = `β(a
∞
0 ,bi )
, i = 0, 1. (3.126)
i
We emphasize that in the right hand side of (3.129), the omitted constant
is independent of T .
As α(p0 ) = 0, the proof is slight different from the above discussions
and we omit the details. So, (3.100) is proved and the other cases of (3.123)
have been discussed in the previous section.
Below we prove (3.124). In fact, (3.129) implies that
kAf kL2 (I,B α(r)/2 ) . kf kL2 (I,B −α(r)/2 ) . (3.130)
r,2 r0 ,2
Using the same way as in the previous section, we have (see Lemmas 3.6,
3.7)
kAf kL∞ (I,L2 ) . kGkL2 (I,B −α(r)/2 ) , (3.131)
r0 ,2
1/2
Corollary 3.3. Let G(t) = eit(I−∆) , θ ∈ [0, 1], n > 3 − θ, 2∗∗ = 2(n −
1 + θ)/(n − 3 + θ), 2 6 r, p 6 2∗∗ , 2σ(θ, ·) = (n + 1 + θ)(1/2 − 1/·), and
2/β(θ, ·) = (n − 1 + θ)(1/2 − 1/·). Then
kG(t)φkLβ(θ,p) (I,B s−σ(θ,p) ) . kφkH s , (3.137)
p,2
Chapter 4
Let us consider the Cauchy problem for the nonlinear Schrödinger equation
(NLS)
75
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
to the standard semi-group theory, if one can solve NLS in H k , then the
uniqueness, persistence and the continuity of the solution map u0 → u can
also be obtained by a standard way. However, this method has much more
assumptions on nonlinearity and initial data. Let us consider the equivalent
integral form of (4.1):
Z t
u(t) = S(t)u0 − i S(t − τ )f (u(τ ))dτ, (4.4)
0
So, we have
kukL4x,t (R2+1 ) . ku0 k2 + k|u|2 ukL4/3(R2+1 )
x,t
So, we can solve (4.9) in the space L4x,t (R2+1 ) at least for small Cauchy
data u0 ∈ L2 . To realize (4.12), one needs to make a delicate balance
between the nonlinearity |u|2 u and the Strichartz space L4x,t . The above
performance contains a very general idea to show the well posedness of
NLS. In the next few sections we will mainly use this technique to study
NLS and nonlinear Klein-Gordon equations.
Proof. If s − δ < 1, the proof is very easy and we omit the details of the
proof. It suffices to consider the case s − δ > 1. We assume that s − δ ∈ / N.
By the equivalent norm in homogeneous Besov spaces,
Z ∞ X 1/2
−2v α 2 dt
kf (u)kḂ s−δ = t sup k M h D f (u)k L r 0 , (4.19)
r0 ,2
0 |h|6t t
|α|=[s−δ]
Hence,
Aq 6 Ckukp−1
s kukḂ v
Ḃ 0
kukḂ s 6 CkukpḂ s0 kukḂ s . (4.22)
r,2 1/a0 ,2 r,2 r,2 r,2
0
By (4.23), we have
q
Y
Aq 6 Ckukp−q
s kukḂ v
Ḃ 0
kukḂ s0 +βi
r,2 1/a00 ,2 r,2
i=1
q
Y
6 Ckukp−q
s
Ḃ 0
kukḂ s0+βi . (4.24)
r,2 r,2
i=0
(ii) Lemma 4.1 covers the nonlinear estimates for NLS and NLKG as
in [19; 26; 41; 173; 191; 236; 237]), where Cazenave and Weissler
[26] considered the case δ = 0, s0 = s; Wang [236; 237] discussed
the nonlinear estimates for NLKG.
2
(iii) If the nonlinearity has an exponential growth, say sinh u, (e|u| −
1)u, the nonlinear mapping estimates can not be covered by Lemma
4.1, one can refer to Nakamura and Ozawa [173], and Wang [239].
4.3.2 Wellposedness in H s
We consider the following equivalent integral equation with respect to
(4.29),
Z t
u(t) = S(t)u0 − i S(t − τ )f (u(τ ))dτ, (4.30)
0
where S(t) = eit∆ . According to the Strichartz inequalities, if u0 ∈ H s , then
(3.75) and (3.76) indicate that the solution should belong to Lγ(p) (I, Bp,2 s
).
Assume that
|f (k) (u)| 6 C|u|σ+1−k , k = 0, 1, ..., [s] + 1, (4.31)
where 0 6 s < n/2, [s] denotes the integer part of s, 0 < σ 6 4/(n − 2s).
We have (see [26])
Theorem 4.1. Let 0 6 s < n/2, 0 < σ < 4/(n − 2s). Assume that
f ∈ C [s]+1 satisfies (4.31), [s] 6 σ. If u0 ∈ H s , then there exists a T ∗ :=
T ∗ (ku0 kH s ) > 0 such that (4.30) has a unique solution
γ(r)
u ∈ Lloc ([0, T ∗ ); Br,2
s
), (4.32)
where
n(2 + σ)
r= . (4.33)
n + sσ
∗
Moreover, for any 2 6 p 6 2 (p 6= ∞), we have
γ(p)
u ∈ Lloc ([0, T ∗ ); Bp,2
s
), (4.34)
∗
and if T < ∞, then
kukLγ(r) ([0,T ∗ );Br,2
s ) = ∞, (4.35)
ku(t)kH s & (T ∗ − t)1−σ(n−2s)/4 , 0 < t < T ∗ . (4.36)
In Theorem 4.1, we obtain the local wellposedness in H s , where σ is
a subcritical power in H s . If σ is a critical power in H s , we can get the
global wellposedness for NLS with small data.
Theorem 4.2. Let 0 6 s < n/2 and σ = 4/(n − 2s). Assume that f ∈
C [s]+1 satisfies (4.31), [s] 6 σ. If u0 ∈ H s , then there exists a T ∗ :=
T ∗ (u0 ) > 0 such that (4.30) has a unique solution u verifying (4.32), where
r = n(2 + σ)/(n + sσ) and γ(r) = 2 + σ. Moreover, for any 2 6 p 6 2∗
(p 6= ∞), the solution satisfies (4.34). If T ∗ < ∞, then (4.35) holds.
If ku0 kḢ s is sufficiently small, then the above solution is a global one,
i.e., T ∗ = ∞ and
kuk∩26p62∗, p6=∞ Lγ(p) (0,∞;Ḃ s ) . Cku0 kḢ s , (4.37)
p,2
Proof. Now we prove Theorems 4.1 and 4.2. Let T > 0, M > 0 and
δ > 0 which will be chosen later. Put
n o
D = u ∈ Lγ(r) (0, T ; Br,2
s
) : kukLγ(r) (0,T ;Ḃ s ) 6 δ, kukLγ(r) (0,T ;Br,2
s ) 6 M ,
r,2
(4.39)
which is equipped with the metric
d(u, v) = ku − vkLγ(r) (0,T ;Lr ) . (4.40)
Considering the mapping
Z t
T : u(t) → S(t)u0 − i S(t − τ )f (u(τ ))dτ, (4.41)
0
So,
kf (u)kLγ(r)0 (0,T ;Ḃ s0 ) . T 1−σ(n−2s)/4 kukσ+1
Lγ(r) (0,T ;Ḃ s )
. (4.45)
r ,2 r,2
Similarly, we have
kT ukLγ(r) (0,T ;Br,2
s )
. T 1−σ(n−2s)/4 (kukσ + kvkσ )Lγ(r) (0,T ;Ḃ s ) ku − vkLγ(r) (0,T ;Lr ) . (4.48)
r,2
1
kT u − T vkLγ(r) (0,T ;Lr ) ku − vkLγ(r) (0,T ;Lr ) . (4.53)
2
So, T : (D, d) → (D, d) is a contraction mapping. So, there exists a u ∈ D
satisfying (4.30). Again, in view of the Strichartz inequalities (3.75) and
(3.76), we have
kukLγ(p) (0,T ;Ḃ s . ku0 kḢ s + T 1−σ(n−2s)/4 kukσ+1 , (4.54)
p,2 ) Lγ(r) (0,T ;Ḃ s )
r,2
So, analogous to the above, one can show the local wellposedness of The-
orem 4.2. For the global wellposedness with small data, it suffices to take
T = ∞ in (D, d) as in Case 1. In fact, analogous to (4.46)–(4.48), we have
kT ukLγ(r) (0,∞;Ḃ s . ku0 kḢ s + kukσ+1 , (4.60)
r,2 ) Lγ(r) (0,∞;Ḃ s )
r,2
kT ukLγ(r) (0,∞;Br,2
s )
kT u − T vkLγ(r) (0,∞;Lr ) ,
. (kukσ + kvkσ )Lγ(r) (0,∞;Ḃ s ) ku − vkLγ(r) (0,∞;Lr ) . (4.62)
r,2
Put
δ = 2Cku0 kḢ s 1, M = 2Cku0 kH s . (4.63)
Using the contraction mapping argument, we can get the result.
1 2λ
E(u(t)) = k∇u(t)k22 + ku(t)k2+α
2+α . (4.66)
2 2+α
From the conservation of energy we see that for the focusing NLS, the
kinetic energy 21 k∇uk22 and the potential energy 2+λ
2λ
kuk2+σ
2+σ have opposite
signs and for the defocusing NLS, the kinetic energy and the potential
energy are both positive. So, the defocusing case is better than the focusing
case. Roughly speaking, the defocusing NLS has global solutions and the
focusing NLS has blowup phenomena.
Theorem 4.3. Let 0 < σ < 4/n. If u0 ∈ L2 , then (4.64) has a unique
solution
γ(r)
u ∈ C([0, ∞); L2 ) ∩ ∩2<r62∗ , r6=∞ Lloc (0, ∞; Lr ) (4.67)
Theorems 4.3 and 4.4 are corollaries of Theorem 4.1 and the conserva-
tion laws in (4.65). Since we have gotten the local wellposedness in Theorem
4.1, using the standard argument, one can get that the local solution of NLS
(4.64) satisfies (4.65) at any local time. By (4.36), we have T ∗ = ∞.
For the energy critical case σ = 4/(n − 2), n > 3, Theorem 4.4 also
holds, see Chapter 7.
where m2 > 0. The main results in this section are the following
Theorem 4.5. Let n > 2, 1/2 6 s < n/2 and 0 < σ < 4/(n − 2s). Assume
that f ∈ C [s+1/2] satisfies
Theorem 4.6. Let n > 2, 1/2 6 s < n/2 and σ = 4/(n− 2s). Assume that
f ∈ C [s+1/2] satisfies (4.70), [s + 1/2] 6 σ + 1. If (u0 , u1 ) ∈ H s × H s−1 ,
then there exists a T ∗ := T ∗ (u0 , u1 ) > 0 such that (4.69) has a unique
solution u verifying (4.71). Moreover, if k(u0 , u1 )kH s ×H s−1 is sufficiently
small, then T ∗ = ∞.
Theorem 4.7. Let n > 2, 0 < σ < 4/(n − 2) and f (u) = |u|σ u. If
(u0 , u1 ) ∈ H 1 × L2 , then (4.69) has a unique solution
γ(r) s−β(r)
u ∈ C([0, ∞); H 1 ) ∩ ∩r∈(2,2∗ ) Lloc ([0, ∞); Br,2 ) , (4.73)
and
1 m2 1 2
E(u, ut ) := k∇uk22 + kuk22 + kut k22 + kuk2+σ
2+σ = E(u0 , u1 ).
2 2 2 2+σ
(4.74)
Theorems 4.5 and 4.6 can be developed to the nonlinear wave equation
(NLW), i.e., m2 = 0 in (4.69). Indeed, substituting H s , Bp,2 a
by relevant
s a
Ḣ , Ḃp,2 , we see that Theorems 4.5 and 4.6 also hold for the following NLW
where K(t) = (m2 − ∆)−1/2 sin(m2 − ∆)1/2 and K 0 (t) = cos(m2 − ∆)1/2 .
Put
s−β(ρ)
D = {u ∈ Lγ(ρ) (0, T ; Bρ,2 ) : kukLγ(ρ) (0,T ;B s−β(ρ) ) 6 M },
ρ,2
and
d(u, v) = ku − vkLγ(ρ) (0,T ;Lρ ) .
Using the Strichartz inequalities together with (4.76), we can show that
Z t
0
T : u(t) → K (t)u0 + K(t)u1 + K(t − τ )f (u(τ ))dτ (4.78)
0
is a contraction mapping from (D, d) into itself. For the details, see Wang
[236; 237].
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Chapter 5
From PDE point of view, one of the main reasons that we like Fourier
transform is that it makes the differential operation into fundamental
algebraic multiplication.
In this section, we introduce Bourgain space and its properties. The Cauchy
problem for a general dispersive equation usually has the following form:
(
∂t u − iφ(D)u = f (u, ∂ α u), (x, t) ∈ Rn × R
(5.1)
u(x, 0) = u0 (x) ∈ H s (Rn ).
91
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
∂t u − iφ(D)u = 0. (5.2)
(τ − φ(ξ))b
u(ξ, τ ) = 0. (5.3)
These modern forms of the spaces were first used to systematically study
nonlinear dispersive wave problems by Bourgain [15; 14]. Klainerman and
Machedon [146] used similar ideas in their study of the nonlinear wave
equation. The spaces appeared earlier in a dierent setting in the works
[194], [7] of Rauch, Reed, and M. Beals. Then the bilinear estimates in
these space were deeply studied by Kenig, Ponce and Vega [132], and by
Tao [216] in an abstract setting. Bourgain spaces X s,b can exploit deeply
dispersive effect of the equation, and analyze the frequency interactions of
waves evolving from the equation. Thus it is usually a powerful tool in the
well-posedness study, see [132; 218]. We take the nonlinear term uux as an
example, then Bourgain space methods reduce to the following two crucial
estimates:
Z t
ψ(t) S(t − s)f (s)ds
6Ckf kX s,b−1 , (5.5)
0 X s,b
k∂x (uv)kX s,b−1 6CkukX s,b kvkX s,b , (5.6)
where ψ(t) is a smooth cut-off function. Intuitively, we can view these two
inequalities in this way: s is the degree of regularity for the space variable,
R t for the operator ∂t − iφ(D). From (5.5) we see the nonlinear
and b is that
evolution 0 S(t − s) · ds gains one derivative with respect to ∂t − iφ(D),
and (5.6) shows this gain can compensate the loss of derivative on the
space variable. How much regularity can be obtained through this process
depends on the strength of dispersive effect of the equation. More precisely,
we will show (5.5) always hold for any φ when 1/2 < b < 1, hence the main
task is to show the second inequality which is the main topics of the next
section.
This section devotes to prove some basic properties of Bourgain X s,b .
Assume ψ ∈ C0∞ (R), suppψ ⊂ [−2, 2], and equals to 1 on [−1, 1]. Let
ψδ (·) = ψ(·/δ) for any δ > 0. We will use the following lemma, which is
proved in Chapter 3.
Proposition 5.1. Assume s ∈ R, 1/2 < b < b0 < 1, 0 < δ < 1. Then
0
kψδ (t)f kX s,b .δ 1/2−b kf kX s,b , kψδ (t)f kX s,b−1 .δ b −b kf kX s,b0 −1 .
Indeed,
kψδ (t)gkḢ c .(kψδ kL∞
t
kgkHtc + k((−∆)c/2 ψδ )gkL2t )
t
Proposition 5.2. (a) Let s ∈ R, u0 ∈ H s (Rn ), 1/2 < b < 1, 0 < δ < 1.
Then
kψδ (t)S(t)u0 kX s,b .δ 1/2−b ku0 kH s . (5.14)
s,b−1
(b) Let s ∈ R, f ∈ X and 1/2 < b < 1. Then
Z t
ψδ (t) S(t − t 0
)f (t 0
) .δ 1/2−b kf kX s,b−1 . (5.15)
0 X s,b
X ktn ψδ (t)kH b Z
g(τ )
(iτ )n b
. dτ .kgkH b−1 . (5.17)
n! |τ |61 (iτ ) t
n>1
June 29, 2011 10:5 World Scientific Book - 9in x 6in booken-main
Lemma 5.3. Assume Y is a space-time Banach space such that for all
u0 ∈ L2 (Rn ) and τ0 ∈ R
The lemma follows from the condition b > 1/2 and Cauchy-Schwartz in-
equality.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
These Besov-type X s,b space was first introduced by Tataru [221], and then
widely used by Ionescu and Kenig [113; 115], Tao [219] and the authors
[96; 90; 91; 92]. F s is a good substitute of X s,1/2 for the following two
reasons: Firstly, it is easy to see X s,1/2+ ⊂ F s ⊂ X s,1/2 , and moreover,
F s space can control many space-time norm such as Strichartz-type space
Lpt Lqx (while X s,1/2 fails logarithmicly), secondly, F s has the same scale in
time as X s,1/2 . This is very similar as the comparison between Besov space
n/2
B2,1 and Sobolev space H n/2 , H n/2+ .
First we prove the linear estimates in the space F s .
kψ(t)S(t)u0 kF s .ku0 kH s .
(b) Assume s ∈ R, k ∈ N ∪ {0} and hτ − φ(ξ)i−1 F u ∈ Xk . Then there
exists C > 0 independent of k such that
Z t
F ψ(t) S(t − s)u(s)ds
6 Ckhτ − φ(ξ)i−1 F ukX .
k
0 Xk
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Proof. To prove (a), from the definition of F s it suffices to show that for
any k ∈ N ∪ {0}
kϕk (ξ)F (ψ(t)S(t)u0 )kXk .kϕk (ξ)b
u0 (ξ)kL2 .
By the definition of Xk we get
X
kϕk (ξ)F (ψ(t)S(t)u0 )kXk = b )c
2j/2 kϕk (ξ)ϕj (τ )ψ(τ u0 (ξ)kL2
j>0
T:
Z b − τ 0 ) − ψ(τ
b − φ(ξ))
ψ(τ
T (fk )(ξ, τ ) = fk (ξ, τ 0 ) 0
hτ 0 − φ(ξ)idτ 0 .
R τ − φ(ξ)
Then it suffices to prove that
kT fk kXk 6 Ckfk kXk
hold uniformly for k > 0.
Denote fk] (ξ, τ ) = fk (ξ, τ + φ(ξ)), (T fk )] (ξ, τ ) = (T fk )(ξ, τ + φ(ξ)). By
change of variables we get
Z b − τ 0 ) − ψ(τ
b )
ψ(τ
(T fk )] (ξ, τ ) = fk] (ξ, τ 0 ) 0
hτ 0 idτ 0 .
R τ
It is easy to see that
ψ(τ
b − τ 0 ) − ψ(τ
b )
0
0
hτ i 6 C[(1 + |τ |)−4 + (1 + |τ − τ 0 |)−4 ].
τ
]
For j ∈ N∪{0}, define fk,j = fk (ξ, τ )ϕj (τ −φ(ξ)) and fk,j = fk,j (ξ, τ +φ(ξ)).
Then we have
"Z #1/2
] −4 j/2 ] 0 2 0
(T fk,j ) (ξ, τ )|.(1 + |τ |) 2 |fk,j (ξ, τ )| dτ
Ij
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
2
X Z
]
+ ϕj+l (τ ) |fk,j (ξ, τ 0 )|(1 + |τ − τ 0 |)−4 dτ 0
l=−2 Ij
:=I + II.
Hence
X 0
kT fk kXk . 2j /2
kϕj 0 (τ )(T fk )] (ξ, τ )kL2
j 0 >0
X 0
. 2j /2 kϕj 0 (τ )(T fk,j )] (ξ, τ )kL2 .
j 0 ,j>0
Lemma 5.4. Assume Y is space-time Banach space such that for all u0 ∈
L2 (Rd ) and τ0 ∈ R
keitτ0 S(t)u0 kY 6 Cku0 kL2 (Rn ) .
Then for any k ∈ N ∪ {0} and u ∈ F 0
\
k4k (u)kY .k4 k (u)kXk .
In the last Section, we already see that the elements of Bourgain space X s,b
are very close to the linear solutions u = S(t)u0 with initial data u0 ∈ H s .
Thus to use X s,b we need to get the estimates for the linear solutions. In
studying the dispersive equations with derivatives in the nonlinear terms,
there are two important kinds of estimates known as Local smoothing effect
and maximal function estimates. In this section, we introduce these two
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
100 The low regularity theory for the nonlinear dispersive equations
α
estimates using S α (t) = Fx−1 eit|ξ| ξ Fx , 1 6 α 6 2 as examples. The
results were due to Kenig-Ponce-Vega, for example see [127; 126].
Proof. Let ω(ξ) = |ξ|α ξ. Then ω is invertible and the inverse is denoted
by ω −1 . By change of variable η = ω(ξ), we have
Z Z
α ixξ itω(ξ) −1 b0 (ω −1 (η))
u
S (t)u0 = C e e û0 (ξ)dξ = C eixω (η) eitη 0 −1 dη.
R R ω (ω (η))
Using Plancherel’s equality and making change of varible η = ω(ξ) (noting
ω 0 (ξ) = (α + 1)|ξ|α ), we get
Z Z
α 2 u0 (ω −1 (η))|2
|c u0 (ξ)|2
|c
kS (t)u0 kL2 . 0 −1 (η))|2
dη. dξ.
t
R |ω (ω R |ξ|α
Therefore, (5.24) is proved.
Next we prove (5.25). The main idea is to permute the position of
x, t, namely, view t as space variable and x is time variable, then L4x L∞ t
becomes Strichartz-type estimate, and (4, ∞) is usually admissible pair.
From the proof of the first inequality we see S α (t)u0 can be viewed as the
linear solution to the dispersive equation with dispersive relation ω −1 (η),
−1
and with the initial data whose Fourier transform is uω c0 (ω (η))
0 (ω −1 (η)) . It is easy
to check that ω −1 (ξ) satisfies (H1), (H2), (H3), (H4) in Section 2.1, and
m1 = m2 = α1 = α2 = 1/(α + 1). Then we get the Strichartz estimate
−1
kFx−1 eitω (ξ)
Fx f kL4t L∞
x
.kf k 1− 1 .
Ḣ 2 4(α+1)
Thus we have
Z !1/2
1 − 1 û0 (ω −1 (η) 2
α
kS (t)u0 kL4x L∞
. |η| dη .
2 4(α+1)
t
R ω 0 (ω −1 (η))
kS α (t)u0 kL4x L∞
t
.ku0 kḢ 1/4 ,
The proof we gave here is a simplified argument for the concrete exam-
ple. In [126], Kenig, Ponce and Vega proved a more precise estimate for a
general class of dispersive equation
Z 41 Z 0 1/2 !1/2
4 2 ω (ξ)
sup |S(t)u0 (x)| dx . |c
u0 (ξ)| dξ .
R t∈R R ω 00 (ξ)
Next we give another maximal function estimate for S α (t), which was
proved in [127].
102 The low regularity theory for the nonlinear dispersive equations
If ξ ∈ supp (1 − ζ), then |h0 (ξ)| = |(α + 1)t|ξ|α + x| > |x|/3. Integrating
by parts, we get for |x| > 1
Z
eih(ξ,x) (1 − ζ)ϕ(ξ)dξ 6 C .
1 + x2
α
Proof. Define Skα (t)u0 = Fx−1 eitξ|ξ| ϕk (ξ)Fx u0 , where {ϕk }∞
k=0 is non-
homogeneous dyadic functions. It suffices to prove
X
∞ 1/2 (1+α)k
sup sup |Skα (t)u0 |2 .2 4 ku0 kL2 .
j=−∞ |t|61 j<x<j+1
(5.29)
June 29, 2011 10:7 World Scientific Book - 9in x 6in booken-main
X∞ ∞ Z
h X j+1 Z 1 2 i1/2
. Hkα (l) |g(x, t0 )|dt0 dx
l=−∞ j=−∞ j −1
∞ Z
h X 1 Z j+1 2 i1/2
(α+1)k
.2 2 |g(x, t0 )|dxdt0 ,
j=−∞ −1 j
104 The low regularity theory for the nonlinear dispersive equations
keit∂xx u0 kL2x L∞
|t|61
.ku0 kH s .
A conjecture of Kenig and Vega states that it still hold when s = 1/2.
Actually we show the inequality holds when s = 1/2 and kukH s is replaced
s
by B2,1 . For these problems we refer the readers to [126; 127].
Then the dispersion relation for KdV equation is φ(ξ) = ξ 3 . KdV equation
is the fundamental equation in shallow water theory, which was founded
in 1895 by Korteweg and de Vries. From the discussion in Section 5.1 we
know the main task is to prove the bilinear estimate. First we prove
Theorem 5.3. If −3/4 < s < −1/2, then there exists b ∈ (1/2, 1) such
that for any b0 ∈ (1/2, b] with b − b0 6 min{−s − 1/2, 1/4 + s/3} we have
We will prove Theorem 5.3 by adopting the [k; Z]-multiplier ideas. The
method is quite standard now. From the definition of X s,b , (5.33) is equal
to
0 0
khξis hτ −ξ 3 ib−1 ξ(b v )(ξ, τ )kL2ξ,τ .khξis hτ −ξ 3 ib u
u ∗b bkL2ξ,τ khξis hτ −ξ 3 ib vbkL2ξ,τ .
For the convenience, we denote
Λs,b (ξ, τ ) = hξis hτ − ξ 3 ib .
Then (5.33) is equal to
ξΛs,b−1 (ξ, τ )[(Λ−1 −1
s,b u) ∗ (Λs,b v)](ξ, τ )
.kukL2(R2 ) kvkL2 (R2 ) . (5.34)
L2ξ,τ
Assume {ϕk }∞ ∞
k=0 ({χk }k=−∞ ) is the nonhomogeneous (homogeneous)
dyadic functions, the intervals {Ik }∞ k=0 are given by (5.20). Decompose
ξi , τi − ξi3 into dyadic pieces, then by duality we get (5.34) is equivalent to
X h2k3 is 2j3 (b−1) 2k3
h2k1 is 2j1 b0 h2k2 is 2j2 b0
ki ∈Z,ji ∈Z+ ,
Z
× [1Dk3 ,j3 (uk1 ,j1 ∗ vk2 ,j2 )f ](ξ, τ )dξdτ .kukL2 kvkL2 kf kL2 , (5.35)
106 The low regularity theory for the nonlinear dispersive equations
other space-time norm, then by Kenig, Ponce, Vega [132] using elementary
calculus. Later, Tao [216] systematically studied that in an abstract setting
by using dyadic decomposition.
We will study the estimate of (5.36) in the next lemma, while for a
general class of dispersion relation φ(ξ) = |ξ|α ξ we refer the readers to [92].
Our proofs are simplified version of the [k; Z]-multiplier, quite elementary
but a little complicated. We suggest the readers to see the crucial ideas in
the calculation. For convenience, if a1 , a2 , a3 ∈ R, we use amin 6 amed 6
amax to denote the maximum, medium, minimum among a1 , a2 , a3 . We
also denote
(f1 ∗ f2 ) · f3 = f1 ∗ f2 · f3 .
where fe(ξ, µ) = f (−ξ, −µ). Note that φ(ξ) = ξ 3 is an odd function4 , Thus
feki ,ji and fki ,ji have the same support. By Cauchy-Schwartz inequality and
support properties of functions fki ,ji we get
J(fk1 ,j1 , fk2 ,j2 , fk3 ,j3 )
Z
jmin /2
.2 kfk1 ,j1 (ξ1 , ·)kL2τ kfk2 ,j2 (ξ2 , ·)kL2τ kfk3 ,j3 (ξ1 + ξ2 , ·)kL2τ dξ1 dξ2
R2
4 If not odd, there are less symmetries, but the methods still work.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
3
Y
.2kmin /2 2jmin /2 kfki ,ji kL2 ,
i=1
which is (a).
For (b), from the support properties we know J(fk1 ,j1 , fk2 ,j2 , fk3 ,j3 ) = 0
unless
kfk]3 ,j3 (ξ2 , τ1 + τ2 + Ω(ξ1 , ξ2 − ξ1 ))kL2 (|ξi |∼Ni , i=1,2) . N2−1 kfk]3 ,j3 kL2 (R2 ) .
(5.42)
It follows from (5.41), (5.42) and Hölder’s inequality that
3
Y
J(fk1 ,j1 , fk2 ,j2 , fk3 ,j3 ) . 2−k3 kfki ,ji kL2 (R2 ) .
i=1
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
108 The low regularity theory for the nonlinear dispersive equations
Now we prove Theorem 5.3 using the lemma. In order to better under-
stand how the conditions are imposed, we divide the proof into many cases.
For each case one has a condition, then intersection of all conditions would
ensure the theorem.
Thus one may assume (5.44) in (5.43). Now we prove (5.35) using (5.43).
Divide the left-hand side of (5.35) into several parts, we may assume
(k2 , k3 ) = (kmed , kmax ), and k2 , k3 > 1, then from (5.43) we get
X Z
h2k3 is 2j3 (b−1) 2k3
[1Dk3 ,j3 (uk1 ,j1 ∗ vk2 ,j2 )f ](ξ, τ )dξdτ
h2k1 is 2j1 b0 h2k2 is 2j2 b0
|k2 −k3 |65,ji ∈Z+ ,
X
. kukL2 k4k2 vkL2 k4k3 f kL2 .kukL2 kvkL2 kf kL2 ,
|k2 −k3 |65
Divide the summation into three parts, the first part is k1 6 −2k2 , denoted
by I, the second part is k1 > −2k2 and 2k1 /2 6 2j2 /2−k2 , denoted by II, the
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
110 The low regularity theory for the nonlinear dispersive equations
third part is k1 > −2k2 and 2k1 /2 > 2j2 /2−k2 , denoted by III. It is easy to
see that I.1. For II, summing on j1 , j2 , j3 we get
X 2(k1 +2k2 )(b−1)
II. .
h2k1 is
−2k2 6k1 6k2
Case 3 (high-low interaction II): k2 > 100, |k3 − k2 | 6 10, k1 > k2 − 10.
Similarly, it follows from Lemma 5.7 (c) that if s > −3/4, 1/2 < b <
3/4 + s/3,
X 2jmax (b−1) 2k3 (1−s) 2jmin /2 2jmed /4 X k(3/4−s+3(b−1))
(5.43). . 2 .1.
ki ∈Z,ji ∈Z+ ,
2jmin b0 2jmed b0 2kmax /4 k>1
Theorem 5.4. If −3/4 < s 6 0, then there exists b ∈ (1/2, 1) such that
k∂x (u1 u2 )kX s,b−1 6 Cku1 kX s,b ku2 kX s,b (5.47)
and
k∂x (u1 u2 )kX s,b−1 .ku1 kX −3/4+,b ku2 kX s,b + ku1 kX s,b ku2 kX −3/4+,b . (5.48)
Remark 5.1. The condition s > −3/4 in Theorem 5.4 is necessary. When
s 6 −3/4, (5.47) fails for any b ∈ R. The case s < −3/4 was due to
Kenig, Ponce and Vega [132], and the case s = −3/4 was due to Nakanishi,
Takaoka and Tsutsumi [180].
Now we prove the local well-posedness for KdV equation using the bi-
linear estimates. First assume u0 ∈ H s , −3/4 < s < −1/2, and define the
operator and set:
Z t
Φu0 (u) = ψ1 (t)S(t)u0 − ψ1 (t) S(t − t0 )ψT (t0 )∂x (u2 )(t0 )dt0 ,
0
112 The low regularity theory for the nonlinear dispersive equations
Next we will show u is the unique solution in XTs,b by using the ideas
in [165]. Assume u1 , u2 ∈ XTs,b are two solutions to KdV equation with
the same initial data u0 , we will prove u1 (t) = u2 (t), t ∈ [−T, T ]. From
symmetry it suffices to prove u1 (t) = u2 (t), t ∈ [0, T ]. For δ > 0 which will
be determined later, i = 1, 2, define uei
ui (t), t ∈ [0, δ],
uei = ui (2δ − t), t ∈ [δ, 2δ], (5.49)
u , otherwise.
0
Thus we get from Proposition 5.2, Proposition 5.1 and Theorem 5.3 that
0
ku1 − u2 kX s,b 6 δ b −b (kukX s,b + kvkX s,b )ke
u1 − u
e2 kX s,b .
δ+ T T
then we have
0
ku1 − u2 kX s,b 6 Cδ b −b (kukX s,b + kvkX s,b )ku1 − u2 kX s,b .
δ+ T T δ+
b0 −b
Choose δ such that Cδ (kukX s,b + kvkX s,b ) < 1/2, then we get u1 (t) =
T T
u2 (t) if t ∈ [0, δ]. Repeating this procedure, we obtained the uniqueness in
[0, T ].
If s is large, then we use the scaling invariance. It is easy to see that
KdV equation has the following scaling invariance: for any λ > 0
u(x, t) → λ2 u(λx, λ3 t), u0 (x) → λ2 u0 (λx). (5.50)
Then Ḣ −3/2 is critical in the sense that: kλ2 u0 (λ·)kḢ −3/2 = ku0 kḢ −3/2 .
From the fact
3
kλ2 u0 (λx)kH −3/4+ .λ 2 + ku0 kH −3/4+ + λ3/4+ ku0 kH −3/4+ ,
thus choosing λ sufficiently small, we may assume
kφkH −3/4+ 6 0 1. (5.51)
The rest argument is similar to the case −3/4 < s < −1/2, and we leave it
to the readers. Therefore, we prove
114 The low regularity theory for the nonlinear dispersive equations
[
By examining the supports of the functions, ϕk3 (ξ)(4 [
k1 u∗ 4k2 v) ≡ 0 unless
| max(k1 , k2 , k3 ) − med(k1 , k2 , k3 )| 6 5.
Thus we have the following several cases depending on the relative sizes
k1 , k2 , k3 : high × low → high, low × high → high, high × high → low,
high × high → high, low × low → low. For each case, we have the corre-
sponding estimtaes.
Using Lemma 5.4, Theorem 5.1 and the proof of Lemma 5.6, we have
the following proposition.
k4k (u)kL2x L∞
t∈I
.23k/4 kF [4k (u)]kXk ,
k4k (u)kL4x L∞
t
.2k/4 kF [4k (u)]kXk ,
−j
k4j (u)kL∞ 2 .2
x Lt
kF [4j (u)]kXj .
d
.k4 [
0 vkX0 k4k2 vkXk2 .
.k 2 3 −k/2 −k1
2 [
k4 d
k1 ukXk1 k4k vkXk .
116 The low regularity theory for the nonlinear dispersive equations
:= I + II + III.
:=II1 + II2 .
d
. 2−3k/4 k4 [
k ukXk k4k2 vkXk2 .
d
. 2−3k/4 k4 [
k ukXk k4k2 vkXk2 .
From the support properties we have 1Dk1 ,j (uk2 ,j1 ∗vk3 ,j2 ) = 0 unless |jmax −
jmed | 6 10 or jmax 6 1000. It follows from Lemma 5.7 (a) that
[
khτ − ξ 3 i−1 ϕk1 (ξ)ξ(4 [ [ [
k2 u ∗ 4k3 v)kXk1 .k4k2 ukXk2 k4k3 vkXk3 .
d
.(2−3k/2 + k2−2k+k1 /2 )k4 [
k ukXk k4k2 vkXk2 .
118 The low regularity theory for the nonlinear dispersive equations
0
X X
. 2−k 2−k3 /2 2k3 2−k/2 2−k3 /2 2j1 /2 2j2 /2 kuk,j1 k2 kvk,j2 k2
k3 =−10k j1 ,j2 >0
d
.k2−3k/2 k4 d
k ukXk k4k vkXk . (5.60)
.2 −3k/2 d
k4 [
k ukXk k4k2 vkXk2 .
.k2 −2k k1 /2
2 d
k4 [
k ukXk k4k2 vkXk2 ,
Thus, (5.62) follows from Propositions 5.5-5.9, the condition −3/4 < s 6 0
and discrete Young’s inequality.
With this bilinear estimates and the scaling (5.50), we can also obtain
the LWP of KdV equation in H s for s > −3/4. From the proof of Theorem
5.6 we see the condition s > −3/4 was only needed in Proposition 5.9 (a).
Thus to study the endpoint case s = −3/4, one naturally asks whether
Theorem 5.6 holds at s = −3/4, more precisely, whether the bound in
Proposition 5.9 (a) can be improved to 2−3k/2 ? The answer is negative by
the following counter-example which was given by N. Kishimoto [141] using
the one in [180].
120 The low regularity theory for the nonlinear dispersive equations
It is easy to see that if (τ, ξ) ∈ (τ (j) , ξ (j) ) − 1/2R0 (j > 1), then |ξ| ∼
2j N −1/2 |τ | ∼ 2j N 3/2 , hence (5.68) is equivalent to
X 0
N 3/2 a0 2−j /2 kϕj 0 (τ − ξ 3 )ϕ0 (ξ)ξ2−j/2 aj 1(τ (j) ,ξ(j) )−1/2R00 kL2
j 0 >0,2j N 3/2 ∼2j0
Xm m
X
∼ N 3/2 a0 (2j N 3/2 )−1/2 2j N −1/2 2−j/2 aj |R00 |1/2 ∼ a0 aj .
j=1 j=1
The proposition above shows that if s = −3/4, one can not obtain local
well-posedness by using F s or X s,b . From the discussion we see the only
problem is the low frequency structure P60 u. Fortunately, by using a new
low frequency structure, we can overcome the logarithmic divergence in
Proposition 5.9 (a). We will use the following structure:
kukX̄0 = kukL2xL∞
t
.
From Proposition 5.4 we have
kϕ0 (t)40 ukX̄0 .k40 ukX0 . (5.69)
Thus it is weaker than X0 , but on the other hand, we have for any 1 6 q 6
∞ and 2 6 r 6 ∞
k40 ukLq|t|6T Lrx ∩Lrx Lq|t|6T .T k40 ukL2x L∞
|t|6T
. (5.70)
For −3/4 6 s 6 0, define
X
F̄ s = {u ∈ S 0 (R2 ) : kuk2F̄ s = 22sk kϕk (ξ)F uk2Xk + k40 uk2X̄0 < ∞}.
k>1
Now we indicate our ideas in constructing F¯s . The starting point is the
bilinear estimates in F s :
k∂x (uv)kN s 6 CkukF s kvkF s .
This estimate fails at s = −3/4. On the other hand, we expect contraction
principle still work at s = −3/4. Thus we need a new space F̄ −3/4 , which
of course satisfies the following
F̄ −3/4 ⊂ C(R : H −3/4 ).
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
122 The low regularity theory for the nonlinear dispersive equations
nition we have
Z t
u(2) = ψ(t)S(t)u0 + Cψ(t) S(t − s)∂x (S(s)u0 · S(s)u0 )ds.
0
Z 3 3 3
3 1 − e−it(ξ −ξ1 −ξ2 ) \ \
= ψ(t)η0 (ξ)eitξ ξ 4k1 u0 (ξ1 )4 k2 v0 (ξ2 )
ξ=ξ1 +ξ2 ξ 3 − ξ13 − ξ23
:= Fx (I) + Fx (II).
Since in the hyperplane ξ = ξ1 + ξ2 one has ξ 3 − ξ13 − ξ23 = 3ξξ1 ξ2 , then we
get
Z \ \
itξ 3 4 k1 u0 (ξ1 )4k2 v0 (ξ2 )
Fx (I) = ψ(t)ϕ0 (ξ)e .
ξ=ξ1 +ξ2 3ξ1 ξ2
Applying Theorem 5.2 we get
Z
\
4 u (ξ )4\ v (ξ )
k1 0 1 k2 0 2
kIkL2x L∞ 6 C
6 C2−3k1 /2 ku0 kL2 kv0 kL2 .
t
ξ=ξ1 +ξ2 3ξ1 ξ2
L2ξ
Rt
Proof. Denote Q(u, v) = ψ(t) 0 S(t − s)40 ∂x [4k1 u(s)4k2 v(s)]ds. Di-
rect computations show that
Z b b − ξ3 )
ψ(τ − τ 0 ) − ψ(τ
F [Q(u, v)] (ξ, τ ) =c ϕ0 (ξ)iξ
R τ 0 − ξ3
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
124 The low regularity theory for the nonlinear dispersive equations
Z
× dτ 0 [
4 [
k1 u(ξ1 , τ1 )4k2 v(ξ2 , τ2 ).
ξ=ξ1 +ξ2 ,τ 0 =τ1 +τ2
[
.2−3k1 k4 [
k1 ukXk1 k4k2 ukXk2 ,
We may assume j3 6 10k1 and |τ1 − ξ13 |&3|ξξ1 ξ2 |. It follows from Lemma
5.7 (b) that
X X
kF −1 (A3 )kL2x L∞
t
. 2k3 2j2 /2 2−k1 kuk1 ,j1 kL2 kvk2 ,j2 kL2
k3 60 j1 >k3 +2k1 −10,j2 ,j3 >0
[
.k1 2−2k1 k4 [
k1 ukXk1 k4k2 ukXk2 ,
where
b(ξ1 , τ1 ),
uk1 (ξ1 , τ1 ) =ϕk1 (ξ1 )1{|τ1 −ξ13 |3·22k1 |ξ|} u
vk2 (ξ2 , τ2 ) =ϕk2 (ξ2 )1{|τ2 −ξ23 |3·22k1 |ξ|} vb(ξ2 , τ2 ).
By change of variable τ10 = τ1 − ξ13 , τ20 = τ2 − ξ23 , we get
Z t Z
−1 itξ 3 −isξ 3
Ft (A2 ) =ψ(t)e ϕ0 (ξ)iξ e eis(τ1 +τ2 )
0 R2
Z
3 3
× eisξ1 uk1 (ξ1 , τ1 + ξ13 )eisξ2 vk2 (ξ2 , τ2 + ξ23 ) dτ1 dτ2 ds.
ξ=ξ1 +ξ2
126 The low regularity theory for the nonlinear dispersive equations
Z
Z
uk1 (ξ1 , τ1 + ξ13 )vk2 (ξ2 , τ2 + ξ23 )
.
ξ
dτ1 dτ2
2
τ1 + τ2 − ξ 3 + ξ13 + ξ23
R ξ=ξ1 +ξ2 L2 ξ
3k1
.2 − 2 [
k4 [
k1 ukXk1 k4k2 ukXk2 .
Since in the integral area |τi | 22k1 |ξ|, i = 1, 2, thus in the hyperplane
X∞ n
1 1 1 τ1 + τ2
− = .
τ1 + τ2 − ξ 3 + ξ13 + ξ23 −ξ 3 + ξ13 + ξ23 n=1
3ξξ1 ξ2 3ξξ1 ξ2
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
then we get
Ft−1 (II1 − II10 )
Z Z ∞
X n
1 τ1 + τ2
= ψ(t)ϕ0 (ξ)ξ eit(τ1 +τ2 )
R2 ξ=ξ1 +ξ2 n=1 3ξξ1 ξ2 3ξξ1 ξ2
it(ξ13 +ξ23 )
× e uk1 (ξ1 , τ1 + ξ13 )vk2 (ξ2 , τ2 + ξ23 ) dτ1 dτ2 .
Decomposing dyadically on the low frequency, {χk (ξ)}∞
k=−∞ denote homo-
geneous decomposition, we get
Ft−1 (II1 − II10 )
∞ Z
X X 3 3
= eit(τ1 +τ2 ) ψ(t)eit(ξ1 +ξ2 ) χk3 (ξ)
n=1 R2
2k3 2−2k1 max(|τ1 |,|τ2 |)
Z n
τ1 + τ2 uk1 (ξ1 , τ1 + ξ13 )vk2 (ξ2 , τ2 + ξ23 )
× dτ1 dτ2 .
ξ=ξ1 +ξ2 3ξξ1 ξ2 3ξ1 ξ2
We rewrite it into the following form
Ft−1 (II1 − II10 )
∞ Z
X X 3 3
= eit(τ1 +τ2 ) ψ(t)eit(ξ1 +ξ2 ) χk3 (ξ)(ξ/2k3 )−n
n=1 R2
2k3 2−2k1 max(|τ1 |,|τ2 |)
Z n
τ1 + τ2 uk1 (ξ1 , τ1 + ξ13 )vk2 (ξ2 , τ2 + ξ23 )
× 2−nk3 dτ1 dτ2 .
ξ=ξ1 +ξ2 3ξ1 ξ2 3ξ1 ξ2
Since χk3 (ξ)(ξ/2k3 )−n is the multiplier for the L2x L∞ 0
t , then as for II1 we
get
kF −1 (II1 − II10 )kL2x L∞
t
∞ Z
X X
. C n |τ1 + τ2 |n 2−nk3
n=1 R2
2k3 2−2k1 max(|τ1 |,|τ2 |)
Z
it(ξ13 +ξ23 ) F (fτ1 )(ξ1 )F (gτ2 )(ξ2 )
×
ψ(t)e
dτ1 dτ2 .
3ξ1n+1 ξ2n+1
ξ=ξ1 +ξ2 L2 L∞ x t
128 The low regularity theory for the nonlinear dispersive equations
∞
X Z
. (C/M )n 2−3k1 /2 kF (fτ1 )kL2 kF (gτ2 )kL2 dτ1 dτ2
n=1 R2
[
.2−3k1 /2 kP [
k1 ukXk1 kPk2 ukXk2 .
Inspired by the proposition, then we may use the structure L2x L∞ t for
the low frequency part. It remains to control the high-low interaction.
In order to apply the contraction principle, the rest task is to show the
boundedness B : F̄ s × F̄ s → F̄ s .
which is sufficient to give (5.81), since in this case k4k ukL∞ 2 .k4k ukX
t Lx k
If max(k2 , k3 ) 6 10, then from Proposition 5.14 and Proposition 5.8 we get
kB(4k2 u, 4k3 v)kX̄0 .k4k2 ukL∞ 2 k4k vkL∞ L2 ,
t Lx 3 t x
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
130 The low regularity theory for the nonlinear dispersive equations
Lemma 5.8. Assume (X, k · k) is a Banach space with the norm k · k. Let
B : X × X→X be a bilinear operator satisfying
kB(x1 , x2 )k 6 ηkx1 kkx2 k, ∀ x1 , x2 ∈ X.
Then for all y ∈ X with 4ηkyk < 1, the equation x = y + B(x, x) has unique
1
solution x ∈ X such that kxk < 2η .
Using Lemma 5.8, Proposition 5.13 and Proposition 5.14, we can prove
Theorem 5.7.
5.5 I-method
In the last two sections, we studied the local well-posedness for the KdV
equation. In this section, we study the global well-posedness, namely ex-
tend the local solution to a global one. KdV equation (5.32) is completely
integrable, and hence has infinite conservation laws. With these conser-
vation laws, one can derive some a priori estimates of the solution. For
example, if u is a smooth solution to the KdV equation (5.32), then for
k ∈ N ∪ {0} and T > 0
ku(t)kH k 6 C(T, ku0 kH k ), ∀ t ∈ [−T, T ]. (5.83)
With these a priori estimates and local well-posedness, one can easily obtain
the following: KdV equation is globally well-posed in H k , where k ∈ N ∪
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
132 The low regularity theory for the nonlinear dispersive equations
4/3
Finally take C = ku0 kL2 .
From this, for general H s norm, we also need to find quantity Qs (u(t)),
they have similar properties as Q1 (u(t)). However, since such quantity does
not exist for s < 0, then we relax the condition to Qs (u(t)) increase much
slower than the norm kukH s during the evolution. The issues that how
to construct Qs (u(t)) and control the increasing rate are systematically
studied in ’I-method’.
Given m : Rk → C, m is said to be symmetric if
m(ξ1 , · · · , ξk ) = m(σ(ξ1 , · · · , ξk ))
134 The low regularity theory for the nonlinear dispersive equations
where
M5 (ξ1 , . . . , ξ5 ) = −2i[σ4 (ξ1 , ξ2 , ξ3 , ξ4 + ξ5 )(ξ4 + ξ5 )]sym .
This procedure can be made for a finite times, but we stop here since it
suffices for our purposes.
In order to control the norm kukH s , it seems to convenient to take
m(ξ) = hξis . However, it is difficult to control its increase. Basing on the
L2 conservation, we take m as following: m is a smooth even function which
has the form
1, |ξ| < N,
m(ξ) = (5.86)
N −s |ξ|s , |ξ| > 2N.
It is easy to see that if N → ∞ then Iu → u, and hence the increase rate
of the modified energy EIk (t) tends to 0 as N → ∞. The basic question is
how fast it tends to 0. If m is of the form (5.86), then it is easy to see m2
satisfies
2 2 0 0
m (ξ) ∼ m (ξ ) |ξ| ∼ |ξ |,
2
(m2 )0 (ξ) = O( m|ξ|(ξ) ), (5.87)
(m2 )00 (ξ) = O( m2 (ξ) ).
|ξ|2
and
|a(ξ + η + λ) − a(ξ + η) − a(ξ + λ) + a(ξ)|.|η||λ| sup |a00 (ξ 0 )|. (5.89)
|ξ 0 |∼|ξ|
Proposition 5.17. If m is of the form (5.86), then for any dyadic num-
bers λ 6 µ there exists an extension of σ3 from {ξ1 + ξ2 + ξ3 = 0, |ξ1 | ∼
λ, |ξ2 |, |ξ3 | ∼ µ} to the entire space {(ξ1 , ξ2 , ξ3 ) ∈ R3 , |ξ1 | ∼ λ, |ξ2 |, |ξ3 | ∼
µ} such that
|∂1β1 ∂2β2 ∂3β3 σ3 (ξ1 , ξ2 , ξ3 )| 6 Cm2 (λ)µ−2 λ−β1 µ−β2 −β3 , (5.90)
Next we give the estimate of σ4 . The proof here was due to [96]. Com-
pared to the proof given in [42], the proof here is more flexible to a general
class of dispersive equations. But the ideas are the same.
Proposition 5.18. Let m be of the form (5.86). Then on the area |ξi | ∼
Ni , |ξj + ξk | ∼ Njk , where Ni , Njk are dyadic numbers, we have
|M4 (ξ1 , ξ2 , ξ3 , ξ4 )| m2 (min(Ni , Njk ))
. . (5.91)
|h4 | (N + N1 )(N + N2 )(N + N3 )(N + N4 )
Proof. From symmetry we may assume N1 > N2 > N3 > N4 . Since ξ1 +
ξ2 +ξ3 +ξ4 = 0, then N1 ∼ N2 . We may also assume N1 ∼ N2 &N , otherwise
M4 vanishes, since m2 (ξ) = 1 if |ξ| 6 N , and if max(N12 , N13 , N14 ) N1 ,
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
136 The low regularity theory for the nonlinear dispersive equations
138 The low regularity theory for the nonlinear dispersive equations
140 The low regularity theory for the nonlinear dispersive equations
Next we will prove two properties. One is to show the modified energy
EI4 (t) increase slowly, the other is to prove it is close to EI2 (t). To control
the increase rate of EI4 (t), it suffices to control its derivative
d 4
E (t) = Λ5 (M5 ).
dt I
Proposition 5.20. Assume I ⊂ R and |I|.1. Let 0 6 k1 6 . . . 6 k5 and
k4 > 10. Then
Z Z
Y5 5
Y
5(k1 +k2 +k3 )
P ki (wi )(x, t)dxdt .2 12 2 −k4 −k5
k4\
kj (wj )kXkj ,
I
i=1 j=1
(5.103)
where on the right-hand side if kj = 0 then replace Xkj by X̄0 .
Proof. It follows from Hölder inequality that the left-hand side of (5.103)
is bounded by
3
Y
k4ki (wi )kL3x L∞
t∈I
· k4k4 (w4 )kL∞ 2 · k4k (w5 )kL∞ L2 .
x Lt 5 x t
i=1
Proof. It follows from Proposition 5.19 that the left-hand side of (5.104)
is bounded by
X Z δ N45 m2 (N∗45 )
Λ5 Q ;
ki >0 0 (N + N1 )(N + N2 )(N + N3 )(N + N45 ) 5i=1 m(Ni )
5
Y
15
Pk1 u1 , · · · , Pk5 u5 dt.N − 4 kuj kF̄ 0 (δ) ,
i=1
ki N∗45
where Ni = 2 . Cancel (N +N45 ) 6 1 and consider only the worst case
m2 (N∗45 ) = 1. it suffices to prove
P R Q
δ 3 1 1 1
k1 ,··· ,k5 >0 0 Λ5 i=1 (N +Ni )m(Ni ) m(N4 ) m(N5 ) ; Pk1 u1 , · · · , Pk5 u5 dt
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
15 Q5
.N − 4 i=1 kuj kF̄ 0 (δ) .
If N2 ∼ N1 &N , N4 .N2 , then consider the worst case N1 > N2 > N4 >
N5 > N3 . We get from (5.103) that
X 5
Y
(5.105).N − 4
15
hN1 i−5/4 hN2 i−5/4 hN3 i1/6 N4 N5
7/6 7/6 [
kPki ukXki
Ni i=1
5
Y
15
.N − 4 kIuj kF̄ 0 (δ) . (5.106)
j=1
The following proposition shows that EI4 (t) is very close to EI2 (t).
and
4
Y
|Λ4 (σ4 ; u1 , u2 , u3 , u4 )|. kIuj (t)kL2 . (5.108)
j=1
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
142 The low regularity theory for the nonlinear dispersive equations
It suffices to prove
XZ −1/2
3
Y 3
Y
N1 ηki (ξi )ubi (ξi ). kui kL2 .
ki >0 ξ1 +ξ2 +ξ3 =0,|ξi |∼Ni i=1 i=1
X 3
Y 3
Y
−1/6 1/6
. N1 N3 kvi kL3 . kui kL2 .
ki >0 i=1 i=1
For any given u0 ∈ H −3/4 and time T > 0, our purpose is to construct
a solution on [0, T ]. If u is a solution to KdV equation with initial data u0 ,
then for any λ > 0, uλ (x, t) = λ−2 u(x/λ, t/λ3 ) is also a solution to KdV
equation with initial data u0,λ = λ−2 u0 (x/λ). Simple calculations show
that
3
kIu0,λ kL2 .λ− 2 −s N −s ku0 kH s .
June 29, 2011 10:20 World Scientific Book - 9in x 6in booken-main
144 The low regularity theory for the nonlinear dispersive equations
2s
For fixed N (N will be determined later), we take λ ∼ N − 3+2s such that
3
λ− 2 −s N −s kφkH s = 0 < 1.
For simplicity of notations, we still denote uλ by u, u0,λ by u0 , and assume
kIu0 kL2 6 0 . The purpose is then to construct solutions on [0, λ3 T ]. By
Proposition 5.23 we get a solution on [0, 1], and we need to extend the
solution. It suffices to control the modified energy EI2 (t) = kIuk2L2 .
First we control EI2 (t) for t ∈ [0, 1], we will prove EI2 (t) < 420 . By
standard bootstrap, we may assume EI2 (t) < 520 . Then from Proposition
5.22 we get
EI4 (0) = EI2 (0) + O(30 )
and
EI4 (t) = EI2 (t) + O(30 ).
From Proposition 5.21 we get for all t ∈ [0, 1]
EI4 (t) 6 EI4 (0) + C50 N −15/4 .
Thus
kIu(1)k2L2 = EI4 (1) + O(30 ) 6EI4 (0) + C50 N −15/4 + O(30 )
=20 + C50 N −15/4 + O(30 ) < 420 .
Thus the solution u can be extended to t ∈ [0, 2]. Iterating this procedure
M steps, then we get for t ∈ [0, M + 1]
EI4 (t) 6 EI4 (0) + CM 50 N −15/4
as long as M N −15/4 .1. Thus we get
EI2 (M ) = EI4 (t) + O(30 ) = 20 + O(30 ) + CM 50 N −15/4 < 420 ,
hence the solution can be extended to t ∈ [0, N 15/4 ]. Take N (T ) sufficiently
large such that
N 15/4 > λ3 T ∼ N 3 T.
Therefore, the u is extended to [0, λ3 T ].
In the end of this section, we prove some properties of the global solu-
tion. By the scaling we get
sup ku(t)kH s ∼λ3/2+s sup kuλ (t)kH s 6 λ3/2+s sup kIuλ (t)kL2 ,
t∈[0,T ] t∈[0,λ3 T ] t∈[0,λ3 T ]
and hence
sup ku(t)kH s .N −s kφkH s .
t∈[0,T ]
In above several sections, we use the Bourgain’s spaces to prove some bi-
linear estimates with respect to KdV equation. However, in some cases,
for example: local well-posedness of the KdV equation in H −3/4 , it is not
enough to get the local result directly using the Bourgain’s spaces. In this
section, we consider the following Cauchy problem of the Schrödinger equa-
tion with derivative such that the reader can understand more the method
of the Bourgain’s spaces
(
iut + uxx = iλ(|u|2 u)x , (x, t) ∈ R2
(5.114)
u(x, 0) = u0 (x).
From Section 5.1 in this chapter, it follows that dispersion relation of (5.114)
φ(ξ) = −ξ 2 . Thus we can obtain the following the definition of the norm
of the Bourgain’s spaces X s,b with respect to Schrödinger equation with
derivative:
kukX s,b = khξi2 hτ + ξ 2 ib u
b(ξ, τ )kL2 .
In order to prove the well-posedness of the Cauchy problem (5.114),
from the standard method, it suffices to show
k∂x (uv̄w)kX s,b−1 .kukX s,b kvkX s,b kwkX s,b . (5.115)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
146 The low regularity theory for the nonlinear dispersive equations
But the above inequality is invalid for any s, b ∈ R, one can refer to [91].
In this section, we will use gauge transform to improve the derivative
nonlinearity, one can refer to [105; 184; 213]. The result in this section was
first given by Takaoka[213]. Via the gauge transformation which was first
used by Hayashi [102] (see also [106])
Rx
|u(y,t)|2 dy
v(x, t) = Gλ (u)(x, t) = e−iλ −∞ u(x, t), (5.116)
(5.114) is formally rewritten as the Cauchy problem
( 2
i∂t v + ∂x2 v = −iλv 2 ∂x v̄ − |λ|2 |v|4 v,
(5.117)
v(x, 0) = v0 (x),
Rx
|u0 (y)|2 dy
where v0 (x) = e−iλ −∞ u0 (x). The corresponding inverse gauge
transformation,
Rx
|v(y,t)|2 dy
u(x, t) = eiλ −∞ v(x, t) = G−λ (v).
Since the map Gλ is a bicontinuous map from H s to H s , we can show that
the global well-posedness of (5.114) in H s is equivalent to that of (5.117).
in the following, we only consider the Cauchy problem (5.117).
Notice that the nonlinear term of the original equation iλ(|u|2 u)x =
2iλ|u|2 ux + iλu2 ūx , via the gauge transformation, the derivative in the
nonlinearity |u|2 ux has been replaced by the quintic nonlinearity |v|4 v. The
Strichartz estimate can control the nonlinearity |v|4 v easy. But, at first
sight, there still exists a derivative in the nonlinearity u2 ūx in equation
(5.117). However a derivative of the complex conjugate of the solution u
can be handled while a derivative of u cannot since ω(ξ) = −ξ 2 is a even
function.
From the definition of Bourgain’s spaces, it follows that for complex
conjugate of u
s b s b
kūkX s,b = khξi hτ − ξ 2 i F ukL2ξ L2τ , kukX s,b = khξi hτ + ξ 2 i F ukL2ξ L2τ .
For the nonlinearityu2 ūx ,we can prove that following estimate holds for
s > 1/2 and b > 1/2
kuv w̄x kXs,b−1 .kukXs,b kvkXs,b kwkXs,b . (5.118)
But for |u|2 ux , the above inequality is invalid.
Now we mainly estimate the nonlinear term v 2 ∂x v̄. Similarly with the
KdV equation, we also need some space-time estimates (Strichartz esti-
mate, local smoothing effect and maximal function estimate) of Schrödinger
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
eqution, the proofs of these estimates are similar with one of the Benjamin-
Ono equation (the dispersion relation ω(ξ) = −|ξ|ξ) in Section 5.2.
2
Lemma 5.9. The group of Schrödinger equation S(t) = Fx−1 eitξ Fx sat-
isfies :
kS(t)v0 kL6x L6t .kv0 kL2 ,
1/2
kD S(t)v0 kL∞ 2 .kv0 kL2 ,
x Lt
Using Lemma 5.3 and Lemma 5.9, we can easily obtain the following
lemma:
kD−1/4 ukL4x L∞
t
.kukX 0,b .
Theorem 5.8. Assume s > 1/2, 1/2 < b < 2/3 and b0 > 1/2. Then
kv1 v2 ∂x v¯3 kXs,b−1 6 Ckv1 kXs1 ,b0 kv2 kXs2 ,b0 kv3 kXs3 ,b0 . (5.119)
148 The low regularity theory for the nonlinear dispersive equations
third largest and minimum of |ξ1 |, |ξ2 |, |ξ3 |, |ξ4 |; and define σj± = τj ± ξj2 ,
j = 1, 2, 3, 4. Let
fi fk
Fi = F −1 , i = 1, 2; Fk = F −1
, k = 3, 4.
hτ + ξ 2 ib0 hτ − ξ 2 ib0
|ξ3 |hξ4 is
K(ξ1 , ξ2 , ξ3 , ξ4 ) = .
hξ1 is hξ2 is hξ3 is
In order to obtain the boundedness of the integral in the left side of (5.120),
we split the domain of integration in several pieces as in [91].
Case 1. If max(|ξ1 |, |ξ2 |, |ξ3 |, |ξ4 |) 6 10, then
K(ξ1 , ξ2 , ξ3 , ξ4 ) 6 C.
By Lemma 5.10 and (5.121), the left side of (5.120) restricted to this domain
is bounded by
Z Q4 Y3 Y4
i=1 fi (ξi , τi )
2 b 0 2 b 0 2 b0 .kf 4 k 2 kFi k L 6. kfi kL2 .
Γ4 hτ1 + ξ1 i hτ2 + ξ2 i hτ3 − ξ3 i i=1 i=1
4
Y
.kJ −s F1 kL4x L∞
t
kJ −s F2 kL4x L∞
t
kf4 kL2x,t kΛ1/2 F3 kL∞ 2.
x Lt
kfi kL2 .
i=1
0
If |σ2+ | = max(|σ1+ |, |σ2+ |, |σ3− |, |σ4− |), then from & hσ4− i1−b hσ2+ ib
− b0
hσ4 i hσ2+ i1−b ,similarly with above, for b 6 3/4 and s > 1/4, the left
side of (5.120) restricted to this domain is bounded by
Z Q
|ξ3 ||ξ3 |2(b−1) 4i=1 fi (ξi , τi )
Q2 s 2 b0 2 b0 2 b0
Γ4 i=1 hξi i hτ1 + ξ1 i hτ3 − ξ3 i hτ4 − ξ4 i
4
Y
.kJ −s F1 kL4x L∞
t
kF4 kL4x L∞
t
kf2 kL2x,t kΛ1/2 F3 kL∞ 2.
x Lt
kfi kL2 ,
i=1
which follows from Lemma 5.10, (5.121) and Hölder inequality.
Subcase 2b. Assume |ξ2 | ∼ |ξ3 | ∼ |ξ|max or |ξ2 | ∼ |ξ4 | ∼ |ξ|max . By
symmetry, we can assume |ξ2 | ∼ |ξ3 | ∼ |ξ|max .
If |ξ2 + ξ3 | 6 1 or |ξ4 | . |ξ1 |, then hξ1 i ∼ hξ4 i. For s > 1/2, we have
K(ξ1 , ξ2 , ξ3 , ξ4 ) 6 C.
Similarly with Case 1, the left side of (5.120) restricted to this domain is
bounded by
Z Q4 Y4
i=1 fi (ξi , τi )
2 b 0 2 b 0 2 b0 . kfi kL2 .
Γ4 hτ1 + ξ1 i hτ2 + ξ2 i hτ3 − ξ3 i i=1
If |ξ2 + ξ3 | > 1 and |ξ4 | |ξ1 |, then
max(|σ1+ |, |σ2+ |, |σ3− |, |σ4− |)&|ξ1 + ξ3 ||ξ2 + ξ3 | ∼ hξ3 ihξ4 i.
By symmetry, we only consider |σ4− | = max(|σ1+ |, |σ2+ |, |σ3− |, |σ4− |). Then
for s > 1/2 and b 6 3/4
K(ξ1 , ξ2 , ξ3 , ξ4 ) |ξ3 |1/2 |ξ3 |1/2
− 1−b . s 1−b 1−b
. .
|σ4 | hξ1 i hξ3 i hξ4 i hξ1 i1/4 hξ2 i1/4
Similarly with Subcase 2a, we can obtain the result.
Subcase 2c. If |ξ1 | ∼ |ξ3 | ∼ |ξ|max or |ξ1 | ∼ |ξ4 | ∼ |ξ|max , then similarly
with Subcase 2b, we can obtain the result.
Subcase 2d. If |ξ1 | ∼ |ξ2 | ∼ |ξ|max , then max(|σ1+ |, |σ2+ |, |σ3− |, |σ4− |)
&|ξ3 |2 , similarly with Subcase 2a, we can obtain the result.
Case 3. If |ξ|min |ξ|thd ∼ |ξ|sub ∼ |ξ|max , then we have
max(|σ1+ |, |σ2+ |, |σ3− |, |σ4− |)&|ξ3 |2 .
Similarly with Subcase 2a, we can obtain the result.
Case 4. If |ξ|min ∼ |ξ|thd ∼ |ξ|sub ∼ |ξ|max , similarly with Case 1, we
can obtain the result.
This completes the proof of theorem.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
150 The low regularity theory for the nonlinear dispersive equations
Next we estimate the nonlinearity |u|4 u. From the above standard ar-
gument, it suffices to show that
5
Y
ku1 u2 u3 ū4 ū5 kX s,b−1 . kui kX s,b . (5.123)
i=1
Notice that there exists the condition s > 1/2 when we estimate the non-
linearity u2 ūx , so we do not need to obtain the sharp index of s for (5.123),
the index s > 1/2 is enough for the local well-posedness. But the author
believe that the sharp index of s for (5.123) is s > 0.
.kukL2/(3−2b) L2 .
T x
Theorem 5.9. Let s > 1/2 and 1/2 < b 6 1. Then there exists some θ > 0
such that
5
Y
kψ(t/T )u1u2 u3 ū4 ū5 kX s,b−1 .T θ kui kX s,b . (5.124)
i=1
Using Lemma 5.11, Lemma 5.10 and the Leibniz rule for fractional deriva-
tives, we have
kψ(t/T )|u|4 ukX s,b−1 .k|u|4 ukL2/(3−2b) H s .ku4 kL∞ 2 kuk 2/(3−2b) s
T Lx L H
T x T ∞
Theorem 5.10. Let λ ∈ R and s > 1/2. For u0 ∈ H s (R), then there exists
b > 1/2 and T = T (ku0 kH 1/2 ) > 0 such that the Cauchy problem (5.114)
admits a unique local solution
u ∈ C([−T, T ] : H s (R)), Gλ (u) ∈ XTs,b .
Moreover, given t ∈ (0, T ), the map u0 → u(t) is Lipschitz continuous from
H s to C(0, T ; H s ).
Using the Leibniz rule for fractional derivatives and Sovolev inequality, for
1/2 6 s < 1, we have
s iλ/2 R−∞ x (n)
|v0 (y)|2 dy
kDs un kL∞
T L x
s
2 .kD vn kL∞ L2 +
D
T x
e vn
∞ 2
LT Lx
Rx
s
s iλ/2 −∞ |v0 (y)|2 dy
(n)
.kD vn kL∞
T Lx
2 +
D e
∞ p kvn kL∞ q
T Lx
LT Lx
s
.kD vn kL∞ 2 +
T Lx
kvn k2L∞ L2p
x
1 kvn kL∞ H 1/2
T T x
where 2 < p < ∞, 1/p + 1/q = 1/2 and 1/p1 = 1/p + 1 − s. Similarly with
above, we have
2
kun − um kL∞ s .(1 + kvn kL∞ H s + kvm kL∞ H s ) kvn − vm kL∞ H s .
T Hx T x T x T x
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
152 The low regularity theory for the nonlinear dispersive equations
Except for KdV equation and Schrödinger equation, there exist some other
important dispersive equations which have been considered by lots of math-
ematicer recently. In this section, the authors give some methods for some
other dispersive equations such that the readers can understand more the
method of Bourgain’s spaces.
From the bilinear estimates of KdV equation and trilinear estimates of
Schrödinger equation with derivative, it follows that we can use not only
the method of Tao’s [k; Z] multiplier [216], but also the method of the
local smoothing effects and maximal function estimates [213] to consider
the well-posedness of dispersive equations. We take the bilinear estimates
of KdV equation for example to show the difference of the two methods.
First, we can use the method of Tao’s [k; Z] multiplier to prove Theorem
5.3. That is
k∂x (u1 u2 )kX s,b−1 6 Cku1 kX s,b0 ku2 kX s,b0 , s > −3/4, b, b0 > 1/2. (5.126)
However, (5.126) holds with the condition s > −5/8 using only Strichartz
estimate, the local smoothing effects and the maximal function estimates
of KdV equation, one can refer to [128]. If −5/8 > s > −3/4, Kenig, Ponce
and Vega [132] used the following inequalities to obtain (5.126).
Z ∞
dx C 1
2b (1 + |x − β|)2b
6 2b
, b> (5.127)
−∞ (1 + |x − α|) (1 + |α − β|) 2
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Z ∞
dx C 1
√ 6 1 , b > (5.128)
−∞ (1 + |x|)2b a − x (1 + |a|) 2 2
Z ∞
dx C 1
6 , b>
−∞ (1 + |x − α|)2(1−b) (1 + |x − β|)2b (1 + |α − β|)2(1−b) 2
(5.129)
Z 1
dx C(1 + β)2(b− 2 ) 1
√ 6 1 , b> . (5.130)
|x|6β (1 + |x|)2(1−b) a−x (1 + |a|) 2 2
As in Section 5.3 of this chapter, we first define ξ = ξ1 + ξ2 , τ = τ1 + τ2
σ = τ − ξ 3 , σ1 = τ1 − ξ13 , σ2 = τ2 − ξ23 . We split the the hyperplane
{(ξ, ξ1 , ξ2 ) × (τ, τ1 , τ2 ) ∈ R3 × R3 : ξ = ξ1 + ξ2 , τ = τ1 + τ2 } into suitable
several parts. For the case: |ξ1 | ∼ |ξ2 | |ξ| and |σ| > |σ1 |, |σ2 |, we only use
the inequality (5.127); for other cases, similarly with the paper [128], we
use the Strichartz estimates, the local smoothing effects and the maximal
function estimates, thus we can also obtain (5.126). This can simplify the
proof of the paper [132], the readers can try it. From this, it follows that the
interaction: high × high → low is worst when one consider low regularity
solution in Sobolev spaces of negative indice. Moreover, corresponding
to the inequality (5.127), there exists the following inequality, which is
also important when one consider the local well-posedness of the dispersive
equations
Z
dx C 1 1
2b 2b
6 4b−1
, <b< . (5.131)
R hx − αi hx − βi hα − βi 4 2
Next we take the fourth-order nonlinear Schrödinger equation for exam-
ple, and show how to consider the well-posedness for the dispersive equation
with complicated phase function (the dispersion relations of KdV equation
φ(ξ) = ξ 3 and Schrödinger equation φ(ξ) = |ξ|2 do not have non-zero sin-
gular points, that is, the solutions of the phase function φ(ξ) = 0, φ0 (ξ) = 0
and φ00 (ξ) = 0 are only zero; the complicated phase functions are generally
assumed to have non-zero singular points). The Cauchy problem for the
fourth-order Schrödinger equation which describes the motion of the vortex
filament as follows:
i∂t u + ∂x2 u + ν∂x4 u = F (u, ū, ∂x u, ∂x ū, ∂x2 u, ∂x2 ū), (x, t) ∈ R × R, (5.132)
where ν is a non-zero real constant, the nonlinear term F is given by
F (u, ū, ∂x u, ∂x ū, ∂x2 u, ∂x2 ū)
1
= − |u|2 u + λ1 |u|4 u + λ2 (∂x u)2 ū + λ3 |∂x u|2 u + λ4 u2 ∂x2 ū + λ5 |u|2 ∂x2 u
2
June 29, 2011 10:22 World Scientific Book - 9in x 6in booken-main
154 The low regularity theory for the nonlinear dispersive equations
with λ1 = − 3µ ν
4 , λ2 = −2µ+ 2 , λ3 = −4µ−ν, λ5 = −2µ+ν, where µ is also a
real constant. The equation (5.132) describes the three-dimensional motion
of an isolated vortex filament, which is embedded in inviscid incompressible
fluid fulfilled in an infinite region. Note that the dispersion relation of the
equation (5.132) φ(ξ) = νξ 4 − ξ 2 has non-zero singular points. we can not
directly use the local smoothing effects and the maximal function estimates
of Section 5.2 in this chapter to consider it. So we need use the following
Fourier restriction operators
Z Z
PNf = eixξ fˆ(ξ)dξ, PN f = eixξ fˆ(ξ)dξ, ∀N > 0, (5.133)
|ξ|>N |ξ|6N
to eliminate the singularity of the phase function. Using the Fourier re-
striction operators and the results of Section 5.2, we have
3
kDx2 P 2a S(t)ϕkL∞ 2 6 CkϕkL2 ,
x Lt
(5.134)
− 41
kDx P a S(t)ϕkL4x L∞ t
6 CkϕkL2 , (5.135)
Z ∞ 1/2
sup |P a S(t)u0 |2 dx 6 CT,s ku0 kH s , s > 1, (5.136)
−∞ [−T,T ]
for the definition of some weaker well-posedness (for example: the solution
mapping is only continuous), some local results of the real BO equation
were obtained. For now, the best local result was obtained by Kenig and
Ionescu[113], who obtained local well-posedness in L2 . Their method based
on the gauge transform and the Besov-type Bourgain’s spaces X s,b . If
k = 3, the best result of the well-posedness was obtained by Kenig and
Takaoka [136], who used the dyadic gauge transform to prove that the
Cauchy problem is globally well-posed in H 1/2 . Moreover,they showed
that the index s = 1/2 is sharp in the sense: the solution map u0 → u
as mapping from H s to C([−T, T ]; H s ) is no longer uniformly continuous
when s < 1/2. Recently, in the paper [91], the author followed the idea in
[113], used the contraction mapping to obtain the global well-posedness in
H 1/2 with small data in L2 norm, here the author did not use the gauge
transform. So this method can apply to complex BO equation.
KdV equation and BO equation can be viewed as the special cases of
the following dispersion generalized BO equation:
∂t u + |∂x |1+α ∂x u = µ∂x (uk ), u(x, 0) = u0 (x), (5.138)
2
where 0 6 α 6 1, u : R →R and |∂x | is the Fourier multiplier operator
with symbol |ξ|. This model is very interesting from mathematical view, it
let us understand the relation between the dispersion effect and the well-
posedness. From the intensity of dispersion effect, it lies between KdV
equation and BO equation, for example: if 0 < α < 1, then it is similar
with BO equation, H s assumption on the initial data is insufficient for the
local well-posedness via Picard iteration by showing the solution mapping
fails to be C 2 smooth from H s to C([−T, T ]; H s ) at the origin for any s,
one can refer to [167]. For the methods to obtained the well-posedness for
(5.138), the method of refined energy was used in [127], the method of short
time X s,b in [115] was used in [92], and a para-differential renormalization
technique was used in [108].
Notice that the spatial dimension of the dispersive equations considered
above in this chapter is one. How about the application of the Bourgain’s
spaces to dispersive equations in higher dimension? For example: the Bour-
gain’s spaces can be applied to the higher dimensional Schrödinger equation.
Compared with one dimension case, cases in higher dimension are more del-
icate. For instance, a quadratic non-linear Schrödinger equation(n = 1, 2)
iut + ∆u = N (u, ū), u(x, 0) = u0 (x),
where (x, t) ∈ R × R, N (u, ū) = c1 |u|2 + c2 u2 + c3 ū2 . Using only the
n
156 The low regularity theory for the nonlinear dispersive equations
Chapter 6
Frequency-uniform decomposition
techniques
W. Orlicz had a small apartment and he once applied to the city ad-
ministration for a bigger one. The answer of an employee was: “Your
apartment is really small but we cannot accept your claim since we
know that you have your own spaces!” —— Lech Maligranda2
In this chapter we study the NLS and the NLKG by using the frequency-
uniform decomposition techniques, see [243; 245; 246; 247].
Let Qk be the unit cube with the center at k, {Qk }k∈Zn constitutes a
decomposition of Rn . Such a kind of decomposition goes back to the work
of N. Wiener [252], and we say that it is the Wiener decomposition of Rn .
We can roughly write
k ∼ F −1 χQk F , k ∈ Zn , (6.1)
where χE denotes the characteristic function on the set E. Since Qk is a
translation of Q0 , k (k ∈ Zn ) have the same localized structures in the
frequency space, which are said to be the frequency-uniform decomposition
operators. Similar to Besov spaces, one can use {k }k∈Zn and `q (Lp ) to
generate a class of function spaces, so called modulation spaces for which
the norm is defined by
!1/q
X
sq q
kf kMp,q
s = hki k k f kp .
k∈Zn
157
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
0
(1) k eit∆ : Lp → Lp satisfies a uniform truncated decay;
(2) k eit∆ is uniformly bounded in Lp .
It is known that S(t) = eit∆ : Lp → Lp if and only if p = 2. This is one
of the main reasons that we can not solve NLS in Lp (p 6= 2). However, if
we consider the frequency-uniform decomposition, there holds
kk S(t)f kp . (1 + |t|)n|1/2−1/p| kk f kp . (6.2)
n
Taking the summation to the above inequality over all k ∈ Z , we can get
the relevant estimate in modulation spaces, which enable us to solve NLS
0
in modulation spaces Mp,1 , 1 6 p 6 ∞.
Now we give a comparison between frequency-uniform and dyadic de-
compositions. Recalling that
4k ∼ F −1 χ{|ξ|∼2k } F
and noticing that |{ξ : |ξ| ∼ 2k }| = O(2nk ) and |Qk | = 1, we see that their
Bernstein’s estimates are quite different:
k4k f kq . 2n(1/p−1/q)k k4k f kp , kk f kq . kk f kp , p 6 q.
We find that there is no regularity increasement for the Bernstein estimate
of k , which leads to the Schrödinger semi-group S(t) = eit∆ satisfies the
following truncated decay estimate,
kk S(t)f kp . (1 + |t|)−n(1/2−1/p) kk f kp0 , p > 2, 1/p + 1/p0 = 1. (6.3)
0
Comparing it with the classical Lp → Lp estimate
k4k S(t)f kp . |t|−n(1/2−1/p) k4k f kp0 , (6.4)
we see that the singularity at t = 0 disappears in (6.3).
Recalling that in the classical Strichartz inequalities, in order to handle
the singularity of |t|−n(1/2−1/p) at t = 0, we need condition n(1/2−1/p) 6 1,
which arises from the Hardy-Littlewood-Sobolev inequality and it is an
essential condition. To solve NLS with the nonlinearity |u|σ u, one need to
improve the spatial regularity at least at the Ḣ n/2−2/σ level for σ > 4/n.
Considering the Strichartz inequalities with the frequency-uniform de-
composition, due to the truncated decay, we can remove the condition
n(1/2 − 1/p) 6 1. As a result, in solving NLS with the nonlinearity |u|σ u,
it is not necessary to improve the spatial derivative regularity if σ is large.
The frequency-uniform decomposition is more delicate than the dyadic
decomposition and it is easier to handle the derivatives in the nonlinearity,
say for k = (k1 , ..., kn ) with k1 1 and j 1,
k∂x1 k f kp ∼ k1 kk f kp , k∂x1 4j f kp . 2j k4j f kp .
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
P n
We see that ρk (ξ) = 1 in Qk and so, k∈Zn ρk (ξ) > 1 for all ξ ∈ R .
Denote
!−1
X
σk (ξ) = ρk (ξ) ρk (ξ) , k ∈ Zn . (6.7)
k∈Zn
Then we have
|σk (ξ)| > c, ∀ ξ ∈ Qk ,
supp σk ⊂ {ξ : |ξ − k|∞ 6 1},
P (6.8)
n σk (ξ) ≡ 1, ∀ ξ ∈ Rn ,
k∈Z
|D σk (ξ)| 6 C|α| , ∀ ξ ∈ Rn , α ∈ (N ∪ {0})n .
α
we obtain that for q = ∞ and 0 < p < 1, the result holds. For general
0 < p 6 q < ∞, by Hölder’s inequality, Lq norm can be controlled by Lp
and L∞ norms, from which we get the result, as desired.
kF −1 ϕF f kr 6 CkϕkH s kf kr (6.13)
we have
X
kσk f kp 6 kϕ σ
k+` (k f )kp .
|`|∞ 61
{σ }
k k {ϕ }
So, kf kMp,q
s 6 kf kMp,q
s .
Proposition 6.2 indicates that one can choose {σk }k∈Zn ∈ Υn according
to our requirement. In applications of PDE, it is convenient to us the
following {σk }k∈Zn ∈ Υn . Let {ηk }k∈Z ∈ Υ1 , we denote
σk (ξ) := ηk1 (ξ1 )...ηkn (ξn ), (6.15)
then we have {σk }k∈Zn ∈ Υn . the above σk (ξ) realizes the separation of
different variables.
From the definitions, we see that Besov spaces and modulation spaces are
rather similar, both of them are the combinations of frequency decomposi-
tion operators and function spaces `q (Lp ). In fact, we have the following
inclusion results.
1
q
p=2
S1
S3
( 12 , 12 )
S2
(0, 0) (1, 0) 1
p
1
q
R1
1
q
=
1
p
(0, 1)
1
p
+
1
q
=
1
R2
( 12 , 12 )
R3
(0, 0) (1, 0) 1
p
For k ∈ Λj , one has that |k| ∼ 2j and Λj overlaps at most O(2nj ) cubes,
which follows that
q
X∞ X
kf kqB 0 . kk f kp
p,q
j=0 k∈Λj
∞
X X
. 2jn(q−1) kk f kqp
j=0 k∈Λj
∞
X X
. hkin(q−1) kk f kqp
j=0 k∈Λj
. kf kq n(1−1/q) . (6.26)
Mp,q
∞
X
. kf kq2 + 2jn(1−q/2) kχ|·|∈(2j−2 ,2j+1 ) fbkq2
j=L
. kf kq n(1/q−1/2) , (6.31)
B2,q
n(1/p−1/2)
holds for all f ∈ LpΩ and M ∈ B2,p .
n(1/p−1/2)
holds for all f ∈ LpB(0,b) and M ∈ B2,p , where C > 0 is independent
of b > 0.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
(F −1 M F f )(x) = [F −1 M (b ·)f\
(b−1 ·)](bx),
Proof. First, we consider the case 0 < p < 1. By Corollary 6.2, for any
|k| 1, |k| ∈ [2j−1 , 2j ),
4
X
−1
kk f kp =
F σk ϕj+` F f
`=−4 p
4
X
. kσk (2j+5 ·)kB n(1/p−1/2) k4j+` f kp . (6.40)
2,p
`=−4
kg(λ ·)kBp,q
s 6 λs−n/p kgkBp,q
s , λ & 1,
we obtain that6
Secondly, we consider the case (1/p, 1/q) ∈ S1 . If (1/p, 1/q) ∈ Ṡ1 (Ṡ1
denotes the set of all inner points of S1 ), then (1/p, 1/q) can be lying in
the segment by connecting (1/∞, 1/∞) and a point (1/p1 , 1/q1 ) in the line
{(1/p, 1/q) : p = 2, q < 2}. By a complex interpolation, we have
n(1/q−1/p)
Bp,q ⊂ Mp,q . (6.44)
Thirdly, if (1/p, 1/q) ∈ S2 , the dual version of the first inclusion implies
the result, as desired.
(Necessity) We need to show that for any 0 < η 1,
τ (p,q)−η
Bp,q 6⊂ Mp,q . (6.45)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
. 2q(n/q−η)j . (6.46)
Assume without loss of generality that
5 j−1 3 j+1
ϕj (ξ) = 1, if ξ ∈ Dj := ξ : ·2 6 |ξ| 6 · 2 .
4 4
Noticing that
√
Λj = k ∈ Zn : B(k, n) ⊂ Dj
contains at least O(2jn ) elements, we have
X X
kf kqMp,q = kk F −1 ϕj kqp > kF −1 σk ϕj kqp & 2nj . (6.47)
k∈Zn k∈Λj
∞ X
X
6 2−εN qj kF −1 σk(j)+` fb(· − k(j))kqp
j=1 |`|∞ 61
∞ X
X
= 2−εN qj k` f kqp . 1, (6.52)
j=1 |`|∞ 61
where ϕj = ϕ(2−j ·). We can assume that ϕ(ξ) = 1 if |ξ| ∈ [3/4, 5/4].
Hence,
F −1 ϕN j χQk(j) fb(· − k(j)) = F −1 χQk(j) fb(· − k(j)). (6.54)
By (6.53) and (6.54),
∞
X
kF krBp,r
s & 2(s−ε)rN j = ∞. (6.55)
j=1
s
The above discussion also implies that F 6∈ B∞,∞ .
Case 1.3. (1/p, 1/q) ∈ S1 . We can assume that σk (ξ) = 0 if ξ ∈ 6 Q̃k :=
{ξ : |ξi − ki | 6 5/8, 1 6 i 6 n}, and the dyadic decomposition function
sequence satisfying ϕj (ξ) = 1 if ξ ∈ Dj (= {ξ : 45 · 2j−1 6 |ξ| 6 43 · 2j+1 }).
Let
Aj = {k ∈ Zn : Q̃k ⊂ Dj }, j 1. (6.56)
It is easy to see that Aj has at most O(2 ) elements. Let f ∈ S (Rn ) be
nj
1/q
X
= kF −1 σ0 (τd
k f )kp
q
& 2jn/q . (6.58)
k∈Aj
For any N 1,
|f (x)| 6 CN (1 + |x|)−N , (6.71)
it follows that
|fa (x)| 6 CN aN |x|−N . (6.72)
By the continuity of f (x) and f (0) = 1, we deduce that there exists a % > 0
such that7
|fa (x)| > 1/2, x ∈ B(0, %). (6.73)
7 In fact, % can be chosen as % = a%0 , %0 > 0 is independent of a.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
that f (x) is a decreasing function on |x|. Since Aj,` has at most O(an 2`n )
elements, weX have for any x ∈ B(k(i),
X X %),
|fa (x − k)| 6 |fa (x − k)|
k∈Aj \{k(i)} `>1 k∈Aj,`
X
6C an 2n` |fa (2` − %)|
`>1
X
. CN an+N 2(n−N )` 6 1/4, (6.75)
`>1
where N > n + 1 and CCN an+N 6 1/4. Hence, it follows from (6.74) and
(6.75) that
|g(x)| > 1/4, x ∈ B(k(i), %). (6.76)
By (6.76), we have
n nj
O(aX 2 )
1
kg(x)kp >
χ
B(k(i),%)
& (a%)
n/p nj/p
2 , (6.77)
i=1 4
p
where % and a are independent of j 1. We can assume that ϕj (ξ) = 1 for
ξ ∈ Dj . Since supp ĝ ⊂ Dj , we have F −1 ϕj F g = g. Hence, (6.77) implies
that
kgkBp,q
0 > kF −1 ϕj F gkp & (a%)n/p 2nj/p . (6.78)
On the other hand,
1/q
X
kgk n(1/p−1/q) =
Mp,q
2nj(q/p−1) kF −1 σk F gkqp
|k|∈[2j−1 ,2j+1 ]
nj/p
62 sup kF −1 σk F gkp . (6.79)
|k|∈[2j−1 ,2j+1 ]
Since supp σk overlaps at most finite many supp τ` (τd
` fa ), using multiplier
estimates, we have
kF −1 σk F gkp . kf kp . (6.80)
Hence, (6.79) and (6.80) imply that
kgkM n(1/p−1/q) . 2nj/p . (6.81)
p,q
By (6.78) and (6.81) we immediately have (6.66). We finish the proof of
Theorem 6.1.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
2
Hence, it suffices to estimate kF −1 (σk eit|ξ| )k1 . Using the multiplier esti-
mate, we have
2 2
kF −1 (σk eit|ξ| )k1 = kF −1 (σ0 eit|ξ| )k1
1−n/2L
X 2 n/2L
. kσ0 k2 kDα (σ0 eit|ξ| )k2
|α|=L
n/2
. (1 + |t| ). (6.83)
Noticing that
kS(t)f kMp,q
s . (1 + |t|)n|1/2−1/p| kf kMp,q
s . (6.86)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Propositions 6.6 and 6.7 are optimal in the sense that the powers of time
variable are sharp, cf. [51]. Now we consider the truncated decay estimate
1/2
for the Klein-Gordon semi-group G(t) = eitω where ω = I − ∆.
X
kk G(t)f kp 6 kF −1 σk+` eithξi σk fˆkp
|`|∞ 61
X
6 kF −1 (σk+` eithξi )k1 kk f kp . (6.91)
|`|∞ 61
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
So, it suffices to estimate kF −1 (σk eithξi )k1 . Using the Fourier multiplier
estimate, we have
kF −1 (σk eithξi )k1 = kF −1 (σ0 eithξ+ki )k1
1−n/2L
X n/2L
. kσ0 k2 kDα (σ0 eithξ+ki )k2
|α|=L
n/2
. (1 + |t| ). (6.92)
Noticing that
kk G(t)f k2 = kk f k2 , (6.93)
by a complex interpolation, we immediately have
kk G(t)f kp . (1 + |t|)n|1/2−1/p| kk f kp . (6.94)
where
1 1
2 6 p < ∞, 2σ(p) = (n + 2) − . (6.97)
2 p
From (6.96) it follows that
k k G(t)f kH −2σ(p) . |t|−n(1/2−1/p) k k f kp0 . (6.98)
p
X
. kF k+` f kp
`∈Λ
X
. k k+` f kp0 . (6.101)
`∈Λ
So, for any θ ∈ [0, 1], it follows from (6.100) and (6.101) that
X
k k G(t)f kp . hki2σ(p)θ |t|−nθ(1/2−1/p) k k+` f kp0 . (6.102)
`∈Λ
holds for all f ∈S (R ) and ψ ∈ C0∞ (R, S (Rn )). Noticing that S (Rn ) and
n
−α/2,q0 0 0
C0∞ (R, S (Rn )) are dense in M2,q and ` (Lγ (R, Lp )), respectively,
(6.111) implies (6.109). By duality,
Z
Z
(U (t)f, ψ(t))dt . kf kM2,q
U (−t)ψ(t)dt
. (6.112)
R R M2,q0
For any k ∈ Zn ,
Z
2
k U (−t)ψ(t)dt
R
2 Z
. k k ψkLγ 0 (R,Lp0 )
k U (t − s)ψ(s)ds
. (6.113)
R Lγ (R,Lp )
Recall that {k }k∈Zn are almost orthogonal. By (6.107), the definition of
k · kMp0 ,q together with the multiplier estimate, we have
X
k k U (t)f kp . hti−δ hki−α k k k+` f kMp0 ,q
|`|∞ 61
−δ −α
. hti hki k k f kp0 . (6.114)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Proposition 6.11. Let U (t) satisfy (6.107) and (6.108). For any γ >
2 ∨ (2/δ), we have
kU f k`q (L∞ (R,L2 )) . kf k`−α/2,q (Lγ 0 (R,Lp0 )) . (6.120)
In addition, if γ 0 6 q, then
kU f kL∞ (R,M2,q ) . kf kLγ 0 (R,M −α/2 ) . (6.121)
p0 ,q
Proof. We only sketch the proof. Applying the same way as in (6.113),
(6.115) and (6.116), we see that
kk U f k22 . hki−α k k f k2Lγ 0 (R,Lp0 ) . (6.122)
(6.122) implies (6.120). By Minkowski’ s inequality, it follows from (6.120)
that (6.121) holds.
8 Due to ht − si−δ < |t − s|−δ , if δ < 1, γ = 2/δ, one can apply Hardy-Littlewood-
Sobolev’s inequality.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Proposition 6.12. Let U (t) satisfy (6.107) and (6.108). For any γ >
2 ∨ (2/δ), we have
kU f k`α/2,q (Lγ (R,Lp )) . kf k`−α/2,q (Lγ 0 (R,Lp0 )) . (6.123)
Rt
Let S(t) = eit∆ , A = 0 S(t − s) · ds. Then
kS(t)ϕk`1 (Lγ (R,Lp )) . kϕkM2,1 , (6.130)
kA f k`1 (Lγ (R,Lp ))∩`1 (L∞ (R,L2 )) . kf k`1 (Lγ 0 (R,Lp0 )) . (6.131)
kG f k`−σ,q (Lγ (R,Lp ))∩`q (L∞ (R,L2 )) . kf k`σ,q (Lγ 0 (R,Lp0 )) . (6.134)
The proof of Theorem 6.2 relies upon the algebra structure of Mp,1 .
The proof of Lemma 6.7 can be shown by following the same technique as
s
that of E2,1 and we omit the details.
We write
Since
k i uk∞ . k i uk2 , i ∈ Zn , (6.154)
in view of (6.153) and (6.154), we have
π(u1+κ )
1 p0 . kuk1+κ
` (Lx,t∈R ) X . (6.155)
(6.160)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Hence,
∞
X C 2k+1
kT ukY . ku0 kM2,1 + kuk2k+1
Y . (6.161)
k!
k=1
Applying (6.161) and combining the proof of Theorem 6.3, we can get the
conclusion of Theorem 6.4.
We now consider the initial value problem for NLKG,
utt + (I − ∆)u + f (u) = 0, u(0) = u0 , ut (0) = u1 . (6.162)
Analogous to NLS, we have
Theorem 6.6. Let n > 1, f (u) = λu1+κ , κ ∈ N and κ > 4/n. Put
n+2
σ= . (6.163)
n(2 + κ)
σ σ−1
Assume that (u0 , u1 ) ∈ M2,1 × M2,1 and there exists a sufficiently small
δ > 0 such that ku0 kM2,1 + ku1 kM σ−1 6 δ. Then (6.162) has a unique
σ
2,1
solution
σ
u ∈ C(R, M2,1 ) ∩ `1 (L2+κ
x,t∈R ). (6.164)
σ
u ∈ C(R, M2,1 ) ∩ `1 (L4x,t∈R ). (6.165)
We study the initial value problem for the derivative nonlinear Schrödinger
equation (gDNLS)
iut + ∆± u = F (u, ū, ∇u, ∇ū), u(0, x) = u0 (x), (6.166)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
For the sake of brevity, we simplify the nonlinearity as F (u, ū, ∇u, ∇ū) =
~λ·∇(|u|2 u). According to the equivalent integral form of gDNLS, one needs
to solve
u(t) = S(t)u0 − iA ~λ · ∇(|u|2 u).
On the basis of the smooth effect of the Schrödinger semi-group in 1D, we
easily get its smooth effect estimates in all dimensions,
1/2
Dxi S(t)u0
∞ 2 2
. ku0 k2 , (6.169)
1+n Lx L(x Lt (R )
i j )j6=i
where
kf kLpx1 Lp2 p
Lt 2 =
kf kLpx2 ,...,x p
L 2 (R×Rn−1 )
p
. (6.171)
i (xj )j6=i 1 i−1 ,xi+1 ,...,xn t Lx1i (R)
(6.169) and (6.170) are scaling-invariant and so, they are optimal estimates.
In view of (6.170), one should choose L∞ 2 2
xi L(xj )j6=i Lt as a framework to
handle the partial derivative ∂xi in the nonlinearity. According to the
integral equation, we have
n
X
k∂x1 ukL∞ 2
x L(x L2 . kDx1/2
1
u0 k2 + k∂xi (|u|2 u)kL1x L2(x L2t . (6.172)
1 j )j6=1 t 1 j )j6=1
i=1
So, one needs to make the following two kinds of nonlinear estimates,
I = k∂x1 (|u|2 u)kL1x L2(x L2t (R1+n ) , II = k∂x2 (|u|2 u)kL1x L2(x L2t .
1 j )j6=1 1 j )j6=1
kukL2x L∞
(x
L∞
t
6 kS(t)u0 kL2x L∞ L∞
t
+ k∇A (|u|2 u)kL2x L∞ L∞
t
.
1 j )j6=1 1 (x j )j6=1 1 (x j )j6=1
1/2
kk S(t)u0 kL2x L∞
(x
L∞
t
. hk1 i kk u0 k2 . (6.174)
1 j )j6=1
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Noticing that the smooth effect estimate in the lower frequency part is
worse than that of the Strichartz estimate, we have thrown away the lower
frequency on the ξ1 orientation in (6.175). If we do not consider the sharp
estimate, (6.174) implies that
X
kuk`1 (L2x L∞ L ∞ ) . kuk
t M
1/2 + hk1 i1/2 kk ∇(|u|2 u)kL1t L2x .
(x ) 1 j j6=1 2,1
k∈Zn
After making nonlinear estimates, we find that the right hand side of (6.179)
can be estimated by (6.175), (6.176) and (6.178). The estimate of II is more
complicated and one needs to consider the interaction between the partial
derivative ∂x2 and the space `1,2 ∞ 2 2
hi (Lx1 L(xj )j6=1 Lt ), see below for details.
We now state the global wellposedness and scattering results for gDNLS.
Let us recall the anisotropic Lebesgue space Lpx1i Lp(x2j )j6=i Lpt 2 is defined by
(6.171). For k = (k1 , ..., kn ), we write
n
X X
kukXαs = hki is−1/2
∂xα` k u
L∞ L2 L2t
xi (xj )j6=i
i, `=1 k∈Zn , |ki |>4
n
X X
+
∂xα k u
m ∞ , (6.180)
` L L xi (xj )j6=i
L∞
t
i, `=1 k∈Zn
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
n X
X
kukSαs = hkis−1
∂xα` k u
L∞ L2 T L3 L6 , (6.181)
t x t x
`=1 k∈Zn
X X
kukX s = kukXαs , kukS s = kukSαs . (6.182)
α=0,1 α=0,1
5/2
Theorem 6.8. Let n > 3, m = 2, u0 ∈ M2,1 and there exists a suitably
small δ > 0 such that ku0 kM 5/2 6 δ. Then (6.166) has a unique solution
2,1
5/2
u ∈ C(R, M2,1 ) ∩ X 5/2 ∩ S 5/2 , kukX 5/2 ∩S 5/2 . δ. Moreover, the scattering
5/2
operator S of (6.166) carries a whole zero neighborhood in C(R, M2,1 ) into
5/2
C(R, M2,1 ).
k∂xi A f kL∞
x L
2 L2t . kf kL1x L2(x L2t , (6.184)
i (xj )j6=i i j )j6=i
Making the change of the variable τ → µ + |ξ| ¯ 2 , the right hand side of
±
(6.189) becomes
−1 ξ1 −it|ξ̄|2±
F (e f )
F
ξ1 ξ 2 − µ t,x1 Fx2 ,...,xn
2 ∞ 2. (6.190)
1 Lξ̄ Lx1 Lµ
Lemma 6.9. Let Dxσi = (−∂x2i )σ/2 . For any σ ∈ R, k = (k1 , ..., kn ) ∈ Zn
with |ki | > 4, we have
kk Dxσi ukLpx1 Lpx2 ,...,xn Lpt 2 . hki iσ kk ukLpx1 Lpx2 ,...,xn Lpt 2 .
1 2 1 2
If σ ∈ N, replacing Dxσi by ∂xσi , the above estimate holds for all k ∈ Zn .
Proof. Using (6.194), we have
X1 Z
σ
k Dxi u = Fξ−1
i
(ηki +` (ξi )|ξi | σ
) (yi )(k u)(xi − yi )dyi .
`=−1 R
On the other hand, integrating by parts, we can get that for |x1 | > 4|t|hk1 i,
2
|Fξ−1
1
eitξ1 ηk1 (ξ1 )| . |x1 |−2 .
So, for any |x1 | > 1,
2
¯ . (1 + |x1 |)−2 + hk1 in/2 (hk1 i + |x1 |)−n/2 .
|F −1 eit|ξ|± ηk1 (ξ1 )ηk̄ (ξ)|
This implies that
2
¯ q/2 ∞ n
kF −1 eit|ξ|± ηk1 (ξ1 )ηk̄ (ξ)k L L (R ) x1 x̄,t
n/2 −n/2
. 1 + hk1 i k(hk1 i + |x1 |) kLq/2 (R)
x1
2/q
. hk1 i .
It follows that (6.195) holds.
By Proposition 6.14, we have
Proposition 6.16 (Frequency-localized smooth effect). For any
k = (k1 , ..., kn ) ∈ Zn , we have
kk A ∂xi f kL∞ L2 L2t . kk f kL1x L2(x L2t , (6.196)
xi (xj )j6=i i j )j6=i
kk A ∂xi f kL∞ L2 . hki i1/2 kk f kL1x L2(x L2t . (6.197)
t x i j )j6=i
. kk f kL1x L2 L2 ,
i (xj )j6=i t
kk A ∂xi f kLγ Lrx . hki i1/2 kk f kL1x L2(x L2t , (6.200)
t i j )j6=i
1/2
kk A ∂xi f kL∞ L2 2 . hki i kk f kLγ 0 Lr0 , (6.201)
xi (xj )j6=i Lt t x
k A ∂ α f
2 ∞ . hki iα+1/2
kk f kL1t L2x . (6.202)
xi L L L∞
xi t
(xj )j6=i
Corollary 6.5. Let 2 6 q < ∞, q > 4/n and 4/n 6 p < ∞. We have the
following results.
1/2
Dx1 k S(t)u0
2
. kk u0 kL2 (Rn ) , (6.205)
L∞ 2
x1 Lx2 ,...,xn Lt
1/2
kk A f kL∞ L2 ∩ L2+p . hk1 i kk f kL1x L2x2 ,...,xn L2t , (6.209)
t x t,x 1
Moreover, if L2+p 3 6
x,t is replaced by Lt Lx in (6.207), (6.209) and (6.212), then
corresponding results also hold.
By (6.208) in Corollary 6.5, the operator k ∂x1 A : L1x1 L2x2 ,...,xn L2t →
L∞ 2 2
x1 Lx2 ,...,xn Lt
and the partial derivative ∂x1 is successfully absorbed. How-
ever, A in the space L∞ 2 2
x1 Lx2 ,...,xn Lt can not handle the partial derivative
∂x2 . So, one needs to use another way to deal with ∂x2 A when it appears
in the space L∞ 2 2
x1 Lx2 ,...,xn Lt .
kk ∂xi A f kLqx L∞ ∞ . hki ihk1 i1/q kk f kL1t L2x . (6.215)
1 x2 ,...,xn Lt
For σ > 1,
X
hk1 iσ
Fξ−1
1 ,ξ2
ψ2 Fx1 ,x2 k ∂x2 A f
2
L∞ 2
x1 Lx2 ,...,xn Lt
k∈Zn , |k1 |>4
X
. hk2 iσ kk f kL1x L2x2 ,...,xn L2t . (6.219)
1
k∈Zn , |k2 |>4
Notice that suppψ2 ⊂ {ξ : |ξ2 | > 2|ξ1 |}. If |k1 | > 4, then |k2 | > 6 and in
the summation of the left hand side of (6.219), one has that |k2 | > |k1 |. So,
X X
hk1 iσ II 6 hk2 iσ−1 hk1 iII.
k∈Zn , |k1 |>4 k∈Zn , |k1 |>4
We have
Y
−1 ξ2 ξ2
F 1 − ψ η (ξ )
ξ1 ,ξ2 2ξ1 ξ1 i=1,2
ki +`i i
1 2
L (R )
X
Y
ξ2 ξ 2
α −1
ηki +`i (ξi )
.
D Fξ1 ,ξ2 1 − ψ 2ξ1 ξ1 i=1,2
|α|62
L2 (R2 )
−1
. hk2 ihk1 i . (6.227)
Combining (6.226) and (6.227), we can get the estimate of II.
So,
kS(t)u0 kX . ku0 kM 5/2 .
2,1
(i)
In order to estimate %1 (A ∂xαj (v1 ...vκ+|ν| )), i, j = 1, ..., n, it suffices to con-
(1) (1)
sider the estimates of %1 (A ∂xα1 (v1 ...vκ+|ν| )) and %1 (A ∂xα2 (v1 ...vκ+|ν| )).
Applying the frequency-uniform decomposition, we have
X
k (v1 ...vκ+|ν| ) = k k(1) v1 ...k(κ+|ν|) vκ+|ν|
(i)
S1
X
+ k k(1) v1 ...k(κ+|ν|) vκ+|ν| , (6.228)
(i)
S2
where
(i) (1) (κ+|ν|)
S1 := {(k (1) , ..., k (κ+|ν|) ) : |ki | ∨ ... ∨ |ki | > 4},
(i) (1) (κ+|ν|) (1) (κ+|ν|)
S2 := {(k , ..., k ): |ki | ∨ ... ∨ |ki | 6 4}.
We will frequently use the almost orthogonality of k ,
k k(1) v1 ...k(κ+|ν|) vκ+|ν| = 0,
if |k − k (1) − ... − k (κ+|ν|) |∞ > κ + |ν| + 1. (6.229)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
X X
+ hk1 i5/2 kk k(1) v1 ...k(κ+|ν|) vκ+|ν| k κ+|ν|+1
κ+|ν|
k∈Zn , |k1 |>4 (1) Lt,x
S2
=: I + II. (6.230)
kk(i) vi kLκ+|ν|−1 L∞ L∞
x1 x̄ t
m m
κ+|ν|−1
1−κ+|ν|−1
6 kk(i) vi k Lm ∞ ∞
x1 Lx̄ Lt
kk(i) vi kL∞
x,t
m m
1−
. kk(i) vi kLκ+|ν|−1
x x̄ t
κ+|ν|−1
m L∞ L∞ kk(i) vi k ∞ 2
L L . (6.232)
1 t x
(1)
Now we estimate %1 (A ∂xα2 (v1 ...vκ+|ν| )). The case α = 0 has been dis-
cussed in the above, it suffices to consider the case α = 1. Let ψi (i = 1, 2)
be as in Lemma 6.10 and Pi = F −1 ψi F . We have
(1)
%1 (A ∂x2 (v1 ...vκ+|ν| ))
X
6 hk1 i2 kP1 k (A ∂x2 (v1 ...vκ+|ν| ))kL∞ 2 2
x Lx̄ Lt 1
k∈Zn , |k1 |>4
X
+ hk1 i2 kP2 k (A ∂x2 (v1 ...vκ+|ν| ))kL∞ 2 2
x Lx̄ Lt 1
k∈Zn , |k1 |>4
(6.238)
(1) (1) (κ+|ν|)
By the symmetry, we can assume that |k1 | = max(|k1 |, ..., |k1 |) in
(1)
S1 . So,
III1
κ+|ν|
X (1)
Y
. C κ+|ν| hk1 i2 kk(1) v1 kL∞ 2 2
x Lx̄ Lt
kk(i) vi kLκ+|ν|−1 L∞ L∞
1 x1 x̄ t
(1)
S1 ,
(1)
|k1 |>4 i=2
κ+|ν|
(1)
Y (1) (1)
. C κ+|ν| %1 (v1 ) (%2 (vi ) + %3 (vi )) . (CkukX )κ+|ν| . (6.239)
i=2
κ+|ν|
Y (1)
. C κ+|ν| %3 (vi ) 6 (CkukX )κ+|ν| . (6.240)
i=1
We have shown that
III . (CkukX )κ+|ν| . (6.241)
Now we estimate IV . Using (6.228), we have
X X
IV 6 hk1 i2 kP2 k (A ∂x2 (k(1) v1 ...k(κ+|ν|) vκ+|ν| ))kL∞ 2 2
x Lx̄ Lt 1
k∈Zn , |k1 |>4 (2)
S1
X X
+ hk1 i2 kP2 k (A ∂x2 (k(1) v1 ...k(κ+|ν|) vκ+|ν| ))kL∞ 2 2
x Lx̄ Lt 1
k∈Zn , |k1 |>4 (2)
S2
(κ+|ν|)
We can choose some k (i) , say k (κ+|ν|) such that |k2 | does not attain
(i)
or does not uniquely attain max16i6κ+|ν| |k2 |, then we can take another
(1) (i)
k (i) , say k (1) such that k2 = max16i6κ+|ν| |k2 |. By Hölder’s inequality,
kk(1) v1 ...k(κ+|ν|) vκ+|ν| kL1x L2x̄ L2t
1
κ+|ν|
Y
6 kk(1) v1 kL∞ 2 2
x Lx̄ Lt
kk(i) vi k(L∞ 2 2
t Lx )∩(Lx L∞
x̄,t )
. (6.244)
1 1
i=2
X κ+|ν|
. C κ+|ν| kk(1) v1 ...k(κ+|ν|) vκ+|ν| kL(2+m)/(1+m) . kukX .
t,x
(2)
S2
(6.246)
Hence, in view of (6.245) and (6.246), we have
IV . (CkukX )κ+|ν| . (6.247)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
(i)
Now we estimate %3 (A (uκ (∇u)ν )). By (6.212),
n
X X
(i)
%3 (A (uκ (∇u)ν )) . hki3/2 kk (uκ (∇u)ν )k 2+m . (6.249)
1+m
Lt,x
i=1 k∈Zn
Using Lemma 6.8, one can control the right hand side of (6.249),
X
hki3/2 kk (v1 ...vκ+|ν| )k 2+m
1+m
Lt,x
k∈Zn
m+1
! κ+|ν|
!
Y X Y X
κ+|ν| 3/2 3/2
.C hki kk vi kL2+m hki kk vi kL∞
t,x
t,x
i=1 k∈Zn i=m+2 k∈Zn
m+1
! κ+|ν|
!
Y X Y X
. C κ+|ν| hki3/2 kk vi kL2+m hki3/2 kk vi kL∞ 2
t Lx
t,x
i=1 k∈Zn i=m+2 k∈Zn
κ+|ν| n
!
Y X (1)
. C κ+|ν| %3 (vi ) 6 (CkukX )κ+|ν| . (6.250)
i=1 i=1
(1)
Now we estimate %2 (A ∂xα1 (uκ (∇u)ν )).
(1)
X
%2 (A ∂xα1 (v1 ...vκ+|ν| )) . hki3/2 kk v1 ...vκ+|ν| kL1t L2x . (6.251)
k∈Zn
× kk A ∂x1 v1 ...vκ+|ν| kL∞ 2 3 6
t Lx ∩ Lt Lx
We can use the contraction mapping argument to finish the proof and omit
the details.
Remark 6.2.
(1) There are some recent works which have been devoted to the study
of the derivative NLKG [54; 55; 76; 144; 185; 200]:
utt + u − ∆u = F (u, ut , uxx , uxt ),
where the methods are quite different from those of derivative NLS.
We do not know whether the frequncy-uniform decomposition tech-
niques are applicable to the derivative NLKG.
10 Notice that |cβ | 6 C |β| .
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Chapter 7
The energy conservation law means that it does not vary with time, in
other word, it has a certain symmetry with respect to the time. This
fundamental connection between conservation laws and symmetries was
first discovered in 1915 (published in 1918) by Emmy Nöther1 and was
called the “Nöther’s theorem”. In short, Nöther’s theorem states informally
that any differentiable symmetry of the action of a physical system has
a corresponding conservation law. Nöther’s theorem is important, both
because of the insight it gives into conservation laws, and also as a practical
calculational tool. It allows us to determine the conserved quantities from
1 Amalie Emmy Nöther (1882-1935) was a German-born mathematician known for her
205
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
where we have expanded the Lagrangian L to the first order in the per-
turbation ε(t). Applying the integration by parts to the last term results
in
t Z t1
∂L 1 ∂L d ∂L
δI = ε · + ε· −ε· dt.
∂ q̇ t0 t0 ∂q dt ∂ q̇
By the boundary condition (7.1), the first term vanishes, and then
Z t1
∂L d ∂L
δI = ε· − dt = 0.
t0 ∂q dt ∂ q̇
Noticing that we do not make any assumptions to the generalized coordi-
nates q except that all of the generalized coordinates are independent of
each other. Thus, we can apply the fundamental theorem of variational
calculus to obtain an Euler-Lagrange equation
d ∂L ∂L
− = 0, (7.2)
dt ∂ q̇ ∂q
which is a system involving n second-order equations with respect to q(t).
The Legendre transformation 3 of L is
H(q(t), p(t), t) = max[p(t) · q̇ − L(q(t), q̇(t), t)].
q̇
Let M (p, q̇) = p(t) · q̇ − L(q(t), q̇(t), t). In order to reach the maximum
of M with respect to q̇, we have to require that the first-order partial
derivative of M (p, q̇) with respect to q̇ is zero and that its second partial
derivative is less than zero (this can be guaranteed by the convexity of L,
2
i.e. ∂∂ q̇L2 > 0). Thus, if p(t) = ∂L∂ q̇ (q(t), q̇(t), t), then M (p, q̇) reaches the
maximum, i.e.
p(t) = ∂L (q(t), q̇(t), t),
∂ q̇
H(q(t), p(t), t) = p(t)q̇ − L(q(t), q̇(t), t),
Substituting (7.6) into the above equation and taking the complex conju-
gate, we can obtain the nonlinear Schrödinger equation (7.5).
Now, we give another statement for Nöther’s theorem:
turns into
Z Z n
X ∂(δξ)j
Ĩ[ũ] = L(ũ, ∂˜ũ)dξ˜ = L(ũ, ∂˜ũ) 1 + + o(ε) dξ
D̃ D j=0
∂ξj
Z Z n
X ∂(δξ)j
= L(ũ, ∂˜ũ)dξ + L(u, ∂u) dξ + o(ε), (7.12)
D D j=0
∂ξj
Denote
n
X
˜ − uk (ξ) =
δũk ≡ ũk (ξ) ∂j uk (δξ)j + δuk (ξ). (7.15)
j=0
Since
n
X n
X
˜ =
∂j ũk (ξ) ˜ ∂ ξ̃l =
∂˜l ũk (ξ) (δjl +
∂(δξ)l ˜ ˜
)∂l ũk (ξ)
∂ξj ∂ξj
l=0 l=0
n
!
X ∂(δξ)l ˜
= ∂˜j + ∂l ˜
ũk (ξ), (7.16)
∂ξj
l=0
we have
∂˜j ũk (ξ)
˜ − ∂j uk (ξ) =(∂˜j − ∂j )ũk (ξ)
˜ + ∂j (ũk (ξ)
˜ − uk (ξ))
n n
!
X ∂(δξ)l X
=− ∂˜l ũk (ξ)
˜ + ∂j ∂l uk (δξ)l + δuk .
∂ξj
l=0 l=0
(7.17)
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
Hence,
2
X Xn
∂L
L(ũ, ∂˜ũ) − L(u, ∂u) = ∂j uk (δξ)j + δuk
∂uk j=0
k=1
2 X
n n n
!!
X ∂L X ∂(δξ)l X
+ − ∂˜l ũk (ξ)
˜ + ∂j ∂l uk (δξ)l + δuk .
∂(∂j uk ) ∂ξj
k=1 j=0 l=0 l=0
(7.18)
Noticed that
2 n
!
∂ ∂(δξ)j X ∂L X ∂L
(L(δξ)j ) = L + ∂j uk + ∂ 2 uk (δξ)j ,
∂ξj ∂ξj ∂uk ∂(∂l uk ) lj
k=1 l=0
(7.19)
and
∂L ∂ ∂L ∂ ∂L
∂j δuk = δuk − δuk , (7.20)
∂(∂j uk ) ∂ξj ∂(∂j uk ) ∂ξj ∂(∂j uk )
it is easily to obtain
L(ũ, ∂˜ũ) − L(u, ∂u)
Xn Xn 2
∂ ∂(δξ)j X ∂L
= (L(δξ)j ) − L + δuk
j=0
∂ξj j=0
∂ξj ∂uk
k=1
2 X
X n
∂ ∂L ∂ ∂L
+ δuk − δuk
j=0
∂ξj ∂(∂j uk ) ∂ξj ∂(∂j uk )
k=1
2 X
X n n
∂L X ∂(δξ)l ˜ ˜ − ∂l uk ).
− (∂l ũk (ξ) (7.21)
∂(∂j uk ) ∂ξj
k=1 j=0 l=0
Therefore, it yields
Z X 2 n
X
∂L ∂ ∂L δuk dξ
δI = −
D k=1 ∂uk j=0 ∂ξj ∂(∂j uk )
Z X n 2
!
∂ X ∂L
+ L(δξ)j + δuk dξ + o(ε). (7.22)
D j=0 ∂ξj ∂(∂j uk )
k=1
(7.8) implies
X ∂ n
∂L ∂L
− = 0, k = 1, 2. (7.23)
∂uk j=0 ∂ξj ∂(∂j uk )
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
˜ − uk (ξ) − ∇ξ uk · δξ.
where δuk (ξ) = ũk (ξ)
Integrating with respect to the spatial variables yields the following result.
Theorem 7.3. If the action functional (7.11) is invariant under the in-
finitesimal transformation
t 7→ t̃ = t + δt(t, x, u), (7.25)
x 7→ x̃ = x + δx(t, x, u), (7.26)
u(t, x) 7→ ũ(t̃, x̃) = u(t, x) + δu(t, x), (7.27)
then
Z
∂L ∂L
(ut δt + ∇u · δx − δu) + (ūt δt + ∇ū · δx − δū) − Lδt dx
Rn ∂ut ∂ ūt
is a conserved quantity. In particular, in the case of nonlinear Schrödinger
∂L
equations, since ∂u t
= 2i ū, ∂∂L i
u¯t = − 2 u, it implies that
Z
i i
ū(ut δt + ∇u · δx − δu) − u(ūt δt + ∇ū · δx − δū) − Lδt dx = const.
Rn 2 2
(7.28)
to obtain, for any time t in the time interval where the solution exists, the
following pseudo-conformal conservation law
Z
|xu + 2it∇u|2 + 4t2 F (|u|2 ) dx
Rn
Z Z t Z
2
= |xu(0, x)| dx + 4 τ (n + 2)F (|u|2 ) − nF 0 (|u|2 )|u|2 dxdτ.
Rn 0 Rn
(7.38)
In the previous subsection, it involves |u|2 , x|u|2 , |x|2 |u|2 , · · · , in the con-
servation laws. Have they some kind of rules to follow? What will happen
if we replace the coefficient in front of |u|2 by a general function? Can we
obtain some other conservation laws or estimates?
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
For convenience, denote G(|u|2 ) = F 0 (|u|2 )|u|2 − F (|u|2 ). By (7.5) and the
integration by parts, we have
Z
Ma (t) = ∂t Va (t) = at |u|2 + 2Im(ū∇a · ∇u) dx, (7.39)
n
ZR
∂t Ma (t) = ∂tt Va (t) = (att − ∆2 a)|u|2 dx
Rn
Z
+4 Im(ū∇at · ∇u)dx
Rn
Z n
X
+4 ajk Re(uk ūj )dx
Rn j, k=1
Z
+2 ∆aG(|u|2 )dx. (7.40)
Rn
On the other hand, in the time dependent case a(t, x) 6= a(x), we get, for
any positive function Q = Q(t, x), that
4Im(ū∇at · ∇u) = |Q−1 (∇at )u − 2iQ∇u|2 − Q−2 |∇at |2 |u|2 − 4Q2 |∇u|2 .
Thus,
Z
∂tt Va (t) = (att − ∆2 a − Q−2 |∇at |2 )|u|2 dx
Rn
Z
+ |Q−1 (∇at )u − 2iQ∇u|2dx
Rn
5 Cathleen Synge Morawetz (1923- ) is a mathematician born in Toronto, Canada.
Morawetz’ research was mainly in the study of the partial differential equations gov-
erning fluid flow, particularly those of mixed type occurring in transonic flow. She is a
professor emerita at the Courant Institute of Mathematical Sciences at the New York
University, where she had also served as the director from 1984 to 1988.
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
Z n
X
+4 (ajk − Q2 δjk )Re(uk ūj )dx
Rn j, k=1
Z
+2 ∆aG(|u|2 )dx. (7.42)
Rn
Example 7.1 (Virial identity). Let a(t, x) = |x|2 , then ajk = 2δjk and
∆2 a = 0, thus from (7.41), it yields the virial identity
Z Z Z
∂tt |x|2 |u|2 dx = 8n |∇u|2 dx + 4n G(|u|2 )dx.
Rn Rn Rn
2
For the defocusing case F (|u|2 ) = p+1 |u|p+1 where p > 1, i.e. G(|u|2 ) =
p−1 p+1
R
p+1 |u| , it implies that the convexity of Rn |x|2 |u|2 dx can be controlled
2
by the energy. For the focusing case F (|u|2 ) = − p+1 |u|p+1 with p > 1 + n4 ,
R
i.e. G(|u|2 ) = − p−1 p+1 |u|
p+1
, the quantity ∂tt Rn |x|2 |u|2 dx can be bounded
from
R above by a multiple of the energy, thus if the energy is negative and
2
Rn |x| |u|2 dx is finite, the solution of the equation necessarily has a blowup
in finite time.
Z
tx
+ |− u − 2iQ∇u|2 dx
Rn Qλ3
Z
t2 /0 u|2 |x|2
|∇
+4 ( 3 − Q2 )|∇u|2 + dx
Rn λ λ3
Z
+2 ∆aG(|u|2 )dx.
Rn
2 t2
Taking Q := λ3 , we can eliminate some terms to reduce the above identity
as
Z Z Z
|xu + 2it∇u|2 /0 u|2 |x|2
|∇
∂t Ma (t) = − ∆2 a|u|2 dx + dx + 4 dx
Rn Rn λ3 Rn λ3
Z
n−1 t2
+2 ( + 3 )G(|u|2 )dx.
Rn λ λ
Noticing ∆2 a = O(λ−3 ), we have, for G(|u|2 ) > 0, that
Z Z
|xu + 2it∇u|2 /0 u|2 |x|2
|∇
dx + 4 dx
Rn λ3 Rn λ3
Z
t2 G(|u|2 )
+2 dx
Rn λ3
Z
. ∂t Ma (t) + O(λ−3 )|u|2 dx.
Rn
Integrating over (1, T ] with respect to t (we can deal with [−T, −1) in a
similar way), we obtain
Z TZ Z TZ
|xu + 2it∇u|2 |∇/0 u|2 |x|2
dxdt + 4 dxdt
1 Rn λ3 1 Rn λ3
Z TZ
t2 G(|u|2 )
+2 dxdt
1 Rn λ3
Z TZ
. Ma (T ) − Ma (1) + O(λ−3 )|u|2 dxdt
1 Rn
Z T Z
. C(E(0), N (0)) + O(λ−3 )|u|2 dxdt,
1 Rn
where H(t) and N (t) are the energy and the L2 -norm defined before, re-
spectively. For |t| > 1, we have
Z Z Z Z
|u|2 |u|2
3
dxdt . 3
dxdt . N (0).
|t|>1 Rn λ |t|>1 Rn |t|
Taking T → ∞, we get
Z Z Z Z
|xu + 2it∇u|2 /0 u|2 |x|2
|∇
3
dxdt + 4 dxdt
|t|>1 Rn λ |t|>1 Rn λ3
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
Z Z
t2 G(|u|2 )
+2 dxdt
|t|>1 Rn λ3
.C(E(0), N (0)).
2
In particular, let F (|u|2 ) = p+1 |u|p+1 . Assume that p ∈ (1, ∞) if n =
4 p−1
1, 2, and p ∈ (1, 1 + n−2 ) if n > 3, then F 0 (|u|2 )|u|2 − F (|u|2 ) = p+1 |u|p+1 .
From the previous estimates, it yields
Z Z
t2 |u|p+1
3
dxdt . C(E(0), N (0)). (7.45)
|t|>1 Rn |(t, x)|
For the case |t| 6 1, we split it into two cases. In the case |x| > 1, we
have
ZZ ZZ
t2 |u|p+1
dxdt . |u|p+1 dxdt . C(E(0), N (0)). (7.46)
|t|61 |(t, x)|3 |t|61
|x|>1 |x|>1
is famous for many things including: Hölder’s inequality, the Jordan-Hölder theorem,
Hölder’s theorem and the Hölder condition which is used e.g. in the theory of partial
differential equations and function spaces.
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
∗ −q
Z 1 ∗
2q2∗
−ε 2q2
∗ −q +n−1
6C r dr kukL2∗ (|x|61))
0
6CkukH 1 .
For the case n = 2, from Hölder’s inequality, we get
k|x|−ε ukLq (|x|61) 6k|x|−ε kL2q (|x|61) kukL2q (|x|61))
Z 1 1/2q
6C r−2qε+n−1 dr kukL2q (|x|61))
0
6CkukH 1 .
From (7.45), (7.45) and (7.45), we obtain the Nakanishi-Morawetz in-
equality (cf. [176])
Z
t2 |u|p+1
3
dxdt . C(E(0), N (0)), ∀n ∈ N. (7.49)
R1+n |(t, x)|
The Morawetz estimates are time-decaying estimates, which are very useful
for us to study the existence of the scattering operators. In this section,
we only state some results on the existence of the scattering operators for
NLS
iut + ∆u = |u|p u, u(0, x) = u0 (x). (7.54)
The existence of the scattering operators in the H 1 -subcritical but L2 -
supercritical cases was shown by Nakanishi [176] in 1 and 2 spatial dimen-
sions and by Ginibre and Velo [79] in higher spatial dimensions n > 3:
Theorem 7.5. Let p ∈ (4/n, 4/(n − 2)) for n > 3, and p ∈ (4/n, ∞) for
n = 1, 2. For any initial datum u0 ∈ H 1 with E(u0 ) < ∞, (7.54) has a
p(n+2)/2
unique global solution u ∈ Ct0 (Ḣx1 ) ∩ Lt,x satisfying
Z ∞Z
|u(t, x)|p(n+2)/2 dxdt 6 C(E(u0 )),
−∞ Rn
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
where
Z
1 2
E(u(t)) = |∇u(t, x)|2 + |u(t, x)|2+p dx
Rn 2 2+p
is the energy conserved quantity. Moreover, there exist finite-energy solu-
tions u± of the linear Schrödinger equation (i∂t + ∆)u± = 0 such that
ku± (t) − u(t)kḢ 1 → 0, as t → ±∞,
x
Next, we introduce briefly the related results for the scattering theory
of H 1 -critical defocusing NLS
4
iut + ∆u = |u| n−2 u, u(0, x) = u0 (x). (7.55)
In 1999, J. Bourgain [19] proved the global wellposedness of the solution
with radial initial data in 3 and 4 dimensions. Then M. Grillakis [84]
gave a different proof for Bourgain’s results in 2000. In 2005, these re-
sults were generalized to any spherically symmetric initial data in any
dimensions n > 3 by T. Tao [217]. At the same time (the paper was
published in 2008), J. Colliander, M. Keel, G. Staffilani, H. Takaoka, and
T. Tao [44] obtained the global well-posedness and the scattering theory
for any general initial data in three dimensions based on the induction
arguments on the energy of Bourgain’s. They applied the induction ar-
gument not only on the frequency space but also on the physical space,
and replaced the original Morawetz’ inequality (7.43) by Morawetz’ in-
teraction inequality (7.53) in order to deal with the non-radial cases. In
addition, the Morawetz’ interaction inequality, together with the almost
conserved quantity of the momentum which controlled the mass in the fre-
quency space, described the possibilities of the energy concentration. For
the higher dimensional case, it is generalized by E. Ryckman and M. Visan
[196] for n = 4 and by M. Visan [232] for n > 5. Thus, the scattering in
the energy class is now known for (7.55) with any general initial data in
all dimensions n > 3, namely, as stated in the following results [44; 196;
232].
Theorem 7.6. Let n > 3. For any initial datum u0 ∈ H 1 with E(u0 ) < ∞,
2(n+2)
then (7.55) yields a unique global solution u ∈ Ct0 (Ḣx1 ) ∩ Lt,xn−2 such that
Z ∞ Z
2(n+2)
|u(t, x)| n−2 dxdt 6 C(E(u0 )),
−∞ Rn
July 7, 2011 10:54 World Scientific Book - 9in x 6in booken-main
where
Z
1 n−2 2n
E(u(t)) = |∇u(t, x)|2 + |u(t, x)| n−2 dx
Rn 2 2n
is the energy conserved quantity.
The above theorem implies the scattering results [44; 232; 196]:
Corollary 7.1. Let n > 3. Assume that u0 has finite energies, u is the
2(n+2)
unique global solution of (7.55) on u ∈ Ct0 (Ḣx1 ) ∩ Lt,xn−2 . Then, there exist
finite-energy solutions u± of the linear Schrödinger equation (i∂t + ∆)u± =
0 such that
ku± (t) − u(t)kḢ 1 → 0, t → ±∞,
x
Along with the perfect solution of the scattering theory of the energy-
critical defocusing NLS, most researchers transfer their interests into an-
other important problem – the global well-posedness and scattering theories
of the mass-critical defocusing NLS
4
iut + ∆u = |u| n u, u(0, x) = u0 (x). (7.56)
At present, there are some results in this aspect:
• n > 1, for sufficiently small initial datum u0 ∈ L2 (Rn ), the solution
of (7.56) is global well-posed and scatters.
• n > 2, for radial initial data u0 ∈ L2 (Rn ), the solution of (7.56) is
global well-posed and scatters, cf. [138; 140].
• n > 1, for any initial data u0 ∈ L2 (Rn ), the solution of (7.56) is
global well-posed and scatters, cf. [64; 65; 66].
For other nonlinearities, there are also some interesting results. K.
Nakanishi established the scattering theory for the mixed power nonlinear-
ity |u|p1 u+|u|p2 u for 4/n < p1 6 p2 < 4/(n−2) and 4/n 6 p1 < p2 6 2∗ −2
in the energy space H 1 , cf. [176; 179; 220]. For the nonlinearity (V ∗ |u|2 )u,
i.e. the Hartree equation, he also proved the existence of the scattering
operator in H 1 for V ∈ Lp1 + Lp2 , p1 , p2 > 1 and n/4 < p2 6 p1 < n/2,
cf. [175]. For the nonlinear Schrödinger equation with the following non-
linearity
X λ2k
f (u) = µ |u|2k u, µ, λ > 0, n = 1, 2, (7.57)
(2k)!
k>1+2/n
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
the scattering operator also exists under some assumptions, cf. [244].
In addition, there is a great process for focusing nonlinear Schrödinger
equations
iut + ∆u = −|u|p u,
(7.58)
u(0, x) = u0 (x).
Chapter 8
The kinetic theory of the gas is a theory devoted to the study of evolutionary
behaviors of the gas in the one-particle phase space of position and velocity.
Recently, the kinetic theory is getting more and more recognized to be
significant both in mathematics and practical applications as a key theory
connecting the microscopic and macroscopic theory of gases and fluids. In
this sense, the kinetic theory is in-between or mesoscopic.
In the macroscopic scales where the gas and fluid are regarded as a
continuum, their motion is described by the macroscopic quantities such
as macroscopic mass density, bulk velocity, temperature, pressure, stresses,
heat flux and so on. The Euler and Navier-Stokes equations, compressible or
incompressible, are the most famous equations among governing equations
proposed so far in fluid dynamics.
The extreme contrary is the microscopic scale where the gas, fluid, and
hence any matter, are looked at as a many-body system of microscopic
particles (atom/molecule). Thus, the motion of the system is governed
by the coupled Newton equations, within the framework of the classical
mechanics. The number of the involved equations is 6N if the total number
of the microscopic particles is N .
Although the Newton equation is the first principle of the classical me-
chanics, it is not of practical use because the number of the equations is
so enormous (N ∼ the Avogadro number 6 × 1023 ) that it is hopeless to
specify all the initial data, and we must appeal to statistics. On the other
hand, the macroscopic (fluid dynamical) quantities mentioned above are
related to statistical average of quantities depending on the microscopic
state. Thus, the kinetic theory that gives the mesoscopic descriptions of
the gas and fluid is noticed to be a key theory that links the microscopic and
227
June 29, 2011 10:24 World Scientific Book - 9in x 6in booken-main
In this section, we will first give the derivation of the Boltzmann equation,
in details, one can refer to [28; 59; 229].
We have
Z Z Z
Q(f, f )(v) = f (v 0 )f (v∗0 ) − f (v)f (v∗ ) p(v, v∗ → v 0 , v∗0 )dv∗0 dv 0 dv∗ .
v∗ v0 v∗0
(8.12)
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Let us explain (8.12) a little bit. This operator can formally be split, in a
self-evident way, into a gain and a loss term, Q(f ; f ) = Q+ (f ; f )−Q−(f ; f ):
The loss term counts all collisions in which a given particle of velocity v will
encounter another particle, of velocity v∗ . As a result of such a collision,
this particle will in general change its velocity, and this will make less
particles with velocity v. On the other hand, each time particles collide
with respective velocities v 0 and v∗0 , then the v 0 particle may acquire v
as new velocity after the collision, and this will make more particles with
velocity v : this is the meaning of the gain term.
Next, we only need to give the expression of p(v, v∗ → v 0 , v∗0 ). Using the
conservation of momentum and kinetic energy in a collision, we have
v + v∗ = v 0 + v∗0 ,
(8.13)
|v|2 + |v∗ |2 = |v 0 |2 + |v∗0 |2 .
v0
θ/2 θ
v∗ v
k
v∗0
separation is not allowed for the non-cutoff cross section but is legitimate
as an intermediate step in the calculation.
In this chapter, we mainly consider the inverse-power law potentials,
the cross section B have the form:
B(|v − v∗ |, σ) = |v − v∗ |γ b(cos θ). (8.25)
If γ > 0, it is called hard potentials; in general, we consider the case of
0 < γ 6 2.
If γ = 0, it is called Maxwellian potentials.
If −n < γ < 0, it is called soft potentials; for −2 < γ < 0, it is called
moderately soft potentials, for γ 6 −2 it is called very soft potentials.
Note that if s = 2 then φ(r) becomes Coulomb potential, which leads
to Fokker-Planck-Landau equation, we do not consider this model in this
book.
In this section, we give some basic tools which one often needs for a fine
study of the Boltzmann operator. In details, one can refer to [59; 229].
1.Symmetrization of the collision kernel. In view of formulas (8.14),
the quantity f∗0 f 0 − f∗ f is invariant under the change of variables σ →
−σ. From the physical point of view this reflects the undiscernability of
particles. Thus one can replace (from the very beginning, if necessary) B
by its symmetrized version
B̄(z, σ) = {B(z, σ) + B(z, −σ)}Iz·σ>0 .
This is the reason why we assume that angle θ varies from 0 to π2 in (8.19).
In fact, we get rid of collisions with deflexion angle larger than π2 by this
symmetrization trick. B in (8.19) is indeed B̄ in the above equality.
2. Pre-postcollisional change of variables. A universal tool in the
Boltzmann theory is the involutive change of variables with unit Jacobian
v−v∗
(v, v∗ , σ) → (v 0 , v∗0 , k), where k = |v−v ∗|
is the unit vector along v − v∗ .
Indeed, denote the mapping Φ : (v, v∗ , σ) → (v 0 , v∗0 , k), then it’s easy to
verify that Φ2 = I. Thus the Jacobian of Φ is 1. Then for any function f
we have
Z Z Z
f (v, v∗ , v 0 , v∗0 , σ)dσdv∗ dv
Rn Rn S n−1
Z Z Z (8.26)
v − v∗
= f (v 0 , v∗0 , v, v∗ , )dσdv∗ dv.
Rn Rn S n−1 |v − v∗ |
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
Of course, the transport operator −v ·∇x does not contribute in any change
of the H functional in time. As a consequence, if f = f (t, x, v) is a solution
of the Boltzmann equation, then H(f ) will evolve in time because of the
effects of the collision operator:
Z
d
H(f (t, ·, ·)) = − D(f (t, ·, ·))dx 6 0. (8.35)
dt Rn
x
ρ − |v−u|
2
M (v) = Mρ,u,T (v) = e 2T , (8.38)
(2πT )n/2
where M is called the Maxwellian and is known to describe the velocity
distribution of a gas in an equilibrium state with the mass density ρ(x), bulk
velocity u(x) and temperature T . Here, (ρ, u, T ) are taken to be parameters.
If (ρ, u, T ) are constants,then M is called a global ( absolute) Maxwellian ;
if they are functions of (x, t), then it is called a local Maxwellian. Evidently,
the global Maxwellian is a stationary solution Boltzmann equation (8.17)
without force F . This is the second part of Boltzmann’s H-theorem.
4. Bobylev’s identities . We now turn to more intricate tools introduced
by Bobylev, who first made Fourier transform an extremely powerful tool
in the study of the Boltzmann operator with Maxwellian collision kernel.
Now we give the Fourier transform of Q(g, f ) in (8.19). We first perform the
calculation of the Fourier transform of the gain term in a general Boltzmann
collision operator Q+ (g, f ).
For any test function ϕ(v), using the pre-postcollisional change of vari-
ables in (8.26), we have
Z
Q+ (g, f )ϕ(v)dv
R n
Z Z (8.39)
v − v∗
= B(|v − v∗ |, · σ)g(v∗ )f (v)ϕ(v 0 )dσdv∗ dv.
Rn ×Rn S n−1 |v − v∗ |
Plugging ϕ(v) = e−ivξ in (8.39), we have
F (Q+ (g, f ))(ξ)
Z Z
v − v∗ v+v∗ |v−v∗ |
= B(|v − v∗ |, · σ)g(v∗ )f (v)e−i 2 ·ξ e−i 2 σ·ξ dσdv∗ dv.
R2n S n−1 |v − v∗ |
(8.40)
Using the following general equality:
Z Z
F (k · σ, l · σ)dσ = F (l · σ, k · σ)dσ, |l| = |k| = 1, (8.41)
S n−1 S n−1
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
v,
|v − v∗ |
|v∗ − ψσ (v)| = . (8.56)
k·σ
Applying this change of variable to the first part in the left-hand side of
(8.52), then changing the name v 0 for v,
Z
B(|v − v∗ |, k · σ)g(v∗ )f 2 (v 0 )dσdv
Rn ×S n−1
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
Z
dv
= B(|v − v∗ |, 2(k 0 · σ)2 − 1)g(v∗ )f 2 (v 0 )| |dσdv 0
Rn ×S n−1 dv 0
Z
2n−1
= B(|ψσ (v 0 ) − v∗ |, 2(k 0 · σ)2 − 1)g(v∗ )f 2 (v 0 ) dσdv 0
k0 ·σ> √12 (k 0 · σ)2
Z
2n−1
= B(|ψσ (v) − v∗ |, 2(k · σ)2 − 1)g(v∗ )f 2 (v) dσdv. (8.57)
k·σ> √12 (k · σ)2
Using the fact that 2n−2 sinn−2 2θ cosn−2 2θ = sinn−2 θ and n-dimensional
spherical coordinates, we have
Z
2n−1
B(|ψσ (v) − v∗ |, 2(k · σ)2 − 1) dσ
k·σ> √12 (k · σ)2
Z
|v − v∗ | 2n−1
= B( , 2(k · σ)2 − 1) dσ
k·σ> √12 k·σ (k · σ)2
Z π4
|v − v∗ | 2n−1
= |S n−2
| sinn−2 θB( , cos 2θ) 2 dθ
0 cos θ cos θ
Z π2
θ |v − v ∗ | 2n−2
= |S n−2 | sinn−2 B( θ
, cos θ) 2 θ dθ
0 2 cos 2 cos 2
Z π2 n−2
sin θ |v − v∗ |
= |S n−2 | θ
B( , cos θ)dθ. (8.58)
0
n
cos 2 cos θ2
Define
A(v − v∗ )
Z π
2 1 |v − v∗ |
=S n−2
sinn−2 θ θ
B( θ
, cos θ) − B(|v − v∗ |, cos θ) dθ.
0 cosn 2 cos 2
(8.59)
This completes the proof.
that if 0 6 γ 6 2
Z
B(|v − v∗ |, k · σ)g(v∗ )(f 2 (v 0 ) − f 2 (v))dσdvdv∗
R2n ×S n−1 (8.61)
6 CkgkL12 kf kL21 ,
θ
which follows from the fact that 1 − cosn 2 6 n(1 − cos 2θ ) = 2n sin 4θ .
Proof. We shall do the proof only in the case when b is integrable and
the result will follow by monotonicity. First, we have
(f (v 0 ) − f (v))2 = f 2 (v 0 ) − 2f (v 0 )f (v) + f 2 (v). (8.63)
0
We begin with the middle term 2f (v )f (v) in (8.63). By the pre-
postcollisional change of variables and Parseval’s identity and Bobylev’s
identity, we have
Z Z
b(k · σ)g(v∗ )f (v 0 )f (v)dσdv∗ dv = Q+ (g, f )f dv
1 + 1
= (Q (g, f ), f )L2 + (f, Q+ (g, f ))L2
2 2 (8.64)
1 1
= (F Q+ (g, f ), F f )L2 + (F f, F Q+ (g, f ))L2
2Z 2
1 ξ
= b( · σ) ĝ(ξ )f (ξ )f (ξ) + ĝ(ξ − )fˆ(ξ + )fˆ(ξ) dξdσ.
− ˆ + ˆ
2 |ξ|
R
For the third term f 2 (v) in (8.63), using the fact that S n−1 b(k · σ)dσ does
not depend on the unit vector k, we have
Z Z Z Z
b(k · σ)g(v∗ )f 2 (v)dσdv∗ dv = b(k · σ)dσ g(v∗ )dv∗ f 2 (v)dv
S n−1 Rn Rn
Z
ξ
= b( · σ)ĝ(0)|fˆ(ξ)|2 dξdσ. (8.65)
|ξ|
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
For the first term f 2 (v 0 ) in (8.63), we first make the change of variables
(v − v∗ , v∗ ) → (v1 , v∗ ), change the name v1 for v, and then use the change
of variables v → v 0 as in Lemma 8.1 to obtain
Z
v − v∗ v + v∗ |v − v∗ |
b( · σ)g(v∗ )f 2 ( + σ)dσdv∗ dv
|v − v∗ | 2 2
Z
v1 v1 + |v1 |σ 2
= b( · σ)g(v∗ )|τ−v∗ f ( )| dσdv∗ dv1
|v1 | 2
Z (8.66)
v v + |v|σ 2
= b( · σ)g(v∗ )|τ−v∗ f ( )| dσdv∗ dv
|v| 2
Z n−1
2
= b(ψ(v 0 , σ)g(v∗ ) v0 |τ−v∗ f (v 0 )|2 dσdv∗ dv 0 ,
( |v0 | · σ)2
where
v0
ψ(v 0 , σ) = 2( 0 · σ)2 − 1, τ−v∗ f = f (v∗ +).
|v |
R
Using the fact that S n−1 b(k · σ) does not depend on k and |F (τh f )| =
|F (f )|, and reversing the change of variables ξ → ξ + as in Lemma 8.1, we
have
Z Z
2n−1
(8.66) = g(v∗ ) b(ψ(ξ, σ) ξ |f (ξ)|2 dσdξ dv∗
( |ξ| · σ)2
Z (8.67)
ξ + 2
= ĝ(0) b( · σ)|f (ξ )| dσdξ.
|ξ|
Collecting (8.63), (8.64), (8.65) and (8.67), we can obtain (8.62).
Lemma 8.3. Assume that b satisfies (8.21), Then there exists a positive
constant Cg depending on n, kgkL11 , kgkL log L and b, such that for |ξ| > 1
Z
ξ
b( · σ)(ĝ(0) − |ĝ(ξ − )|)dσ > Cg |ξ|ν . (8.69)
S n−1 |ξ|
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
Lemma 8.3 is a consequence of the two lemmas below, next we will use
Cg to denote a positive constant depending on n, kgkL11 , kgkL log L and b.
ε > 0 and r > 0 being chosen in such a way that this quantity is positive.
ε
If |ξ| 6 1, let δ = |ξ| in (8.71), Lemma 8.4 holds with the following
constant
Z
2 sin2 (δ|ξ|) n kgkL11 o
Cg = 2δ inf kgk L 1 − − sup g(v)dv ,
|ξ|61 δ 2 |ξ|2 r r
|A|64ε(2r)n−1 (1+ π ) A
(8.73)
where δ > 0 and r > 0 are chosen in such a way that this quantity is
positive.
Lemma 8.5. If b satisfies
K(ν)
sinn−2 θb(cos θ) ≈ as θ → 0, K(ν) > 0; (8.74)
θ1+ν
then for |ξ| > 1,
Z
ξ
b( · σ)(|ξ − |2 ∧ 1)dσ > K(ν)|ξ|ν . (8.75)
S n−1 |ξ|
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
In this section, we will give the application of the results of Section 3 in this
chapter; and prove the regularity of solutions for the spatially homogeneous
Boltzmann equation without angular cutoff
∂t f (t, v) = Q(f, f )(t, v), t > 0, v ∈ Rn ,
(8.79)
f (0, v) = f0 (v),
where cross section B
v − v∗ v − v∗
B(|v − v∗ |, · σ) = |v − v∗ |γ b( · σ),
|v − v∗ | |v − v∗ |
K(ν)
sinn−2 θb(cos θ) ≈ when θ → 0, K(ν) > 0.
θ1+ν
June 29, 2011 10:26 World Scientific Book - 9in x 6in booken-main
From (8.33), it follows that the solutions f (t, v) of the Cauchy problem
(8.79) have the following conservation of mass, momentum and energy,
Z Z
f (t, v)dv = f0 (v)dv, (8.80)
Rn Rn
Z Z
f (t, v)vj dv = f0 (v)vj dv, j = 1, 2...n, (8.81)
Rn Rn
Z Z
|v|2 |v|2
f (t, v) dv = f0 (v) dv. (8.82)
Rn 2 Rn 2
For the existence of weak solutions for Cauchy problem (8.79), in 1998
Villani [230] showed that if the initial data have the finite mass, energy and
entropy,
Z
f0 (v)[1 + |v|2 + log(1 + f0 (v))]dv < +∞, (8.83)
Rn
then he constructed a weak solution f (t, v) of the Cauchy problem (8.79),
which satisfies the following:
f (t, v) > 0, f (t, v) ∈ C(R+ , S 0 ); t > 0, f (t, v) ∈ L12 ∩ LlogL, (8.84)
These make sense provided that f satisfies (8.84) and test function ϕ ∈
L∞ ([0, T ]; W 2,∞ ).
In this section, we assume that a weak solution to the Boltzmann equa-
tion (8.79) has already been constructed and that it satisfies the usual
entropic estimates, to say,
Z
f (t, v)[1 + |v|2 + log(1 + f (t, v))]dv < +∞, (8.90)
Rn
we are interested in regularity issues associated to such solutions. That is,
is this weak solution more regular than the initial datum, and if so, can we
have estimates on this regularity?
In this section, we will show that the weak solutions above constructed
f (t, v) of the Cauchy problem (8.79) with Maxwellian molecules case be-
longs to Hv+∞ . For hard potentials case, there exist some results only
in modified hard potentials cases now, in details, one can refer to [4; 60;
111]. For soft potentials cases, there seems no any results now.
In this section, we will introduce two methods of multiplier to consider
it. That is, we first choose a team of suitable multiplier, then mainly give
the sharp estimates of the commutators of the collision operator Q(f, f )
and pseudo-differential operators composed by multipliers.
First, we give the method of the Littlewood-Paley decomposition, in
details, one can refer to [3]. The Littlewood-Paley decomposition is defined
as in Section 3 in Chapter 1: Assume that supp ϕk ⊂ {ξ : 2k−1 6 |ξ| 6
P∞
2k+1 }, k ∈ N, supp ϕ0 ⊂ {ξ : |ξ| 6 2}, k=0 ϕk = 1; we can assume
ϕk is a radial function. Recall that for every multi-index α, there exists a
positive number Cα such that
2k|α| |Dα ϕk (ξ)| 6 Cα , k = 0, 1, 2...; ξ ∈ Rn . (8.91)
Then the Littlewood-Paley projection operator is defined by
d
4 ˆ n
k f (ξ) = ϕk (ξ)f (ξ), k = 0, 1, 2...; ξ ∈ R .
Lemma 8.6. Assume that the initial data f0 (v) satisfies (8.83), cross sec-
tion B(|v − v∗ |, k · σ) = b(k · σ) with b satisfying the following
K
sinn−2 θb(cos θ) ≈ 1+ν when θ → 0, K > 0. (8.92)
θ
Let f (t, v) be any weak non negative solution of the Cauchy problem (8.79),
satisfying (8.80)-(8.82) and (8.90). Then
k+1
X
(4k Q(f, f ), 4k f )L2 − (Q(f, 4k f ), 4k f )L2 6 Ckf0 kL1 ( k4k f kL2 ),
j=k−2
(8.93)
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
k+1
X
(4k f, 4k Q(f, f ))L2 − (4k f, Q(f, 4k f ))L2 6 Ckf0 kL1 ( k4k f kL2 ).
j=k−2
(8.94)
Proof. We only prove (8.93), the proof of (8.94) is similar with one of
(8.93). Applying Fourier transform on Boltzmann equation (8.79), using
(8.49) and (8.50) (Bobylev identity), we have
Z
\ ξ
∂t fˆ(ξ) = Q(f, f )(ξ) = b( · σ) fˆ(ξ − )fˆ(ξ + ) − fˆ(0)fˆ(ξ) dσ. (8.95)
S n−1 |ξ|
d
∂t 4 \
k f (ξ) = 4k Q(f, f )(ξ)
Z
ξ
= b( · σ) fˆ(ξ − )fˆ(ξ + )ϕk (ξ) − fˆ(0)fˆ(ξ)ϕk (ξ) dσ.
S n−1 |ξ|
(8.96)
Then,
Z Z
d
∂t (4 d
k f , 4k f )L2 =
d
(∂t 4 d
k f )4k f dξ +
d
4 d
k f ∂t 4k f dξ
Rn Rn
= (4k\Q(f, f ), 4d d \
k f )L2 + (4k f , 4k Q(f, f ))L2
Z Z
ξ
= b( · σ) fˆ(ξ − )fˆ(ξ + )ϕ2k (ξ)fˆ(ξ) − fˆ(0)fˆ(ξ)ϕ2k (ξ)fˆ(ξ) dσdξ
n n−1 |ξ|
ZR ZS
ξ
+ b( · σ) fˆ(ξ − )fˆ(ξ + )ϕ2k (ξ)fˆ(ξ) − fˆ(0)fˆ(ξ)ϕ2k (ξ)fˆ(ξ) dσdξ.
Rn S n−1 |ξ|
(8.97)
Moreover, we have
Z
\ ξ
Q(f, 4k f )(ξ) = b( · σ) fˆ(ξ − )ϕk (ξ + )fˆ(ξ + ) − fˆ(0)ϕk (ξ)fˆ(ξ) dσ,
S n−1 |ξ|
(8.98)
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
\
(Q(f, d
4k f ), 4 d \
k f )L2 + (4k f , Q(f, 4k f ))L2
ZZ
ξ
= b( · σ) fˆ(ξ − )ϕk (ξ + )fˆ(ξ + )ϕk (ξ)fˆ(ξ) − fˆ(0)fˆ(ξ)ϕ2k (ξ)fˆ(ξ) dσdξ
|ξ|
ZZ
ξ
+ b( · σ) fˆ(ξ − )ϕk (ξ + )fˆ(ξ + )ϕk (ξ)fˆ(ξ) − fˆ(0)fˆ(ξ)ϕ2k (ξ)fˆ(ξ) dσdξ.
|ξ|
(8.99)
Thus, from (8.97) and (8.99) it follows that
(4k\ d
Q(f, f ), 4 \ d
k f )L2 − (Q(f, 4k f ), 4k f )L2
Z Z
ξ
= b( · σ)fˆ(ξ − )fˆ(ξ + )(ϕk (ξ) − ϕk (ξ + ))ϕk (ξ)fˆ(ξ)dσdξ.
R n S n−1 |ξ|
(8.100)
From the definitions of ξ + and ξ − , it follows that
ξ + |ξ|σ θ ξ π
ξ+ = , |ξ + | = |ξ| cos , · σ = cos θ, 0 6 θ 6 ,
2 2 |ξ| 2
|ξ|2 θ
6 |ξ + |2 6 |ξ|2 , |ξ|2 − |ξ + |2 = |ξ − |2 = |ξ|2 sin2 .
2 2
From the integral in (8.100), we note that 2k−1 6 |ξ| 6 2k+1 . From the
definition of ξ + , it follows that 2k−2 6 |ξ + | 6 2k+1 . Thus
\
fˆ(ξ + ) = 4 + \ + d + \ +
k−2 f (ξ ) + 4k−1 f (ξ ) + 4k f (ξ ) + 4k+1 f (ξ ). (8.101)
Plugging this expression of f (ξ + ) in (8.100), and by the fact that |fˆ(ξ − )| 6
kf kL1 6 kf0 kL1 , (8.100) is bounded by
Z Z
ξ d + k d
kf0 kL1 b( · σ)|4 k f (ξ )||Aξ ||4k f (ξ)|dσdξ
2 k−1 6|ξ|62k+1 S n−1 |ξ|
Z Z
ξ \ + k d
+ kf0 kL 1 b( · σ)|4 k−2 f (ξ )||Aξ ||4k f (ξ)|dσdξ
2k−1 6|ξ|62k+1 S n−1 |ξ|
Z Z
ξ \ + k d
+ kf0 kL1 b( · σ)|4 k−1 f (ξ )||Aξ ||4k f (ξ)|dσdξ
2k−1 6|ξ|62k+1 S n−1 |ξ|
Z Z
ξ \ + k d
+ kf0 kL1 b( · σ)|4 k+1 f (ξ )||Aξ ||4k f (ξ)|dσdξ,
2k−1 6|ξ|62k+1 S n−1 |ξ|
(8.102)
where Akξ = ϕk (ξ) − ϕk (ξ + ). By (8.91), we have
|Akξ | = |ϕk (ξ) − ϕk (ξ + )| 6 |ϕ̃k (|ξ|) − ϕ̃k (|ξ + |)|
1 + 1 |ξ|2 − |ξ + |2 θ (8.103)
. (|ξ| − |ξ |) . . sin2 .
2k 2k |ξ| + |ξ + | 2
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
We only estimate the first term in (8.102), the estimates of the other terms
in (8.102) can be obtained similarly. By (8.103) and Hölder inequality,
(8.102) is bounded by
Z Z
ξ d + k d
kf0 kL1 b( · σ)|4 k f (ξ )||Aξ ||4k f (ξ)|dσdξ
2k−1 6|ξ|62k+1 S n−1 |ξ|
Z Z
d ξ θ d +
. kf0 kL1 |4 k f (ξ)| b( · σ) sin2 |4 k f (ξ )|dσdξ
Rn S n−1 |ξ| 2
Z 1/2 nZ Z ξ θ d + 2 o1/2
. kf0 kL1 |4d 2
k f (ξ)| dξ b( · σ) sin2 |4 k f (ξ )|dσ dξ
|ξ| 2
. kf0 kL1 k4k f k2L2 . (8.104)
+
which follows from the change of variables ξ → ξ as in Lemma 8.1.
Theorem 8.1. Under the hypothesis of Lemma 8.6, Let f (t, v) be any weak
non negative solution of the Cauchy problem (8.79), satisfying (8.90). Then
for any s ∈ R+ and t > 0, one has f (t, v) ∈ H s (Rn ).
From 1 − cosn θ
2 6 n(1 − cos 2θ ) = 2n sin2 θ
4 it follows that
I1 . k4k f k2L2 . (8.109)
For the second term in (8.106), using Corollary 8.1 and Lemma 8.3 in this
chapter, we have
Z
1 v − v∗
I2 = b( · σ)f (v∗ )(4k f (v 0 ) − 4k f (v))2 dσdv∗ dv
2 R2n ×S n−1 |v − v∗ |
Z
> Cf0 (1 + |ξ|ν )ϕk (ξ)fˆ(ξ)dξ > Cf0 k4k f k2 ν2 .
H
Rn
(8.110)
Thus collecting (8.105), (8.106), (8.109), (8.110) and Lemma 8.6, we get
k+1
X
∂t k4k f kL2 + Cf0 k4k f k2H ν2 6 Ckf0 kL1 ( k4j f k2L2 ). (8.111)
j=k−2
for all k > 2(p − 1) + k0 + 2. Thus, for fixed s > 0, t > 0 and an integer
p > 1 which we shall choose just below; using the definition of Sobolev
spaces, we have
∞
X
kf k2H s ∼ 22ks k4k f k2L2
k=0
2(p−1)+k0 +1 ∞
X X
= 22ks k4k f k2L2 + 22ks k4k f k2L2 (8.116)
k=0 k=2(p−1)+k0 +2
2(p−1)+k0 +1 ∞
X X
= 22ks k4k f k2L2 + 2k(2s+n) Uk .
k=0 k=2(p−1)+k0 +2
In order to prove f ∈ H s , it remains to show that the last series appearing
above is a convergent one. From (8.115) it follows that
∞
X
2k(2s+n) Uk
k=2(p−1)+k0 +2
∞
X
6 2k(2s+n) M Ap e−Ck−2(p−1)t (8.117)
k=2(p−1)+k0 +2
∞
X 1
+ 2k(2s+n) M Dp .
(Ck−2(p−1) )p
k=2(p−1)+k0 +2
Using the definition of Ck , we have
Ck−2(p−1) ∼ C(f0 , p)2kνp , if k is large enough. (8.118)
Choosing p such that νp > 2s + n + 1 for instance, yields that the two series
on the right hand side of the last inequality are convergent.
Lemma 8.7. (Iteration) Let β, M be two non negative numbers. Given
positive integer k0 , assume that {Ck }k>k0 is a sequence of positive numbers,
and is non-decreasing in k, and satisfies: there exists a positive α such that
Ck+1 − Ck > α for all k > k0 . {Uk }k>k0 = {Uk (t)}k>k0 is another sequence
with t ∈ R+ satisfying:
∂t Uk (t) + Ck Uk (t) 6 β(Uk−2 (t) + Uk−1 (t) + Uk+1 (t)),
0 6 Uk (t) 6 M, ∀k > k0 + 2, ∀t > 0.
Then for any integer p > 1, there exist constants Ap , Dp such that, for all
k > 2(p − 1) + k0 + 2, one has
1
Uk (t) 6 M Ap e−Ck−2(p−1)t + M Dp , t > 0.
(Ck−2(p−1) )p
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
Lemma 8.8. Under the hypothesis of Lemma 8.6, Let f (t, v) be any weak
non negative solution of the Cauchy problem (8.79), satisfying (8.90). Then
(Q(f, f ), Mδ2 f )L2 − (Q(f, Mδ f ), Mδ f )L2 6 Cf kf0 kL1 kMδ2 f kL2 , (8.119)
where the constant Cf is independent of 0 < δ < 1.
From the mean value theorem, it follows that there exists s+ < s̃ < s such
that
∂ M̃δ
M̃δ (t, s) − M̃δ (t, s+ ) = (t, s̃)(s − s+ ).
∂s
By the definition of M̃δ ,
∂ M̃δ Nt − 4 δN0
(t, s) = − M̃δ (t, s).
∂s 2(1 + s) 1 + δs
By the following
s δs M̃ (t, s̃) θ
δ N T0 +4
, 6 1; 6 2 2 , s − s+ = s sin2 ,
1 + s 1 + δs +
M̃δ (t, s ) 2
we have
N T0 +4 θ
M̃δ (t, s) − M̃δ (t, s+ ) 6 N0 2 2 M̃δ (t, s+ ) sin2 .
2
Then
Z Z
ξ
· σ)fˆ(ξ − )fˆ(ξ + )(Mδ (t, ξ) − Mδ (t, ξ + ))Mδ (t, ξ)fˆ(ξ)dσdξ
b(
Rn S n−1 |ξ|
Z Z
ξ θ
6 b( · σ) sin2 |fˆ(ξ − )||fˆ(ξ + )|Mδ (t, ξ + )Mδ (t, ξ)|fˆ(ξ)|dσdξ
R n S n−1 |ξ| 2
6 Ckf0 kL1 kMδ f k2L2 .
Thus using Lemma 8.8, we give another proof of Theorem 8.1. We
note that any weak solution f has the following properties: Mδ2 f ∈
L∞ ([0, T0 ]; W 2,∞ (R3 )) and Mδ f ∈ C([0, T0 ]; L2 (R3 )). From the definition
of weak solution, we choose ψ(t, v) = Mδ2 f (t, v) as test function. For any
t ∈ (0, T0 ),
Z Z
f (t, v)Mδ2 f (t, v)dv − f0 (v)Mδ2 f (0, v)dv
R3 R3
Z t Z
− dτ f (τ, v)∂τ (Mδ2 f (τ, v))dv (8.122)
0 R3
Z t Z
= dτ Q(f, f )(τ, v)Mδ2 f (τ, v)dv.
0 R3
Moreover, we have
Z t Z
dτ f (τ, v)∂τ (Mδ2 f (τ, v))dv
0 R3
Z t Z
Mδ2 f (τ + h, v) − Mδ2 f (τ, v)
= lim dτ (f (τ, v) + f (τ + h, v)) dv
h→0 0 R3 2h
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
Z t Z
= lim dτ
h→0 0 R3
n M f (τ + h, v)
δ
Mδ (τ + h)f (τ, v) + Mδ (τ + h)f (τ + h, v)
2h
M f (τ, v) o
δ
− Mδ (τ )f (τ, v) + Mδ (τ )f (τ + h, v) dv
2h
Z t Z
(Mδ f )2 (τ + h, v) − (Mδ f )2 (τ, v)
= lim dτ dv
h→0 0 R3 2h
Z t Z
1 n
+ lim dτ Mδ (τ + h)(f (τ, v))Mδ (τ + h)(f (τ + h, v))
h→0 0 R3 2h
o
− Mδ (τ )(f (τ, v))Mδ (τ )(f (τ + h, v)) dv
= J1 + J2 . (8.123)
For J1 , we have Z Z
t
(Mδ f )2 (τ + h, v) − (Mδ f )2 (τ, v)
J1 = lim dτ dv
h→0 0 R3 2h
Z Z h oZ
1 n t+h
= lim dτ − dτ (Mδ f )2 (τ, v)dv
h→0 2h 0 0 R3
Z Z
1 1
= (Mδ f )2 (t, v)dv − (Mδ f )2 (0, v)dv. (8.124)
2 R3 2 R3
For J2 , we have
Z t Z
1 n
J2 = lim dτ Mδ (τ + h)(f (τ, v))Mδ (τ + h, v)(f (τ + h, v))
h→0 0 R3 2h
o
− Mδ (τ )(f (τ, v))Mδ (τ )(f (τ + h, v)) dv
Z t Z
1 n
= lim dτ f (τ, v)Mδ2 (τ + h, v)(f (τ + h, v))
h→0 0 R3 2h
o
− f (τ, v)Mδ2 (τ )(f (τ + h, v)) dv
Z t Z n o
1
= lim dτ f (τ, v) Mδ2 (τ + h) − Mδ2 (τ ) (f (τ + h, v)) dv
h→0 0 R3 2h
Z t Z
1
= dτ f (τ, v)(∂τ Mδ2 (τ ))(f (τ, v))dv. (8.125)
2 0 R 3
Similarly with the first proof of Theorem 8.1, or by the content of Section
3 in this chapter, we have
n o
kMδ f k2H ν2 6 Cf (−Q(f, Mδ f ), Mδ f )L2 + kMδ f k2L2 . (8.127)
From Lemma 8.8, it follows that
n o
ef (−Q(f, f ), M 2 f )L2 + kMδ f k2 2 .
kMδ f k2H ν2 6 C (8.128)
δ L
Since
∂t Mδ (t, ξ) = N Mδ (t, ξ) loghξi,
we obtain
Z t Z Z t
1
2
dτ f (τ, v)(∂τ Mδ (τ ))(f (τ, v))dv 6 2N k(log Λ) 2 Mδ f (τ )k2L2 dτ,
0 R3 0
(8.129)
−1 −1
where log Λ = F loghξiF and Λ = F hξiF . This together with (8.126)
and (8.128), imply
Z t
1 ν
kMδ f (t)k2L2 + kΛ 2 Mδ f (τ )k2L2 dτ
2Cf 0
Z t Z t
1
2 2
6 kMδ f (0)kL2 + 2N k(log Λ) Mδ f (τ )kL2 dτ +
2 kMδ f (τ )k2L2 dτ.
0 0
(8.130)
Using loghξi 6 hξi and Gagliardo-Nirenberg interpolation inequality, for
any ε > 0, we have
1 Z t ν
2
kMδ f (t)kL2 + −ε kΛ 2 Mδ f (τ )k2L2 dτ
2Cf 0
Z t (8.131)
2 2
6 kMδ f (0)kL2 + Cε,N kMδ f (τ )kL2 dτ.
0
1
By choosing ε = 4Cf > 0, there exists a constant Cf,N depending only on
Cf , N , T0 and being independent of δ ∈ (0, 1), such that for any t ∈ (0, T0 ),
Z t
kMδ f (t)k2L2 6 kMδ f (0)k2L2 + Cf,N kMδ f (τ )k2L2 dτ. (8.132)
0
Then Gronwall inequality yields
kMδ f (t)k2L2 6 eCf,N,t kMδ (0)f0 )k2L2 . (8.133)
Since
kMδ f (t)kL2 = k(1 − δ∆)−N0 f (t)kH N t−4 ,
June 15, 2011 11:58 World Scientific Book - 9in x 6in booken-main
and
kMδ (0)f0 kL2 (R3 ) = k(1 − δ∆)−N0 f0 kH −4 (R3 ) 6 kf0 kH −4 (R3 ) 6 Ckf0 kL1 (R3 ) ,
we obtain
k(1 − δ∆)−N0 f (t)kH N t−4 6 C̃eCf,N,t kf0 kL1 (R3 ) ,
where the constant C̃ is independent of δ ∈ (0, 1). Finally, for any given
t > 0 since N can be arbitrarily large, by letting δ → 0, we have f (t) ∈
H +∞ (R3 ).
Remark 8.3. Recently, the existence of the classical global solutions of the
Boltzmann equation without angular cut-off was obtained by Gressman
and Strain [83]. There are some recent works by Chen, Li and Xu [34;
36] which have been devoted to the study of the regularity for the Landau
equation, which can be regarded as a limit of the Boltzmann equation when
the collisions become grazing. Mouhot and Villani [172] for the first time
establish the Landau damping in a nonlinear context.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Appendix A
Notations
We list some notations used in this book, most of them are familiar for the
PDE readers.
259
This page intentionally left blank
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Appendix B
Remark B.1. The scattering theory involves two essential factors: the ex-
istence of the forward/backward wave operators and the asymptotic com-
pleteness. Generally speaking, the existence of wave operators is relatively
easy to establish as long as the power p is suitable, and especially if a small-
ness condition is assumed, both in focusing and defocusing cases. However,
261
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
the asymptotic completeness is a bit harder and requires some decay esti-
mates even in the defocusing case.
Remark B.2. To reduce the number of cases slightly, we shall only consider
scattering from t = 0 to t = +∞ or vice versa. One can certainly consider
scattering back and forth between t = 0 to t = −∞, or between t = −∞
and t = +∞, but the theory is more or less the same in each of these cases.
Appendix C
263
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Proof. We can assume that {aj } = {aj }j>0 , aj > 0 and s1 < s0 . Put
Ci = supj>0 2si j aj , i = 0, 1. It suffices to consider the case C1 > 0. We
have C1 6 C0 . Take j0 such that
C0 /2s0 j , j > j0 ,
min(C0 /2s0 j , C1 /2s1 j ) =
C1 /2s1 j , j 6 j0 .
It is easy to see that C0 ∼ C1 2(s0 −s1 )j0 . So,
k2s0 j aj k1−θ
`∞ k2
s1 j
aj kθ`∞ ∼ C1 2(s0 −s1 )j0 (1−θ) . (C.4)
On the other hand,
aj 6 C0 /2s0 j , j > j0 ; aj 6 C1 /2s1 j , 0 6 j 6 j0 .
A simple calculation yields
k2sj aj k`q . C1 2(s0 −s1 )j0 (1−θ) , (C.5)
which implies the result, as desired.
where |{· · · }| denotes the measure of the set {· · · }. It is easy to see that
∞
X
2ks2 |4k f | . 2K(s2 −s1 ) sup 2ks1 |4k f |. (C.9)
k
k=K+1
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
By Corollary 1.1,
T : Lpi → Lqi , i = 0, 1.
1 1−θ θ 1 1−θ θ
= + , = + .
p p0 p1 q q0 q1
kT kLp→Lq 6 kT k1−θ θ
Lp0 →Lq0 kT kLp1 →Lq1 .
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Proposition C.1. Let 1 < p, q < ∞ and 0 < α < n satisfy 1/p = 1/q +
α/n. Then we have
kIα f kq . kf kp . (C.14)
Lemma C.2. Let ϕ ∈ C0∞ (R). Assume that P ∈ C 2 (R) satisfies that for
any ξ ∈ supp ϕ, |P (k) (ξ)| > 1. Moreover, we assume that the following
alternative condition holds
(i) k > 2;
(ii) k = 1 and P 0 (x) is a monotone function.
Then we have Z
eiλP (ξ) ϕ(ξ)dξ . λ−1/k (kϕk∞ + kϕ0 k1 ).
In particular,
kukLp ∼ kukḞ 0 , kukLp ∼ kukFp,2
0 . (C.16)
p,2
Recall that the result of Theorem C.4 is quite similar to Besov spaces,
indeed, if (C.17) holds, then we have
(Bps00 ,q0 , Bps11 ,q1 )θ = Bp,q
s
. (C.18)
See [245].
The Christ-Kiselev lemma (cf. [46]) is very useful for the study of nonlinear
dispersive equations. There are a series of generalizations to the Christ-
Kiselev lemma in recent years, see Molinet-Ribaud [166] and Smith-Sogge
[201]. The following result is due to Wang-Han-Huang [243]. Denote
Z ∞ Z t
0 0 0
T f (t) = K(t, t )f (t )dt , Tre f (t) = K(t, t0 )f (t0 )dt0 . (C.19)
−∞ 0
Bibliography
269
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
270 Bibliography
[18] J. Bourgain, Global solutions of nonlinear Schrödinger equation, Amer. Math. Soc.,
1999.
[19] J. Bourgain, Global well posedness of defocusing critical nonlinear Schrödinger
equation in the radial case, J. Amer. Math. Soc., 12 (1999), 145–171.
[20] J. Bourgain, J. Colliander, On well-posedness of the Zakharov system, Int. Math.
Res. Not., 11 (1996), 515–546.
[21] P. Brenner, On space-time means and everywhere defined scattering operators for
nonlinear Klein–Gordon equations, Math. Z., 186 (1984), 383–391.
[22] P. Brenner, On scattering and everywhere-defined scattering operators of nonlonear
Klein-Gordon equations, J. Differential Equations, 56 (1985), 310–344.
[23] H. Brezis and P. Mironescu, Gagliardo–Nirenberg, composition and products in
fractional Sobolev spaces, J. Evol. Equ., 1 (2001), 387–404.
[24] L. Carleson, Some analytical problems related to statistical mechanics, Euclidean
Harmonic Analysis, Lecture Notes in Math., Vol. 779, Springer, Berlin, 1980, 5–45.
[25] T. Cazenave, F. B. Weissler, Some remarks on the nonlinear Schrödinger equation
in the critical case, Lecture Notes in Math., 1394 (1989), 18–29.
[26] T. Cazenave and F.B. Weissler, The Cauchy problem for the critical nonlinear
Schrödinger equation in H s , Nonlinear Anal. TMA, 14 (1990), 807–836.
[27] T. Cazenave, Semilinear Schrödinger equations, Courant Lecture Notes in Mathe-
matics, Vol. 10, 2003.
[28] C. Cercignani, The Boltzmann Equation and Its Applications, Springer–Verlag,
1988.
[29] J. Y. Chemin, Théormes d’unicité pour le système de Navier–Stokes tridimensionnel,
Journal d’Anal. Math. 77 (1999), 27–50.
[30] J. Y. Chemin, I. Gallagher, On the global wellposedness of the 3D Navier–Stokes
equations with large initial data. Ann. Sci. École Norm. Sup. 39 (2006), 679–698.
[31] J. Y. Chemin, I. Gallagher, Wellposedness and stability results for the Navier-Stokes
equations in R3 , Ann. Inst. H. Poincaré, Anal. Non Linéaire, 26 (2009), 599–624.
[32] J. Y. Chemin, I. Gallagher, Large, global solutions to the Navier–Stokes equations,
slowly varying in one direction, Tran. Amer. Math. Soc., 362 (2010), 2859–2873.
[33] J. Y. Chemin, P. Zhang, On the global wellposedness to the 3-D incompressible
anisotropic Navier–Stokes equations. Comm. Math. Phys., 272 (2007), 529–566.
[34] H. Chen, W. X. Li, C. J. Xu, Propagation of Gevrey regularity for solutions of
Landau equations, Kinet. Relat. Models, 1 (2008), 355–368.
[35] H. Chen, W. X. Li, C. J. Xu, Gevrey hypoellipticity for linear and non-linear Fokker–
Planck equations, J. Differential Equations, 246 (2009), 320–339.
[36] H. Chen, W. X. Li, C. J. Xu, Analytic smoothness effect of solutions for spatially
homogeneous Landau equation, J. Differential Equations, 248 (2010), 77–94.
[37] H. Chen, W. X. Li, C. J. Xu, Gevrey hypoellipticity for a class of kinetic equations,
Commun. Part. Diff. Equations, 36 (2011), 693–728.
[38] J. C. Chen, D. S. Fan and L. Sun, Asymptotic estimates For unimodular Fourier
multipliers on modulation spaces, Preprint.
[39] Cheskidov A., Shvydkoy R, The regularity of weak solutions of the 3D Navier-Stokes
−1
equations in B∞,∞ , Arch. Ration. Mech. Anal., 195 (2010), 159–169.
[40] J. Colliander, Nonlinear Schrödinger equations, Lecture notes, 2004.
[41] J. E. Colliander, J. M. Delort, C. E. Kenig and G. Staffilani, Bilinear estimates and
applications to 2D NLS, Trans. Amer. Math. Soc., 353 (2001), 3307–3325.
[42] J. Colliander, M. Keel, G. Staffilani, H. Takaoka and T. Tao, Sharp global well-
posedness for KdV and modified KdV on R and T, J. Amer. Math. Soc., 16 (2003),
705–749.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Bibliography 271
[43] J. Colliander, M. Keel, G. Staffilani, H. Takaoka and T. Tao, Viriel, Morawetz, and
interaction Morawetz inequalities, Lecture notes, 2004.
[44] J. Colliander, M. Keel, G. Staffilani, H. Takaoka, and T. Tao, Global well-posedness
and scattering for the energy-critical nonlinear Schrödinger equation in R3 , Ann. of
Math., 167 (2008), 767–865.
[45] J. Colliander, M. Keel, G. Staffilani, H. Takaoka and T. Tao, Resonant de-
compositions and the I-method for cubic nonlinear Schrödinger equation on R2 ,
arXiv:0704.2730.
[46] M. Christ, A. Kiselev, Maximal functions associated to filtrations, J. Funct. Anal.,
179 (2001), 406–425.
[47] M. Christ, J. Colliander and T. Tao, Asymptotics, frequency modulation and low
regularity ill-posedness for canonical defocusing equations, Amer. J. Math., 125
(2003), 1235–1293.
[48] P. A. Clarkson and J. A. Tuszyriski, Exact solutions of the multidimensional deriva-
tive nonlinear Schrödinger equation for many-body systems near criticality, J. Phys.
A: Math. Gen. 23 (1990), 4269–4288.
[49] E. Cordero, F. Nicola, Some new Strichartz estimates for the Schrödinger equation.
J. Differential Equations, 245 (2008), 1945–1974.
[50] E. Cordero, F. Nicola, Remarks on Fourier multipliers and applications to the wave
equation, J. Math. Anal. Appl., 353 (2009), 583–591.
[51] E. Cordero, F. Nicola, Sharpness of some properties of Wiener amalgam and mod-
ulation spaces. Bull. Aust. Math. Soc., 80 (2009), 105–116.
[52] S. B. Cui, Pointwise estimates for a class of oscillatory integrals and related Lp –Lq
estimates, J. Fourier Anal. Appl., 11 (2005), 441–457.
[53] S. B. Cui, Pointwise estimates for oscillatory integrals and related Lp -Lq estimates,
II Multidimensional case, J. Fourier Anal. Appl., 12 (2006), 605–627.
[54] J. M. Delort, Existence globale et comportement asymptotique pour léquation de
Klein–Gordon quasi-linéaire à données petites en dimension 1, Ann. Sci. École
Norm. Sup., 34 (2001), 1–61.
[55] J. M. Delort, D. Y. Fang, R. Y. Xue, Global existence of small solutions for quadratic
quasilinear Klein-Gordon systems in two space dimensions. J. Funct. Anal., 211
(2004), 288–323.
[56] C. Deng, J. H. Zhao, S. B. Cui, Well-posedness of a dissipative nonlinear elec-
trohydrodynamic system in modulation spaces, Nonlinear Anal., TMA, 73 (2010),
2088–2100.
[57] L. Desvillettes, About the regularization properties of the non cut-off Kac equation,
Comm. Math. Phys., 168 (1995), 417–440.
[58] L. Desvillettes, Regularization properties of the 2-dimensional non radially symmet-
ric non cutoff spatially homogeneous Boltzmann equation for Maxwellian molecules,
Trans. Theory Stat. Phys., 26 (1997), 341–357.
[59] L. Desvillettes, About the use of the Fourier transform for the Boltzmann equation,
Summer School on Methods and Models of Kinetic Theory (MMKT 2002). Riv.
Mat. Univ. Parma, 7 (2003), 1–99.
[60] L. Desvillettes and B. Wennberg, Smoothness of the solution of the spatially homo-
geneous Boltzmann equation without cutoff, Comm. Partial Differential Equations,
29 (2004), 133–155.
[61] W. Y. Ding, Y. D. Wang, Schrödinger flow of maps into symplectic manifolds,
Science in China Ser. A, 41 (1998), 746–755.
[62] W. Y. Ding and Y. D. Wang, Local Schrödinger flow into Kähler manifolds, Sci.
China Ser. A, 44 (2001), 1446–1464.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
272 Bibliography
Bibliography 273
2001.
[87] B. L. Guo, Viscosity Elimination Method and the Viscosity of Difference Scheme,
Chinese Sci. Publisher, 2004.
[88] B. L. Guo, S. J. Ding, Landau–Lifshitz equations, Frontiers of Research with the
Chinese Academy of Sciences, 1. World Scientific, 2008.
[89] B. L. Guo and B. X. Wang, The Cauchy problem for the Davey–Stewartson systems,
Comm. Pure Appl. Math., 52 (1999), 1477–1490.
[90] Z. H. Guo, Global Well-posedness of Korteweg-de Vries equation in H −3/4 (R), J.
Math. Pures Appl., 91 (2009), 583–597.
[91] Z. H. Guo, Local well-posedness and a priori bounds for the modified Benjamin–Ono
equation without using a gauge transformation, arXiv:0807.3764.
[92] Z. H. Guo, Local Well-posedness for dispersion generalized Benjamin–Ono equations
in Sobolev spaces, arXiv:0812.1825.
[93] Z. H. Guo, The Cauchy problem for a class of derivative nonlinear dispersive equa-
tions, Doctoral thesis, Peking Univ., (2009).
[94] Z. H. Guo, L. Z. Peng, B. X. Wang, Decay estimates for a class of wave equations,
J. Funct. Anal., 254 (2008), 1642–1660.
[95] Z. H. Guo, L. Z. Peng and B. X. Wang, On the local regularity of the KP-I equation
in anisotropic Sobolev space, J. Math. Pure Appl., 94 (2010), 414–432.
[96] Z. H. Guo and B. X. Wang, Global well posedness and inviscid limit for the
Korteweg–de Vries–Burgers equation, J. Differential Equations, 246 (2009), 3864–
3901.
[97] M. Hadac, Well-posedness for the Kadomtsev–Petviashvili-II equation and general-
izations, Trans. Amer. Math. Soc., 360 (2008), 6555–6572.
[98] M. Hadac, S. Herr and H. Koch, Well-posedness and scattering for the KP-II equa-
tion in a critical space, Ann. Inst. H. Poincaré, Anal. Non Linéaire, 26 (2009),
917–941.
[99] H. Hajaiej, L. Molinet, T. Ozawa, B. X. Wang, Necessary and sufficient conditions
for the fractional Gagliardo–Nirenberg inequalities and applications to the Navier–
Stokes and generalized boson equations, arXiv:1004.4287.
[100] C. C. Hao, L. Hsiao and B. X. Wang, Wellposedness for the fourth order nonlinear
Schrödinger equations, J. Math. Anal. Appl., 320 (2006), 246–265.
[101] C. C. Hao, L. Hsiao and B. X. Wang, Wellposedness of Cauchy problem for the
fourth order nonlinear Schrödinger equations in multi-dimensional spaces, J. Math.
Anal. Appl., 328 (2007), 58–83.
[102] N. Hayashi, The initial value problem for the derivative nonlinear Schrödinger equa-
tion in the energy space. Nonlinear Anal., TMA, 20 (1993), 823–833.
[103] N. Hayashi, E.I. Kaikina, P.I. Naumkin, Damped wave equation in the subcritical
case, J. Differential Equations, 207 (2004), 161–194.
[104] N. Hayashi, E.I. Kaikina, P.I. Naumkin, Damped wave equation with a critical
nonlinearity, Trans. Amer. Math. Soc., 358 (2006), 1117–1163.
[105] N. Hayashi and T. Ozawa, Finite energy solutions of nonlinear Schrödinger equa-
tions of derivative type, SIAM J. Math. Anal., 25 (1994), 1488–1503.
[106] N. Hayashi, T. Ozawa, Modified wave operators for the derivative nonlinear
Schrödinger equation. Math. Ann., 298 (1994), 557–576.
[107] N. Hayashi, T. Ozawa, Remarks on nonlinear Schrödinger equations in one space
dimension. Differential Integral Equations 7 (1994), 453–461.
[108] S. Herr, A. Ionescu, C. Kenig and H. Koch, A para-differential renormalization
technique for nonlinear dispersive equations, arXiv:0907.4649.
[109] Z. H. Huo and Y. L. Jia, The Cauchy problem for the fourth-order nonlinear
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
274 Bibliography
Bibliography 275
[131] C. E. Kenig, G. Ponce and L. Vega, Quadratic forms for the 1-D semilinear
Schrödinger equation, Trans. Amer. Math. Soc., 346 (1996), 3323–3353.
[132] C. E. Kenig, G. Ponce and L. Vega, A bilinear estimate with applications to the
KdV equation, J. Amer. Math. Soc., 9 (1996), 573–603.
[133] C. E. Kenig, G. Ponce and L. Vega, Smoothing effects and local existence theory
for the generalized nonlinear Schrödinger equations, Invent. Math., 134 (1998),
489–545.
[134] C. E. Kenig, G. Ponce and L. Vega, On the ill-posedness of some canonical dispersive
equations, Duke Math. J., 106 (2001), 617–633.
[135] C. E. Kenig, G. Ponce, L. Vega, The Cauchy problem for quasi-linear Schrödinger
equations, Invent. Math., 158 (2004), 343–388.
[136] C. E. Kenig and H. Takaoka, Global wellposedness of the modified Benjamin–Ono
equation with initial data in H 1/2 , Int. Math. Res. Not., 2006, Art. ID 95702,
1–44.
[137] R. Killip, D. Li, M. Visan and X. Y. Zhang, Characterization of minimal-mass
blowup solutions to the focusing mass-critical NLS, SIAM J. Math. Anal., 41
(2009), 219–236.
[138] R. Killip, T. Tao, and M. Visan, The cubic nonlinear Schrödinger equation in two
dimensions with radial data, J. Euro. Math. Soc., 11 (2009), 1203–1258.
[139] R. Killip, M. Visan, The focusing energy-critical nonlinear Schrödinger equation in
dimensions five and higher, Amer. J. Math. 132 (2010), no. 2, 361–424..
[140] R. Killip, M. Visan, and X. Y. Zhang, The mass-critical nonlinear Schrödinger
equation with radial data in dimensions three and higher, Anal. of PDE, 1 (2008),
229–266.
[141] N. Kishimoto, Well-posedness of the Cauchy problem for the Korteweg-de Vries
equation at the critical regularity, Preprint.
[142] N. Kishimoto, Low-regularity Local Well-posedness for Quadratic Nonlinear
Schrödinger Equations, Master Thesis, Kyoto University, 2008.
[143] S. Klainerman, Long-time behavior of solutions to nonlinear evolution equations,
Arch. Rational Mech. Anal., 78 (1982), 73–98.
[144] S. Klainerman, Global existence of small amplitude solutions to nonlinear Klein–
Gordon equations in four space-time dimensions, Comm. Pure Appl. Math., 38
(1985), 631–641.
[145] S. Klainerman and M. Machedon, Space-time estimates for null forms and the local
existence theorem, Comm. Pure Appl. Math., 46 (1993), no. 9, 1221–1268.
[146] S. Klainerman and M. Machedon, Smoothing estimates for null forms and applica-
tions, Duke Math. J., 81 (1995), 99–133.
[147] S. Klainerman, G. Ponce, Global small amplitude solutions to nonlinear evolution
equations, Commun. Pure Appl. Math., 36 (1983), 133–141.
[148] M. Kobayashi, Modulation spaces M p,q for 0 < p, q 6 ∞, J. of Funct. Spaces and
Appl., 4 (2006), no.3, 329–341.
[149] M. Kobayashi, Dual of modulation spaces, J. of Funct. Spaces and Appl., 5 (2007),
1–8.
[150] M. Kobayashi, Y. Sawano, Molecular decomposition of the modulation spaces, Os-
aka J. Math., 47 (2010), 1029–1053.
[151] H. Koch and D. Tataru, Well-posedness for the Navier–Stokes equations, Adv.
Math., 157 (2001), 22–35.
[152] B. G. Konopelchenko, B. T. Matkarimov, On the inverse scattering transform of
the Ishimori equations, Phys. Lett., 135 (1989), 183–189.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
276 Bibliography
[153] O. Ladyzhenskaya, The solvability “in the large” of the boundary value problem for
the Navier-Stokes equation in the case of two space variables, Comm. Pure Appl.
Math., 12 (1959), 427–433.
[154] O. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow, Sec-
ond English edition, Mathematics and its Applications, Vol. 2, Gordon and Breach,
Science Publishers, New York–London–Paris, 1969.
[155] P. G. Lemarié-Rieusset, Recent developments in the Navier–Stokes problem, A CRC
Press Company, 2002.
[156] S.P. Levandosky, Decay estimates for fourth order wave equations, J. Differential
Equations, 143 (1998), 360–413.
[157] F. Linares and G. Ponce, On the Davey–Stewartson systems, Ann. Inst. H.
Poincaré, Anal. NonLinéaire, 10 (1993), 523–548.
[158] H. Lindblad and C.D. Sogge, On existence and scattering with minimal regularity
for semilinear wave equations, J. Funct. Anal., 130 (1995), 357–426.
[159] J. L. Lions, Some Methods for Solving Nonlinear Boundary Value Problems, Trans-
lated by B. L. Guo and L. R. Wang, Sun Yat-sen University Press, 1992.
[160] P. L. Lions, Regularity and compactness for Boltzmann collision operator without
cut-off. C. R. Acad. Sci. Paris Series I, 326 (1998), 37–41.
[161] W. Littman, Fourier transforms of surface-carried measures and differentiability of
surface averages, Bull. of Amer. Math. Soc., 69 (1963), 766–770.
[162] S. Machihara and T. Ozawa, Interpolation inequalities in Besov spaces, Proc. Amer.
Math. Soc., 131 (2002), 1553–1556.
−1
[163] R. May, Rôle de l’space de Besov B∞,∞ dans le contrôle de l’explosion évetuelle
en temps fini des solutions réguli‘eres des équations de Navier–Stokes, C. R. Acad.
Sci. Paris, 323 (2003), 731–734.
[164] A. Miyachi, F. Nicola, S. Riveti, A. Taracco and N. Tomita, Estimates for unimodu-
lar Fourier multipliers on modulation spaces, Proc. Amer. Math. Soc., 137 (2009),
3869–3883.
[165] L. Molinet and F. Ribaud, On the low regularity of the Korteweg–de Vries–Burgers
equation, Int. Math. Res. Not., 37 (2002), 1979–2005.
[166] L. Molinet, F. Ribaud, Well-posedness results for the generalized Benjamin–Ono
equation with small initial data, J. Math. Pures Appl., 83 (2004), 277–311.
[167] L. Molinet, J.C. Saut and N. Tzvetkov, Ill-posedness issues for the Benjamin–Ono
and related equations, SIAM J. Math. Anal., 33 (2001), 982–988.
[168] L. Molinet, J. C. Saut and N. Tzvetkov, Global well-posedness for the KP-I equation,
Math. Ann., 324 (2002) 255–275, 328 (2004), 707–710.
[169] L. Molinet, J. C. Saut and N. Tzvetkov, Well-posedness and ill-posedness results
for the Kadomtsev–Petviashvili-I equation, Duke Math. J., 115 (2002), 353–384.
[170] C. S. Morawetz and W. A. Strauss, Decay and scattering of solutions of a nonlinear
relativistic wave equation, Comm. Pure Appl. Math.,25 (1972), 1–31.
[171] Y. Morimoto, S. Ukai, C. J. Xu and T. Yang, Regularity of the solution to the
spatially homogeneous Boltzmann equation without angular cutoff, Discrete Contin.
Dyn. Syst., 24 (2009), 187–212.
[172] C. Mouhot and C. Villani, On the Landau damping, arXiv:0904.2760.
[173] M. Nakamura and T. Ozawa, Nonlinear Schrödinger equations in the Sobolev space
of critical order, J. Funct. Anal., 155 (1998), 364–380
[174] M. Nakamura and T. Ozawa, The Cauchy problem for nonlinear Klein–Gordon
equations in the Sobolev spaces, Publ. Res. Inst. Math. Sci., 37 (2001), 255–293.
[175] K. Nakanishi, Energy scattering for Hartree equations, Math. Res. Lett., 6 (1999),
107–118.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Bibliography 277
[176] K. Nakanishi, Energy scattering for nonlinear Klein–Gordon and Schrödinger equa-
tions in spatial dimensions 1 and 2, J. Funct. Anal., 169 (1999), 201–225.
[177] K. Nakanishi, Scattering theory for nonlinear Klein–Gordon equation with Sobolev
critical power, Int. Math. Res. Notices, 1999, 31–60.
[178] K. Nakanishi, Unique global existence and asymptotic behavior of solutions for wave
equations with non-coercive nonlinearity, Comm. Partial Differential Equations, 24
(1999), 185–221.
[179] K. Nakanishi, Remarks on the energy scattering for nonlinear Klein–Gordon and
Schrödinger equations, Tohoku Math. J., 53 (2001), 285–303.
[180] K. Nakanishi, H. Takaoka and Y. Tsutsumi, Counterexamples to bilinear estimates
related to the KdV equation and the nonlinear Schrödinger equation, Methods of
Appl. Anal., 8 (2001) 569–578.
[181] L. Nirenberg, On elliptic partial differential equations, Ann. Sc. Norm. Sup. Pisa,
Ser. III, 13, (1959), 115–162.
[182] H. Ono, Algebraic solitary waves in stratified fluids, J. Phys. Soc. Japan, 39 (1975)
1082–1091.
[183] F. Oru, Rôle des oscillations dans quelques probl‘emes d’nalyse non-linéaire, Doc-
torat de Ecole Normale Supérieure de Cachan, 1998.
[184] T. Ozawa, On the nonlinear Schrödinger equations of derivative type, Indiana Univ.
Math. J., 45 (1996), 137–163.
[185] T. Ozawa, K. Tsutaya, Y. Tsutsumi, Global existence and asymptotic behavior of
solutions for the Klein–Gordon equations with quadratic nonlinearity in two space
dimensions, Math. Z., 222 (1996), 341–362.
[186] T. Ozawa and J. Zhai, Global existence of small classical solutions to nonlinear
Schrödinger equations, Ann. I. H. Poincaré, Anal. Non Linéaire, 25 (2008), 303–
311.
[187] M. Paicu, Équation anisotrope de Navier–Stokes dans des espaces critiques, Rev.
Mat. Iberoamericana, 21 (2005), 179–235.
[188] Y. P. Pao, Boltzmann collision operator with inverse power intermolecular poten-
tial,, I, II. Commun. Pure Appl. Math., 27 (1974), 407–428, 559–581.
[189] H. Pecher, Low energy scattering for nonlinear Klein–Gordon equations, J. Funct.
Anal., 63 (1985), 101–122.
[190] H. Pecher, Lp -Abschätzungen und klassische Lösungen für nichtlineare Wellengle-
ichungen. I, Math. Z., 150 (1976), 159–183.
[191] H. Pecher, Nonlinear small data scattering for the wave and Klein–Gordon equa-
tions, Math.Z., 185 (1984), 261–270.
[192] J. Peetre, Applications de la théorie des espaces d’interpolation dans l’analyse har-
monique, Ricerche Mat., 15 (1966), 1–34.
[193] J. Rauch, Partial Differential Equations, Graduate Texts in Math., Springer–Verlag,
1991.
[194] J. Rauch and M. Reed, Nonlinear microlocal analysis of semilinear hyperbolic sys-
tems in one spatial dimension, Duke Math. J., 49 (1982), 397–475.
[195] M. Ruzhansky, J. Smith, Dispersive and Strichartz estimates for hyperbolic equa-
tions with constant coefficients, MSJ Memoirs, 22, Mathematical Society of Japan,
Tokyo, 2010.
[196] E. Ryckman, M. Visan, Global well-posedness and scattering for the defocusing
energy-critical nonlinear Schrödinger equation in R1+4 , Amer. J. Math., 129 (2007),
1–60.
[197] I.E. Segal, Dispersion for nonlinear relativistic equation II, Ann. Sc. Ec. Norm
Sci., 4 (1968), 459–497.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
278 Bibliography
[198] I.E. Segal, Space-time decay for solutions of wave equations, Adv. Math., 22 (1976),
304–311.
[199] J. Shatah, Global existence of small classical solutions to nonlinear evolution equa-
tions, J. Differential Equations, 46 (1982), 409–423.
[200] J. Shatah, Normal forms and quadratic nonlinear Klein–Gordon equations, Comm.
Pure Appl. Math., 38 (1985), 685–696.
[201] H. F. Smith, C. D. Sogge, Global Strichartz estimates for nontrapping perturbations
of Laplacian, Comm. Partial Differential Equations, 25 (2000), 2171–2183.
[202] E. M. Stein, Singular Integrals and Differentiability Properties of Functions, Prince-
ton Univ. Press, Princeton, New Jersey, 1970.
[203] E. M. Stein, Harmonic Analysis, Princeton Univ. Press, Princeton, New Jersey,
1992.
[204] W. Strauss, Nonlinear scattering theory at low energy, J. Funct. Anal., 42 (1981),
110–133 and 43 (1981), 281–293.
[205] W. Strauss, Nonlinear Wave Equations, Lecture Notes, Vol. 73, Amer. Math.
Soc., Providence, RI. 1989.
[206] R. S. Strichartz, Restriction of Fourier transform to quadratic surfaces and decay
of solutions of wave equations, Duke Math. J., 44 (1977), 705–714.
0
[207] M. Sugimoto, Lp –Lp -estimates for hyperbolic equations, Math. Res. Lett., 2
(1995), 171–178.
[208] M. Sugimoto, Estimates for hyperbolic equations with non-convex characteristics,
Math. Z., 222 (1996), 521–531.
[209] M. Sugimoto, M. Ruzhansky, A smoothing property of Schrödinger equations and
a global existence result for derivative nonlinear equations, Advances in analysis,
315–320, World Sci. Publ., Hackensack, NJ, 2005.
[210] M. Sugimoto amd N. Tomita, The dilation property of modulation spaces and their
inclusion relation with Besov spaces, J. Funct. Anal., 248 (2007), 79–106.
[211] C. Sulem and P.L. Sulem, The Nonlinear Schrödinger Equation: Self-Focusing and
Wave Collaps, Appl. Math. Sci., 139, Springer–Verlag, 1999.
[212] H. Takaoka, Well-posedness for the Kadomtsev–Petviashvili-II equation, Adv. Dif-
ferential Equations, 5 (2000), 1421–1443.
[213] H. Takaoka, Well-posedness for the one-dimensional nonlinear Schrödinger equation
with the derivative nonlinearity, Adv. Differential Equations, 4 (1999), 561–580.
[214] H. Takaoka and N. Tzvetkov, On the local regularity of the Kadomtsev–Petviashvili-
II equation, Int. Math. Res. Not., 2001 (2001) 77–114.
[215] T. Tao, Spherically averaged endpoint Strichartz estimates for the two-dimensional
Schrödinger equation, Comm. Partial Differential Equations, 25 (2000), 1471–1485.
[216] T. Tao, Multilinear weighted convolution of L2 -functions, and applications to non-
linear dispersive equations, Amer. J. Math., 123 (2001), 839–908.
[217] T. Tao, Global well-posedness and scattering for the higher-dimensional energy-
critical non-linear Schrödinger equation for radial data, New York J. Math., 11
(2005), 57–80.
[218] T. Tao, Nonlinear dispersive equations, volume 106 of CBMS Regional Conference
Series in Mathematics, Published for the Conference Board of the Mathematical
Sciences, Washington, DC, 2006.
[219] T. Tao, Scattering for the quantic generalized Korteweg-de Vries equation, J. Dif-
ferential Equations, 232 (2007), 623–651.
[220] T. Tao, M. Visan and X. Y. Zhang, The nonlinear Schrödinger equation with com-
bined power-type nonlinearities, Comm. Partial Differential Equations, 32 (2007),
1281–1343.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Bibliography 279
[221] D. Tataru, Local and global results for wave maps I, Comm. Partial Differential
Equations, 23 (1998), 1781–1793.
[222] J. Toft, Continuity properties for modulation spaces, with applications to pseudo-
differential calculus, I, J. Funct. Anal., 207 (2004), 399–429.
[223] P. Tomas, A restriction theorem for the Fourier transform, Bull. Amer. Math. Soc.,
81 (1975), 477–478.
[224] H. Triebel, Theory of Function Spaces, Birkhäuser–Verlag, 1983.
[225] H. Triebel, Modulation spaces on the Euclidean n-spaces, Z. Anal. Anwendungen,
2 (1983), 443–457.
[226] Y. Tsutsumi, L2 -solutions for nonlinear Schrödinger equations and nonlinear
groups. Funkcial. Ekvac., 30 (1987), 115–125.
[227] J. A. Tuszynski and J. M. Dixon, Coherent structures in strongly interacting many-
body systems: I, Derivation of dynamics, J. Phys. A: Math. Gen., 22 (1989),
4877–4894.
[228] S. Ukai, Local solutions in Gevrey classes to the nonlinear Boltzmann equation
without cutoff, Japan J. Appl. Math., 1 (1984), 141–156.
[229] C. Villani, A Review of Mathematical Topics in Collisional Kinetic Theory, Hand-
book of Fluid Mechanics, eds. S. Friedlander and D. Serre (North–Holland, 2002).
[230] C. Villani, On a new class of weak solutions to the spatially homogeneous Boltzmann
and Landau equations, Arch. Rational Mech. Anal., 143 (1998), no. 3, 273–307.
[231] C. Villani, Regularity estimates via entropy dissipation for the spatially homoge-
neous Boltmann equation, Rev. Mat. Iberoamericana, 15 (1999), 335–352.
[232] M. Visan, The defocusing energy-critical nonlinear Schrödinger equation in higher
dimensions, Duke Math. J., 138 (2007), 281–374.
[233] H. Wadade, Remarks on the Gagliardo–Nirenberg type inequality in the Besov and
the Triebel-Lizorkin spaces in the limiting case, J. Fourier Anal. Appl., 15 (2009),
857–870.
[234] B. X. Wang, The Cauchy Problem for the Nonlinear Schrödinger, Klein–Gordon
Equations and Their Coupled Equations, Doctoral Thesis, Inst. Appl. Phys. and
Comput. Math., Dec. 1993, 1–224.
[235] B. X. Wang, Bessel (Riesz) potentials on Banach function spaces and their applica-
tions I, Theory, Acta Math. Sinica (N.S.), 14 (1998), 327–340.
[236] B. X. Wang, On existence and scattering for critical and subcritical nonlinear Klein–
Gordon equations in H s , Nonlinear Analysis, TMA, 31 (1998), 173–187.
[237] B. X. Wang , On scattering of solutions for the critical and subcritical nonlinear
Klein-Gordon equations, Discrete Contin. Dyn. Syst., 5 (1999), 753–763.
[238] B. X. Wang, The limit behavior of solutions for the complex Ginzburg–Landau
equation, Commun. Pure Appl. Math., 55 (2002), no.4, 481–508.
[239] B. X. Wang, The smoothness of scattering operators for the sinh–Gordon and non-
linear Schrödinger equations, Acta Math. Sinica (Engl. Ser.), 18 (2002), 549–564.
[240] B. X. Wang, Exponential Besov spaces and their applications to certain evolution
equations with dissipation, Commun. Pure and Appl. Anal., 3 (2004), 883–919.
[241] B. X. Wang, Nonlinear scattering theory for a class of wave equations in H s , J.
Math. Anal. Appl., 296 (2004), 74–96.
[242] B. X. Wang, Concentration Phenomenon for the L2 Critical and Super Critical
Nonlinear Schrödinger Equation in Energy Spaces, Commun. Contemp. Math., 8
(2006), 309–330.
[243] B. X. Wang, L. J. Han, C. Y. Huang, Global well-Posedness and scattering for
the derivative nonlinear Schrödinger equation with small rough data, Ann. I. H.
Poincaré, Anal. Non Linéaire, 26 (2009), 2253–2281.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
280 Bibliography
[244] B. X. Wang, C. C. Hao and H. Hudzik, Energy scattering for the nonlinear
Schrödinger equations with exponential growth in lower spatial dimensions, J. Dif-
ferential Equations, 228 (2006), 311–338.
[245] B. X. Wang and C. Y. Huang, Frequency-uniform decomposition method for the
generalized BO, KdV and NLS equations, J. Differential Equations, 239 (2007),
213–250.
[246] B. X. Wang and H. Hudzik, The global Cauchy problem for the NLS and NLKG
with small rough data, J. Differential Equations, 231 (2007), 36–73.
[247] B. X. Wang, L. F. Zhao, B. L. Guo, Isometric decomposition operators, function
λ and their applications to nonlinear evolution equations, J. Funct. Anal.,
spaces Ep,q
233 (2006), 1–39.
[248] W. K. Wang, T. Yang, Point-wise estimates and Lp convergence rates to diffusion
waves for p-system with damping, J. Differential Equations 187 (2003), 310–336.
[249] W. K. Wang, W. J. Wang, The pointwise estimates of solutions for semilinear
dissipative wave equation in multi-dimensions. J. Math. Anal. Appl., 366 (2010),
226–241.
[250] F. B. Weissler, Existence and nonexistence of global solutions for a semilinear heat
equation, Israel Math. J. 39 (1981), 29–40.
[251] F. B. Weissler, The Navier–Stokes initial value problem in Lp , Arch. Rational Mech.
Anal. 74 (1980), 219–230.
[252] N. Wiener, Tauberian theorems, Ann. of Math., 33 (1932), 1–100.
[253] J. Xiao, Homothetic variant of fractional Sobolev space with application to Navier–
Stokes system, Dyn. Partial Differ. Equ. 4 (2007), 227–245.
[254] D. C. Yang, W. Yuan, A new class of function spaces connecting Triebel–Lizorkin
spaces and Q spaces, J. Funct. Anal., 255 (2008), 2760–2809.
[255] D. C. Yang, W. Yuan, New Besov-type spaces and Triebel–Lizorkin-type spaces
including Q spaces, Math. Z., 265 (2010), 451–480.
[256] K. Yosida, Functional Analysis, Springer–Verlag, 1981.
[257] W. Yuan, W. Sickel, D. C. Yang Morrey and Campanato Meet Besov, Lizorkin and
Triebel, Lect. Note in Math., 2005 (2010), 1-281.
[258] V. E. Zakharov, E. A. Kuznetson, Multi-scale expansions in the theory of systems
integrable by inverse scattering method, Physica D, 18 (1986), 455–463.
[259] V. E. Zakharov, E. I. Schulman, Degenerated dispersion laws, motion invariant and
kinetic equations, Physica D, 1 (1980), 185–250.
[260] Y. L. Zhou, B. L. Guo, Weak solution of system of ferromagnetic chain with several
variables. Sci. Sinica Ser. A, 30 (1987), 1251–1266.
[261] Y. L. Zhou, B. L. Guo and S. Tan, Existence and uniqueness of smooth solution for
system of ferro-magnetic chain. Sci. China Ser. A 34 (1991), 257–266.
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
Index
s
Bp,q ,9 Boltzmann equation
C (Rn ), 2
∞
Global (absolute) Maxwellian, 238
s
E2,1 , 45 Bobylev’s identities , 238
s
Fp,q , 9 Boltzmann collision operator, 233
Lp (`q ), 281 Boltzmann equation, 233
Lpx1i Lp(x2j )j6=i Lpt 2 , 189 Cross section, 233
s
Mp,q , 160 Cutoff cross section, 234
s
Xp,q , 10 Hard potentials, 235
k , 160 Homogeneous Boltzmann equation,
S, 2 233
S 0, 2 Local Maxwellian, 238
s
Ḃp,q , 20 Maxwellian potentials, 235
s
Ḟp,q , 20 Non cutoff cross section, 234
s Soft potentials, 235
Ẋp,q , 20
S˙ , 20
S˙ 0 , 20 Christ-Kiselev lemma, 267
`q (Lp ), 9 Complex interpolation in modulation
`s,q γ p spaces, 267
(L (R, L )), 179
φ ∗ ψ, 3 Convexity Hölder’s inequality, 21
φ̃, 3 Critical power in Ḣ s for NLS, 83
4k , 9 Critical space Ḣ s for NLS, 83
e k , 192
Critical space for NS equation, 35
{σk }k∈Zn , 160
p ∧ q, p ∨ q, 11 Dilation operator, 3
Dispersion, 51
Action functional, 206 Dispersion relation, 92
Affine transform, 7 Dyadic decomposition operator, 9
Almost orthogonality of k , 198
Asymptotic completeness, 261 Embedding theorem
Asymptotic state, 261 Embedding between Besov and
modulation spaces, 164
Bernstein’s inequality, 14 Embedding in homogeneous
281
June 15, 2011 10:2 World Scientific Book - 9in x 6in booken-main
282 Index
Index 283