Inviscid limit for the energy-critical complex

Ginzburg-Landau equation
Chunyan Huang
1
, Baoxiang Wang
2
LMAM, School of Mathematical Sciences, Peking University,
Beijing 100871, Peoples Republic of China.
Abstract. In this paper, we consider the limit behavior for the solution of the Cauchy
problem of the energy-critical complex Ginzburg-Landau equation in R
n
, n 3. In
lower dimension case (3 n 6), we show that its solution converges to that of the
energy-critical nonlinear Schrodinger equation in C(0, T,

H
s
(R
n
)), T > 0, s = 0, 1, as
a by-product, we get the regularity of solutions in H
3
for the nonlinear Schrodinger
equation. In higher dimension case (n > 6), we get the similar convergent behavior in
C(0, T, L
2
(R
n
)). In both cases we obtain the optimal convergent rate.
Key words and phrases. Complex Ginzburg-Landau equation; nonlinear Schodinger
equation; inviscid limit; energy-critical power.
2000 Mathematics Subject Classications. 35K 55, 35Q 55.
1 Introduction
We are interested in the inviscid limit behavior between the energy-critical complex
Ginzburg-Landau equation(CGL)
u
t
( + i)u + (a + i)[u[

u = 0, u(0, x) = u
0
(x), (1.1)
and the nonlinear Schrodinger equation (NLS)
v
t
iv + i[v[

v = 0, v(0, x) = v
0
(x). (1.2)
where u(t, x) and v(t, x) are complex valued functions of (t, x) R
+
R
n
, R
+
=
[0, ), i =

1, > 0, a > 0, u
t
= u/t, v
t
= v/t, =
2
/
2
x
1
+... +
2
/
2
x
n
,
u
0
and v
0
are known complex valued functions of x R
n
. In Eqs. (1.1) and (1.2),
0 < < 4/(n 2) (0 < < , if n = 1, 2) is said to be an energysubcritical
power, = 4/(n 2) is said to be an energycritical power. Eq. (1.1) was rst
discovered by Ginzburg and Landau for a phase transition in superconductivity
[8], which was also subsequently derived in many other areas, such as instability
1
Email: hcy@math.pku.edu.cn
2
Email: wbx@math.pku.edu.cn
1
waves in hydrodynamics and pattern formation; cf. [10]. When is an energy
subcritical or an energycritical power, the global well-posedness of (1.1) has been
established, see for instance, Ginibre and Velo [7] and references therein. Eq. (1.2)
is a basic model equation for diverse physical phenomena, including Bose-Einstein
condensates, description of the envelop dynamics of a general dispersive wave in a
weakly nonlinear medium (cf. [11, 16]). The global well-posedness of solutions of
(1.2) has been studied by many authors (cf. [3, 4, 5, 13]). When is an energy
subcritical power, the global well-posedness of (1.2) can be found in Kato [12]. When
is an energycritical power, Bourgain [3] and Grillakis [9] showed the global well-
posedness and the existence of scattering operators for Eq. (1.2) with radial and
nite energy data in three spatial dimensions, and the radial condition was removed
by Colliander, Keel, Stalani, Takaoka and Tao [5]. Recently, Ryckman and Visan
(cf. [15, 18]) have generalized the global well-posedness to higher spatial dimensions.
In this paper, we consider the following question: Does the solution of (1.1)
tend to the solution of (1.2) as the parameters and a tend to zero? This problem
has been studied by several authors; cf. Wu [22], Bechouche and J ungel [1], Wang
[19] and Machihara and Nakamura [14]. But as far as the authors can see, in the
energycritical case = 4/(n 2), except that Bechouche and J ungel [1] obtained
the convergent behavior in the weak-star topology, the inviscid limit behavior in
the strong topology between (1.1) and (1.2) remains unsolved; cf. [19]. When
= 4/(n 2), (1.1) and (1.2) can be rewritten as
u
t
( + i)u + (a + i)[u[
4
n2
u = 0, u(0, x) = u
0
(x), (1.3)
v
t
iv + i[v[
4
n2
v = 0, v(0, x) = v
0
(x). (1.4)
We will show that when the spatial dimension n 3, the solution of (1.3) strongly
converges to that of (1.4) in C(0, T; L
2
) for any T > 0. Moreover, when 3 n 6,
we can prove the stronger convergent behavior in C(0, T;

H
1
) for any T > 0. Let
L
r
(R
n
) be the Lebesgue space, |f|
r
:= |f|
L
r
(R
n
)
, |f|

H
k
r
:= |
k
f|
r
,

H
k
:=

H
k
2
,
H
k
:= L
2

H
k
. Put
E(u(t)) =

R
n
1
2
[u(t, x)[
2
+
n 2
2n
[u(t, x)[
2n
n2
dx, (1.5)
which is said to be the energy for the CGL (1.3) and the NLS (1.4). E(u) is invariant
under the scaling
u(t, x) u

(t, x) =
n2
2
u(
2
t, x), > 0. (1.6)
Our main results are the following:
2
Theorem 1.1 Let 3 n 6. Let u
0
H
1
, v
0
H
3
, 0 < < 1. Assume that
T > 1 is arbitrary and
|u
0
v
0
|
H
1 , 0 < T, a . (1.7)
Then the solutions of (1.3) and (1.4) satisfy the following approximate behavior:
|
k
(u v)|
L

t
L
2
x
([0,T]R
n
)
C(|
k
(u
0
v
0
)|
2
+ T + a), k = 0, 1. (1.8)
where C := C(|u
0
|
H
1, |v
0
|
H
3) denotes a constant that depends only on |u
0
|
H
1 and
|v
0
|
H
3.
Theorem 1.2 Let n > 6. Let u
0
H
1
, v
0
H
1
, 0 < <1. Assume that T > 1 is
arbitrary and
|u
0
v
0
|
H
1 , 0 < aT, T . (1.9)
Then the solutions of (1.3) and (1.4) satisfy the following approximate behavior:
|u v|
L

t
L
2
x
([0,T]R
n
)
C(|u
0
v
0
|
2
+ (T)
1/2
+ a); (1.10)
where C := C(n, |u
0
|
H
1, |v
0
|
H
1) denotes a constant that depends only on n, |u
0
|
H
1
and |v
0
|
H
1.
Theorem 1.3 Let n > 6. u
0
H
1
, 0 < <1, 0 < a, 1. Then the solutions u
of (1.3) have the following upper bound estimate:
|u|
L
2(n+2)
n2
t,x
(R
+
R
n
)
C(E(u
0
)); (1.11)
where C(E(u
0
)) denotes a constant that depends only on E(u
0
) and is independent
of , a.
We now rst briey sketch the proofs of Theorems 1.1 and 1.2. On the basis of
our earlier work [19], if we get the upper bounds of the solutions u and v of Eqs. (1.3)
and (1.4) in the spaces L
2(n+2)/(n2)
(R
+
R
n
), then we can prove our Theorems 1.1
and 1.2 by iteration method. Using Bourgains, Colliander, Keel, Stalani, Takaoka
and Taos, and Ryckman and Visans results (cf. [3, 5, 15, 18]), we see that v in the
space L
2(n+2)/(n2)
(R
+
R
n
) has an upper bound that depends only on |v
0
|

H
1
. A
natural idea seems to apply a similar way as in [5, 15, 18] obtaining a uniform upper
bound of u in L
2(n+2)/(n2)
(R
+
R
n
) for all 0 < , a <1. However, in [5, 15, 18], the
3
proofs of the global well-posedness of (1.4) rely upon the symmetric property of the
Schrodinger semi-group e
it
(i.e., e
it
has the same dispersive properties in the cases
t < 0 and t > 0) and this symmetric property is invalid for the Ginzburg-Landau
semi-group e
(+i)t
. Moreover, the proof will become very complicated if we imitate
the procedures as in [5, 15, 18].
When 3 n 6, we will give a simple proof of Theorem 1.1 by using the
perturbation analysis method, i.e., treating the NLS as a perturbative equation of
the CGL:
w
t
( + i)w + g(v, w) = 0,
where w = u v, g(v, w) = v + (a + i)([w + v[
4
n2
(w + v) [v[
4
n2
v) + a[v[
4
n2
v.
Hence, it suces to show that w is uniformly bounded in L
2(n+2)/(n2)
(R
+
R
n
) for
all 0 < , a <1 as the initial data w
0
is small enough. Following the idea as in [19],
we treat v as a part of the nonlinear terms. We consider the following integral
equation:
w = e
(+i)t
w
0

t
0
e
(+i)(t)
g(v, w)()d.
In view of the Strichartz type estimates for the Ginzburg-Landau semi-group (which
are similar to those for the Schrodinger semi-group), in order to control w in the
space L
2(n+2)/(n2)
(R
+
R
n
), we need to control v in H
3
. For n = 3, 4, the upper
bound of v in H
3
can be easily obtained, since the nonlinearity [v[
4/(n2)
v is a C

function (cf. [3, 19]). For n = 5, 6, [v[

4/(n2)
v is in C
2
and not in C
3
, we need
to use the formal time-dierentiation method to overcome the regularity loss of
the nonlinearity. After showing the upper bound of v in H
3
, we can perform the
perturbation analysis and nally obtain our result, as desired. For higher spatial
dimensions n > 6, since [v[
4/(n2)
v is not in C
2
, the formal time-dierentiation
method seems invalid to get the upper bound of v in H
3
. This is why we assume
that n 6 in Theorem 1.1.
When n > 6, we shall bound u in L
2(n+2)/(n2)
(R
+
R
n
) by applying the ideas
as in Bourgain [3], Colliander, Keel, Stalani, Takaoka and Tao [5], and Visan [18],
i.e., the induction method on the energy E. In detail, for any energy E 0 and
T > 0, dene
M
T
(E) := sup |u|
L
2(n+2)
n2
t,x
([0,T]R
n
)
,
4
where u ranges over all of the Schwartz solutions
3
of (1.3) on [0, T] R
n
with
E(u) E and the parameters 0 , a 1. Put
M(E) := supM
T
(E) : T > 0.
Noticing that both E(u) and M
T
(E) are invariant under the scaling (1.6), we im-
mediately have
Proposition 1.4 For any E, T > 0, we have
M(E) = M
T
(E) M
1
(E). (1.12)
For E < 0, dene M(E) = 0. Our goal is to show M(E) < for any energy E.
To be more comprehensive, we write
Q = E : M(E) < .
The perturbation analysis implies that if E Q then E + Q for any 0 < <1,
and so, Q is open, see Lemma 2.9 below for details. Moreover, Q contains zero and
is connected. We use the induction-on-energy argument as in [3, 5, 18]. Assume for
contradiction that M(E) can be innite, consider the minimal energy
E
c
= infE : M(E) = < ,
then naturally M(E) < for all E < E
c
. Since Q is open and contains zero, then
E
c
> 0. Using this critical energy E
c
, we can construct a near blowup solution whose
energy is near to E
c
with an enormous L
2(n+2)
n2
t,x
norm. We will prove that such near
blowup solution doesnt exist, then eventually we get a contradiction. In detail, by
the critical property of E
c
and Lemma 2.9, we have
Lemma 1.5 (Induction hypothesis) Let 0 < , a 1 be variable parameters. Let
t
0
R
+
, suppose u(t
0
) is an

H
1
function with E(u(t
0
)) E
c
for some > 0,
then there exists a global solution u to (1.3) on [t
0
, +) R
n
with initial data u(t
0
)
at time t
0
satisfying
|u|
L
2(n+2)
n2
t,x
([t
0
,+)R
n
)
M(E
c
). (1.13)
3
The global existence of the Schwartz solutions of (1.3), which correspond to the initial data
u
0
in the Schwartz space, has been shown in [7].
5
We need eight small parameters
1
0

1

2

3

4

5

6

7
> 0
where each
j
is chosen suciently small depending on previous
0
, . . . ,
j1
and
the critical energy E
c
. Since M(E
c
) is innite, we can nd a Schwartz solution u of
(1.3) with E
c
/2 E(u) E
c
such that for some T > 0, , a (0, 1],
|u|
L
2(n+2)
n2
t,x
([0,T]R
n
)
=
1

7
, (1.14)
which leads to the following denition:
Denition 1.6 A minimal energy blowup solution (MEBS for short) to (1.3) is a
Schwartz solution which satises (1.14) and with energy
1
2
E
c
E(u) E
c
. (1.15)
Using the energy estimate for the CGL, we claim that for the MEBS, both
and a must be very small. In fact, we have
Proposition 1.7 If u is a MEBS, then we have max(, a)
6
.
Proof. By the energy estimate for the CGL (see Theorem 2.7 below), we have

T
0

R
n
[u(, x)[
2
dxd + a

T
0

R
n
[u(, x)[
2(n+2)
n2
dxd E(u
0
).
If a
6
, we see that
|u|
L
2(n+2)
n2
t,x
([0,T]R
n
)

E(u
0
)

(n2)/2(n+2)
,
which contradicts (1.14). So, it suces to consider the case
6
. Interpolating
L
2(n+2)
n2
t,x
between L

t
L
2n
n2
x
and L
2
t
L
2n
n4
x
, we have
|u|
L
2(n+2)
n2
t,x
([0,T]R
n
)
|u|
4/(n+2)
L

t
L
2n
n2
x
|u|
(n2)/(n+2)
L
2
t
L
2n
n4
x
E(u
0
)
4/(n+2)
|u|
(n2)/(n+2)
L
2
t,x
E(u
0
)
4/(n+2)

E(u
0
)

(n2)/2(n+2)
,
which also contradicts (1.14).
6
In the following we always assume that
0 , a
6
.
By choosing a subinterval [0, T

] [0, T], such that

|u|
L
2(n+2)
n2
t,x
([0,T

]R
n
)
=
1

5
. (1.16)
We divide
[0, T

] = I
0
I
1
(1.17)
with each subinterval containing half of the L
2(n+2)
n2
t,x
([0, T

] R
n
)) norm. We want
to show that
|u|
L
2(n+2)
n2
t,x
(I
0
R
n
)
C(
1
, ,
4
, E(u
0
)). (1.18)
Following [5, 18], the proof of (1.18) proceeds in the following four steps. (1)
Using the same way as in [5, 18], we can show that the MEBS is localized in physical
and frequency space for any time t I
0
. (2) We generalize the frequency localized
Morawetz estimate (FLME) of the NLS to the CGL and get a uniform version of
the FLME which is independent of , a. Our idea is to apply the known Morawetz
estimate for the NLS [5, 18] and treat u and a[u[
4
n2
u as the perturbative terms,
which can be controlled by the energy estimates for the CGL. (3) We show the
frequency-localized L
2
mass conservation law to prevent energy evacuation to the
high frequency. Comparing with the NLS, the new diculty lies in the fact that the
L
2
norm of the solutions of the CGL has some dissipation (see (2.13) for details) and
we need to show that this dissipation vanishes when , a ` 0. Another diculty
is that the Ginzburg-Landau semi-group e
(+i)t
is not symmetric with respect to
time t, one needs to carefully deal with the backward case t < 0. (4) Following [5]
and using the perturbation analysis, we can nish the proof of (1.18).
2 Notations and known results
In this section, we recall some notations, basic facts and the Strichartz estimates for
the NLS and the CGL.
7
2.1 Notations
Let I R be an interval. We use L
q
t
L
r
x
to denote the space-time norm:
|u|
L
q
t
L
r
x
(IR
n
)
:=

R
n
[u(t, x)[
r
dx

q/r
dt

1/q
(2.1)
with the usual modication when q or r is equal to innity. We denote |u|
L
q
(I,

H
k
r
)
:=
|
k
u|
L
q
t
L
r
x
(IR
n
)
. If there is no confusion, we will write L
q
tI
L
r
x
:= L
q
t
L
r
x
(I R
n
) and
L
r
tI,x
:= L
r
t
L
r
x
(I R
n
).
We introduce the Littlewood-Paley projection operators in following way: Let
() be a radially symmetric bump function adapted to the ball R
n
: [[ 2
which equals 1 on the unit ball. Let () := () (2). For each dyadic number
N, we dene the Fourier multipliers:
P
N
f := T
1
(/N)Tf,
P
>N
f := T
1
(1 (/N))Tf,
P
N
f := T
1
(/N)Tf.
Recall that for any dyadic number N, there hold the telescoping identities
P
N
f =

MN
P
M
f; P
>N
f =

M>N
P
M
f; f =

M
P
M
f,
for all Schwartz f, where M ranges over dyadic numbers. Dene
P
M<N
:= P
N
P
M
=

M<N

N
P
N
,
where N

, M N are all dyadic numbers. Similarly we dene P

<N
, P
N
and
P
MN
, etc. It is easy to verify that the Littlewood-Paley operators satisfy the
following Bernstein inequalities for s 0, 1 p q :
|P
N
f|
L
p N
s
|[[
s
P
N
f|
L
p, (2.2)
|P
N
[[
s
f|
L
q N
s+n(
1
p

1
q
)
|P
N
f|
L
p, (2.3)
|P
N
[[
s
f|
L
p N
s
|P
N
f|
L
p. (2.4)
We dene the

S
0
(I R
n
) Strichartz norm by
|u|

S
0
(IR
n
)
:=

|u|
L

t
L
2
x
L
2
t
L
2n
n2
x
(IR
n
)
, n 6;

N2
Z
|P
N
u|
2
L

t
L
2
x
L
2
t
L
2n
n2
x
(IR
n
)

1/2
, n > 6.
8
and for any s > 0, we dene the Strichartz norm

S
s
(I R
n
) by
|u|

S
s
(IR
n
)
:= |[[
s
u|

S
0
(IR
n
)
.
One can easily check that for any admissible pairs (q, r),
|
k
u|
L
q
t
L
r
x
(IR
n
)
|u|

S
k
(IR
n
)
, k = 0, 1.
2.2 Strichartz Inequalities
Denote
2

= 2 +
4
n
, 2

=
2n
n 2
, n 3.
Throughout this paper, p

will stand for the dual number of p [1, ], i.e. 1/p +

1/p

= 1. Let us follow Cazenave and Weisslers notation (cf. [4]):

Denition 2.1 Let n 3. (q, r) is said to be an admissible pair if 2/q = n(1/2
1/r), 2 q, r 2

. We say that (q

, r

) is a dual admissible pair if (q, r) is an

admissible pair.
From the denition above, we can easily nd that (, 2), (2

, 2

), (2, 2

) and
(q, r) =

2(n + 2)
n 2
,
2n(n + 2)
n
2
+ 4

are admissible pairs and (1, 2), ((2

, (2

), (2, (2

) are dual admissible pairs.

We denote
S(t) := e
it
= F
1
e
it||
2
F, A :=

t
0
S(t ) d,
S

(t) := e
(+i)t
= F
1
e
(+i)t||
2
F, A

:=

t
0
S

(t ) d.
Eqs. (1.3) and (1.4) can be formally rewritten as
u(t) = S

(t)u
0
+A

(u
t
( + i)u), (2.5)
v(t) = S(t)v
0
+A(v
t
iv). (2.6)
We use X Y to denote the estimate X CY , C is a universal positive constant.
It is well known that S(t) satises the decay estimate
|S(t)|
p
[t[
n(1/21/p)
||
p
, 1/p + 1/p

= 1, 2 p . (2.7)
9
Since e
||
2
is an L
p
multiplier, i.e., e
||
2
M
p
, M
p
is Homanders multiplier space,
and M
p
is isometrically invariant under surjective ane transformation, we have
|e
||
2
|
M
p
= |e
t||
2
|
M
p
(cf. [2]). It follows from (2.5) that for any 2 p < and
t > 0,
|S

(t)|
p
C|S(t)|
p
Ct
n(1/21/p)
||
p
, (2.8)
where C is independent of and t, L
p

, 1/p + 1/p

= 1. Combining (2.8) with

the standard Strichartz estimates for the Schrodinger equation, one can prove the
following time-space L
p
L
p

estimates. We omit the details here, one can see Wang

[19] for nonendpoint case and [20] for endpoint case.
Lemma 2.2 Let n 3, k N 0. Let (q, r) be any admissible pair and (q

1
, r

1
)
be any dual admissible pair. Then we have
|S

(t)|
L
q
(0,;

H
k
r
)
||

H
k
, (2.9)
|A

f|
L
q
(0,T;

H
k
r
)
|f|
L
q

1(0,T;

H
k
r

1
)
, (2.10)
for all

H
k
, f L
q

1
(0, T;

H
k
r

1
), 0 < T . When = 0, it reduces to the
standard Strichartz estimates for the Schrodinger equation (see, [13]).
We emphasize that the estimates (2.9) and (2.10) are independent of > 0. Using
Lemma 2.2 together with Duhamels formula (2.5) and the triangle inequality, we
can verify the following multi-linear form of the Strichartz estimates:
Corollary 2.3 Let I be a time interval, k = 0, 1, and let u : I R
n
C be a
solution to
u
t
( + i)u =
M

m=1
F
m
for some functions F
1
, . . . , F
M
, where 0, then for any admissible pair (q, r),
|
k
u|
L
q
t
L
r
x
(IR
n
)
C(|u(t
0
)|

H
k
(R
n
)
+
M

m=1
|
k
F
m
|
L
q

m
t
L
r

m
x
(IR
n
)
), (2.11)
where t
0
= min I, (q

1
, r

1
), . . . (q

m
, r

m
) are dual admissible pairs.
We need the following improved Strichartz estimate for the nonhomogeneous term,
which can be proved by following the method in [6] and the Strichartz estimates for
the CGL.
10
Proposition 2.4 Let I [0, +) be a compact time interval. Let (q, r) and ( q, r)
be two Schr odinger-acceptable pairs satisfying the scaling condition
1
q
+
1
q
=
n
2
(1
1
r

1
r
) and either
1
q
+
1
q
= 1,
n 2
n
<
r
r
<
n
n 2
,
1
r

1
q
, and
1
r

1
q
or
1
q
+
1
q
< 1,
n 2
n

r
r

n
n 2
.
Then

s<t
e
(+i)(ts)
F(s)ds

L
q
t
L
r
x
(IR
n
)
|F|
L
q

t
L
r

x
(IR
n
)
.
Here we say (q, r) Schr odinger-acceptable means that
1
q
< n(
1
2

1
r
), 1 q, r ,
or (q, r) = (, 2).
Remark 2.5 We point out here that the pairs (q, r) = (2,
2n(n2)
n
2
3n2
), ( q, r) = (2,
2n(n2)
n
2
5n+10
)
satisfying the hypothesis in Proposition 2.4.
Remark 2.6 When we prove a property of the CGL, if we only use the (improved)
Strichartz estimates and solve the integral equation (2.5) forward, then we can use
an analogous way to the NLS, for the CGL satises the same (improved) Strichartz
estimates as the NLS. This fact is very important for our later discussion.
2.3 Global well-posedness
Theorem 2.7 (Global well-posedness for CGL) (cf. [7, 19]) Let , a 0 with
0 < max(, a) < , u
0


H
1
. Then there exists a unique global solution u with
u C
t
L
2
x
L
2
t
L
2n
n2
x
to (1.3) such that
E(u(t)) +

t
0

R
n
[u(, x)[
2
dxd + a

t
0

R
n
[u(, x)[
2(n+2)
n2
dxd E(u
0
). (2.12)
Moreover, if u
0
L
2
, then u C
t
L
2
x
L
2
t
L
2n
n2
x
and
1
2
|u(t)|
2
2
+

t
0
|u()|
2
2
d + a

t
0
|u()|
2n
n2
2n
n2
d =
1
2
|u
0
|
2
2
. (2.13)
In [7], the upper bound of |u|
L
2(n+2)
n2
t,x
(R
+
R
n
)
in Theorem 2.7 relies upon max(, a)
and it tends to innity as max(, a) 0 (which can be seen from the proof of
Proposition 1.7).
11
We also need the following result which can be found in Colliander, Keel, Sta-
lani, Takaoka and T. Tao [5] for n = 3, and Ryckman and Visan (cf. [15, 18]) for
n 4.
Theorem 2.8 (Global well-posedness for NLS) Let n 3. If E(v
0
) < , then
there exists a unique global solution v C
t

H
1
x
L
2(n+2)
n2
t,x
to (1.4) such that
|v(t, x)|
L
2(n+2)
n2
t,x


S
1
(RR
n
)
C(E(v
0
))
for some constant C(E(v
0
)) depending only on the energy.
2.4 Perturbation lemma
We need the following perturbation lemma which is useful in the proof of the main
Theorem 1.3.
Lemma 2.9 (Long-time perturbation theory) Let n > 6. Let I [0, +) be a
compact time interval and let u be an approximate solution to (1.3) on I R
n
in
the sense that
i u
t
+ (1 i) u = (1 ai)[ u[
4
n2
u + e,
for some function e. Assume that
| u|
L
2(n+2)
n2
t,x
(IR
n
)
M
| u|
L

H
1
x
(IR
n
)
E
for some constants M, E > 0. Let t
0
= min I and let u(t
0
) close to u(t
0
) in the sense
that
|u(t
0
) u(t
0
)|

H
1
(R
n
)
E

(2.14)
for some E

> 0. Assume also the smallness conditions

(

N
|P
N
e
(+i)(tt
0
)
(u(t
0
) u(t
0
))|
2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(IR
n
)
)
1
2
, (2.15)
|e|
L
2
t
L
2n
n+2
x
(IR
n
)
, (2.16)
for some 0 < <
1
, where
1
=
1
(E, E

, M) > 0 is a small constant. Then there

exists a solution u to (1.3) on I R
n
with initial data u(t
0
) satisfying
|(u u)|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(IR
n
)
C(E, E

, M)( +
7
(n2)
2
) (2.17)
12
|u u|

S
1
(IR
n
)
C(E, E

, M)(E

+ +
7
(n2)
2
) (2.18)
|u|

S
1
(IR
n
)
C(E, E

, M). (2.19)
Here C(E, E

, M) > 0 is a non-decreasing function of E, E

, M, and the dimension

n.
Remark 2.10 Recall that the Ginzburg-Landau semi-group S

(t) satises the same

Strichartz estimates as the corresponding Schrodinger semi-group S(t). Noticing
that t
0
= min I, we only need to consider the CGL forward in Lemma 2.9, then the
proof is parallel to that of the NLS (see Tao and Visan [17]) and we omit the details
of the proof.
3 Proof of Theorem 1.1
In this section, we prove the main Theorem 1.1. We will use the perturbation
analysis to give a short proof of Theorem 1.1 when 3 n 6.
3.1 Nonlinear estimates
The purpose of this subsection is to derive some nonlinear estimates, which will be
applied in the next subsection. Write F(z) = [z[
4
n2
z, let F
z
, F
z
be the complex
derivatives
F
z
:=
1
2

F
x
i
F
y

, F
z
:=
1
2

F
x
+ i
F
y

,
then we see that
F
z
(z) =
n
n 2
[z[
4
n2
, F
z
(z) =
2
n 2
[z[
4
n2
z
2
[z[
2
, (3.1)
which implies that F
z
(z) and F
z
(z) are both O([z[
4
n2
). Note that the dierence of
two nonlinear terms satises the integral identity:
F(u) F(v) =

1
0
F
z
(v + (u v))(u v) + F
z
(v + (u v))(u v)d. (3.2)
Hence,
[F(u) F(v)[ [u v[([u[
4
n2
+[v[
4
n2
). (3.3)
Observe that F(u) obeys the fractional chain rule:
F(u(x)) = F
z
(u(x))u(x) + F
z
(u(x))u(x). (3.4)
By (3.1), (3.4) and Holders inequality, we can get the following estimate:
13
Proposition 3.1 For k=0,1, we have:
|
k
F(u)|
L
2
t
L
2n
n+2
x
(IR
n
)
|u|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)
|
k
u|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(IR
n
)
. (3.5)
Similarly, we can invoke Holders inequality and (3.1), (3.3),(3.4) to get:
Proposition 3.2 For k = 0, 1, we have
|
k
F(u)|
L
(2

t,x
(IR
n
)
|u|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)
|
k
u|
L
2

t,x
(IR
n
)
, (3.6)
|F(u) F(v)|
L
(2

t,x
(IR
n
)

|u|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)
+|v|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)

|u v|
L
2

t,x
(IR
n
)
. (3.7)
Now we estimate the dierence with one order derivative. Write w = u v. When
n 6, in view of the chain rule (3.4), we calculate
(F(u) F(v))
= F
z
(u)u + F
z
(u) u F
z
(v)v F
z
(v) v
= (F
z
(u) F
z
(v))v + (F
z
(u) F
z
(v)) v + F
z
(u)(u v) + F
z
(u)(u v)
([F
z
(u) F
z
(v)[ +[F
z
(u) F
z
(v)[)[v[ +[F
z
(u) + F
z
(u)[[(u v)[
[v[[u v[([u[
6n
n2
+[v[
6n
n2
) +[(u v)[[u[
4
n2
[v[[w[([v[
6n
n2
+[w[
6n
n2
) +[w[[w + v[
4
n2
. (3.8)
Therefore by Holders inequality and (3.8), we get
Proposition 3.3 When n 6, there hold
|(F(u) F(v))|
L
(2

t,x
|v|
L
2

t,x

|v|
6n
n2
L
2(n+2)
n2
t,x
+|w|
6n
n2
L
2(n+2)
n2
t,x

|w|
L
2(n+2)
n2
t,x
+|w|
L
2

t,x

|v|
4
n2
L
2(n+2)
n2
t,x
+|w|
4
n2
L
2(n+2)
n2
t,x

, (3.9)
|(F(u) F(v))|
L
2
t
L
2n
n+2
x
|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x

|v|
6n
n2
L
2(n+2)
n2
t,x
+|w|
6n
n2
L
2(n+2)
n2
t,x

|w|
L
2(n+2)
n2
t,x
+|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x

|v|
4
n2
L
2(n+2)
n2
t,x
+|w|
4
n2
L
2(n+2)
n2
t,x

. (3.10)
14
We need the following estimates:
Proposition 3.4 We have
|F(u)|
L
2 C|u|
4
n2

H
1
|u|

H
2
, (3.11)
|F(u)|
L
2 C|u|
6n
n2

H
1
|u|
2

H
2
C|u|
4
n2

H
1
|u|

H
3, n 6. (3.12)
Proof. By Holders inequality and Sobolevs embedding,
|F(u)|
L
2 C|u|
n+2
n2
L
2n+4
n2
C(|u|
4
n+2
L
2
|u|
n2
n+2
L
2
)
n+2
n2
.
Similarly when n 6,
|F(u)|
L
2 C|[u[
4
n2
u|
L
2 C|[u[
4
n2
|
L
n|u|
L
2n
n2
C|u|
4
n2
L
4n
n2
|u|
L
2 C(|u|
6n
4

H
1
|u|
n2
4

H
2
)
4
n2
|u|

H
2. (3.13)
When n 6, we need to compute F(u). Noticing that F
zz
(z) =
2n
(n2)
2
[z[
6n
n2
|z|
z
,
F
z z
= F
zz
=
2n
(n2)
2
[z[
6n
n2
z
|z|
, F
z z
=
82n
(n2)
2
[z[
6n
n2
z
|z|
, we nd that F(u(x)) obeys the
chain rule:
F(u(x)) =F
zz
(u(x))(u(x))
2
+ 2F
z z
(u(x))[u(x)[
2
+ F
z z
(u(x))(u(x))
2
+ F
z
(u(x))u(x) + F
z
(u(x))u(x). (3.14)
Therefore, it satises the estimate:
[F(u(x))[ C([u[
6n
n2
[u[
2
+[u[
4
n2
[u[), n 6. (3.15)
3.2 Proof of Theorem 1.1
Suppose that u and v are solutions of (1.3) and (1.4), respectively. Dening w =
u v, then we see that w satises:

w
t
( + i)w + v + (a + i)([w + v[
4
n2
(w + v) [v[
4
n2
v) + a[v[
4
n2
v = 0,
w(0) = w
0
.
w can be rewritten as
w = S

(t)w
0
A

[v + (a + i)([w + v[
4
n2
(w + v) [v[
4
n2
v) + a[v[
4
n2
v]. (3.16)
If the initial data w
0
is small enough, then we can prove the following estimate for
the solution of (3.16), whose proof will be given in the end of this section:
15
Proposition 3.5 Let 3 n 6. Suppose that w is a solution to (3.16), T > 0 is
arbitrary, |w
0
|
H
1 <1, u
0
H
1
, v
0
H
3
, 0 < T, a . Then we have
|w|
L
2(n+2)
n2
t,x
([0,T]R
n
)
C(|u
0
|
H
1, |v
0
|
H
3).
Proposition 3.6 Let 3 n 6. Suppose v is a solution to (1.4) and v
0
H
3
,
then for k = 1, 2, 3,
|v(t)|
L

H
k
x
(R
+
R
n
)
C. (3.17)
Proof. If n = 3, 4, (3.17) is shown by Bourgain [3]. If n = 5, 6, since [v[
4/(n2)
v
is in C
2
and not in C
3
, we need to use the formal time-dierentiation method
to overcome the derivative loss of the nonlinearity. This method has been used by
Kato [12], Cazenave and Weissler [4]. The case k = 1 follows immediately from the
energy preserving fact E(v(t)) = E(v
0
) and Sobolevs embedding, hence
|v|
L

H
1
x
(R
+
R
n
)
C(|v
0
|

H
1
). (3.18)
When k = 2, v can be written as the following integral form:
v = S(t)v
0
i

t
0
S(t )F(v)d.
In view of Theorem 2.8, we see that
|v|
L
2(n+2)
n2
t,x
(R
+
R
n
)
+|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(R
+
R
n
)
C(|v
0
|

H
1
). (3.19)
So, for a given small parameter 0 < <1, we can divide R
+
into J disjoint intervals
I
j
such that R
+
=
J
j=1
I
j
and
|v|
L
2(n+2)
n2
t,x
(I
j
R
n
)
+|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
, j = 1, ..., J. (3.20)
By Corollary 2.3 and (3.15),
|v|

S
0
(I
j
R
n
)
|v(minI
j
)|
2
+|[v[
4
n2
v|
L
(2

t,x
(I
j
R
n
)
+|[v[
6n
n2
[v[
2
|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
. (3.21)
Using Holders inequality and Sobolevs embedding, one has that
|[v[
6n
n2
[v[
2
|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
16
|v|
6n
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
|v|
6n
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
|
2
v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)

4
n2
|v|

S
0
(I
j
R
n
)
, (3.22)
and
|[v[
4
n2
v|
L
(2

t,x
(I
j
R
n
)
|v|
4
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2

t,x
(I
j
R
n
)

4
n2
|v|

S
0
(I
j
R
n
)
. (3.23)
Hence, in view of (3.20)(3.23), we have
|v|

S
0
(I
j
R
n
)
|v(minI
j
)|
2
+
4
n2
|v|

S
0
(I
j
R
n
)
. (3.24)
Taking j = 1 and noticing that v(minI
1
) = v
0
, we obtain that
|v|

S
0
(I
1
R
n
)
2|v
0
|

H
2
, |v|
L

t
L
2
x
(I
1
R
n
)
2|v
0
|

H
2
, (3.25)
which implies that |v(minI
2
)|
2
2|v
0
|

H
2
. Repeating the procedure above, we
have
|v|

S
0
(I
2
R
n
)
2
2
|v
0
|

H
2, |v|
L

t
L
2
x
(I
2
R
n
)
2
2
|v
0
|

H
2.
By iteration, for any j = 1, ..., J, we see that
|v|

S
0
(I
j
R
n
)
2
j
|v
0
|

H
2, |v|
L

t
L
2
x
(I
j
R
n
)
2
j
|v
0
|

H
2
.
Hence,
|v|
L

t
L
2
x
(R
+
R
n
)
C(|v
0
|

H
1)|v
0
|

H
2, (3.26)
which concludes the conclusion in the case k = 2.
Finally, we consider the case k = 3 and use the equation
v
t
iv = i([v[
4
n2
v) = iF(v) (3.27)
to estimate |v|

H
3
. It suces to calculate |v
t
|
2
and |(F(v))|
2
. It is easy to see
that v
t
satises the following integral equation:
v
t
= iS(t)v
0
iS(t)F(v
0
) i

t
0
S(t )

(F(v()))d. (3.28)
17
Let I
j
(j = 1, ..., J) be the decomposition of R
+
as in (3.20). We have
|v
t
|

S
0
(I
j
R
n
)
|v(minI
j
)|

H
3 +|F(v(minI
j
))|
2
+|[v[
4
n2
v
t
|
L
(2

t,x
(I
j
R
n
)
+|[v[
6n
n2
vv
t
|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
. (3.29)
Using the same way as in Proposition 3.2, we get
|[v[
4
n2
v
t
|
L
(2

t,x
(I
j
R
n
)
|v|
4
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v
t
|
L
2

t,x
(I
j
R
n
)

4
n2
|v
t
|

S
0
(I
j
R
n
)
. (3.30)
In view of Holders inequality and Sobolevs embedding, we have
|[v[
6n
n2
vv
t
|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
|v|
6n
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
|v
t
|
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
6n
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
|v
t
|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)

4
n2
|v
t
|

S
0
(I
j
R
n
)
. (3.31)
Inserting the estimates above into (3.29) and using Proposition 3.4, we conclude
that
|v
t
|

S
0
(I
j
R
n
)
|v(minI
j
)|

H
3 +|F(v(minI
j
))|
2
|v(minI
j
)|

H
3
(1 +|v(minI
j
)|
4/(n2)

H
1
). (3.32)
Taking notice of v(minI
1
) = v
0
, we have
|v
t
|
L

t
L
2
x
(I
1
R
n
)
|v
0
|

H
3
(1 +|v
0
|
4
n2

H
1
). (3.33)
On the other hand, from Proposition 3.4, (3.18) and (3.26), it follows that for any
t > 0,
|F(v(t))|
L
2 |v(t)|
6n
n2

H
1
|v(t)|
2

H
2
C(|v
0
|

H
1)|v
0
|

H
3. (3.34)
Hence, by equation (3.27),
|v|
L

t
L
2
x
(I
1
R
n
)
|v
t
|
L

t
L
2
x
(I
1
R
n
)
+|F(v(t))|
L

t
L
2
x
(I
1
R
n
)
C(|v
0
|

H
1)|v
0
|

H
3. (3.35)
18
(3.35) implies that
|v(minI
2
)|

H
3
C(|v
0
|

H
1
)|v
0
|

H
3
. (3.36)
Combining (3.32) with (3.34) and (3.36), we have
|v
t
|

S
0
(I
2
R
n
)
C(|v
0
|

H
1)|v
0
|

H
3.
Again, using the same way as in (3.35), we obtain that
|v|
L

t
L
2
x
(I
2
R
n
)
C(|v
0
|

H
1)|v
0
|

H
3.
Using standard iteration method, we get for any j = 1, J,
|v|
L

t
L
2
x
(I
j
R
n
)
C(|v
0
|

H
1
)|v
0
|

H
3
,
which implies that
|v|
L

t
L
2
x
(R
+
R
n
)
C(|v
0
|

H
1
)|v
0
|

H
3
. (3.37)
This completes the conclusion.
Now we can prove our main result:
Proof of Theorem 1.1. First, we consider the convergence in L
2
space. In view
of Corollary 2.5 and Proposition 3.2, we have
|w|

S
0
(IR
n
)
C|w(min I)|
2
+ C|v|
L
1
t
L
2
x
(IR
n
)
+ aC|[v[
4
n2
v|
L
(2

t,x
(IR
n
)
+ C|[w + v[
4
n2
(w + v) [v[
4
n2
v|
L
(2

t,x
(IR
n
)
C|w(min I)|
2
+ C[I[|v|
L

t
L
2
x
(IR
n
)
+ aC|v|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)
|v|
L
2

t,x
(IR
n
)
+ C

|w + v|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)
+|v|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)

|w|
L
2

t,x
(IR
n
)
. (3.38)
By Theorem 2.8 and Proposition 3.5, we can divide [0, T] into N = N(|u
0
|
H
1, |v
0
|
H
3)
disjoint time intervals I
k

N
k=1
such that
C

|w + v|
4
n2
L
2n+4
n2
t,x
(I
k
R
n
)
+|v|
4
n2
L
2n+4
n2
t,x
(I
k
R
n
)

1
2
, 1 k N. (3.39)
Replace I by I
k
in (3.38), and with (3.39), then
|w|

S
0
(I
k
R
n
)
C|w|
L

t
L
2
x
(I
k1
R
n
)
+ C[I
k
[|v|
L

t
L
2
x
(I
k
R
n
)
19
+
a
2
|v|
L
2

t,x
(I
k
R
n
)
+
1
2
|w|
L
2

t,x
(I
k
R
n
)
. (3.40)
Using iteration, we get
|w|

S
0
(I
k
R
n
)
2C|w|
L

t
L
2
x
(I
k1
R
n
)
+ 2C[I
k
[|v|
L

t
L
2
x
(I
k
R
n
)
+ a|v|
L
2

t,x
(I
k
R
n
)
2C(2C|w|
L

t
L
2
x
(I
k2
R
n
)
+ 2C[I
k1
[|v|
L

t
L
2
x
(I
k1
R
n
)
+ a|v|
L
2

t,x
(I
k1
R
n
)
) + 2C[I
k
[|v|
L

t
L
2
x
(I
k
R
n
)
+ a|v|
L
2

t,x
(I
k
R
n
)

(2C)
k
|w
0
|
2
+ (2C)[I
k
[|v|
L

t
L
2
x
(I
k
R
n
)
+ (2C)
2
[I
k1
[|v|
L

t
L
2
x
(I
k1
R
n
)
+ + (2C)
k
[I
1
[|v|
L

t
L
2
x
(I
1
R
n
)
+ a|v|
L
2

t,x
(I
k
R
n
)
+ 2Ca|v|
L
2

t,x
(I
k1
R
n
)
+ + (2C)
k
a|v|
L
2

t,x
(I
1
R
n
)
(2C)
k
(|w
0
|
2
+

jk
[I
j
[|v|
L

t
L
2
x
(
jk
I
j
R
n
)
)
+ a(sup
jk
|v|
L
2

t,x
(I
j
R
n
)
)(1 + 2C + + (2C)
k
)
(2C)
k+1

|w
0
|
2
+

jk
[I
j
[|v|
L

t
L
2
x
(
jk
I
j
R
n
)
+ a sup
jk
|v|
L
2

t,x
(I
j
R
n
)

. (3.41)
In view of Corollary 2.3 and inequality (3.39), one can easily check that
sup
jk
|v|
L
2

t,x
(I
j
R
n
)
2C(E(v
0
))|v
0
|
2
. (3.42)
Summarizing the above discussion, we obtain that
|w|
L

t
L
2
x
([0,T]R
n
)
(2C)
N+2
(|w
0
|
2
+ T|v|
L

t
L
2
x
([0,T]R
n
)
+ a|v
0
|
2
), (3.43)
where N = N(|u
0
|
H
1, |v
0
|
H
3) is a large constant that depends only on|u
0
|
H
1 and
|v
0
|
H
3, and is independent of and a. Combining (3.43) with (3.26), we get for
any T > 0
|w|
L

t
L
2
x
([0,T]R
n
)
(2C)
N+2
(|w
0
|
2
+ T|v
0
|
H
2 + a|v
0
|
2
) (3.44)
which implies the desired results. The proof of (1.8) for k = 1 proceeds in a similar
way as that for k = 0, which is essentially implied by the proof of Proposition 3.5,
see below and we omit the details.
Proof of Proposition 3.5. Let T > 0 and 0 < <1 be a suciently small para-
meter. Following the proof of Proposition 3.6, we can divide [0, T] into J (depending
20
on |v
0
|
H
3) disjoint intervals I
j

J
j=1
such that on each interval I
j
there holds
|v|
L
2(n+2)
n2
t,x
(I
j
R
n
)
+|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
. (3.45)
Choose 0 < < 1 small enough satisfying (4C)
J
. Let |w
0
|
H
1 /2, 0 <
T, a /4. By Sobolevs embedding, Corollary 2.3, Propositions 3.13.3 and
Proposition 3.6,
|w|
L
2(n+2)
n2
t,x
(I
j
R
n
)
+|w|
L

t
L
2
x
(I
j
R
n
)
|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
+|w|
L

t
L
2
x
(I
j
R
n
)
C|w(minI
j
)|

H
1 + C|v|
L
1
t
L
2
x
(I
j
R
n
)
+ Ca|F(v)|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
+ C|(F(u) F(v))|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
C|w(minI
j
)|

H
1 + C[I
j
[|v|
L

t
L
2
x
(I
j
R
n
)
+ Ca|v|
4
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
|v|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
+ C|v|
L
2(n+2)
n2
tI
j
L
2n(n+2)
n
2
+4
x

|v|
6n
n2
L
2(n+2)
n2
tI
j
,x
+|w|
6n
n2
L
2(n+2)
n2
tI
j
,x

|w|
L
2(n+2)
n2
tI
j
,x
+ C|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)

|v|
4
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)
+|w|
4
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)

C|w(minI
j
)|

H
1 + C
+ C

6n
n2
+|w|
6n
n2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)

|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
+ C|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)

4
n2
+ C|w|
4
n2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)

C|w(minI
j
)|

H
1 + C + C
4
n2
|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
+ C|w|
4
n2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
+ C|w|
n+2
n2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
. (3.46)
Taking j = 1 and noticing that w(minI
1
) = w
0
, by the standard continuity method,
we get
|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
1
R
n
)
+|w|
L

t
L
2
x
(I
1
R
n
)
2C, (3.47)
21
where the constant C depends only on |u
0
|
H
1 and |v
0
|
H
3. By (3.47), we see that
|w(minI
2
)|
2
2C. Again, it follows from (3.46) that
|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
2
R
n
)
+|w|
L

t
L
2
x
(I
2
R
n
)
4C
2
+ C
4
n2
|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
2
R
n
)
+ C|w|
4
n2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
2
R
n
)
+ C|w|
n+2
n2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
2
R
n
)
. (3.48)
The continuity method yields that
|w|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
2
R
n
)
+|w|
L

t
L
2
x
(I
2
R
n
)
(4C)
2
.
Noticing that (4C)
J
, we can repeat the procedure above step by step and
nally get the result, as desired.
4 Proof of Theorem 1.3
4.1 Concentration for the MEBS
Using the induction hypothesis (Lemma 1.5) and bilinear estimate, we construct a
near solution to (1.3), then applying the long-time perturbation lemma, one can
prove the following proposition which means that the MEBS can not be included in
this class, that is to say, the MEBS can only concentrate in one particular frequency
in the phase space. Since the proof is similar to NLS (1.4) as in [5, 18], we omit the
details.
Proposition 4.1 (Frequency localization of energy at each time) Let u be a minimal
energy blowup solution to (1.3), then for any t I
0
, there exists a dyadic frequency
N(t) 2
Z
such that for every
4

0
, we have
|P
c()N(t)
u(t)|

H
1
x
, (4.1)
|P
C()N(t)
u(t)|

H
1
x
, (4.2)
and
|P
c()N(t)<<C()N(t)
u(t)|

H
1
x
1, (4.3)
where 0 < c() <1 <C() < .
22
On the other hand, the minimal energy blowup solution is concentrated in physical
space. First, we show that:
Proposition 4.2 (Potential energy is bounded below) For any minimal energy
blowup solution to (1.3), for any t I
0
, there holds |u(t)|
L
2n
n2
x

1
.
Outline of the proof. Suppose for a contrary that there exists some time t
0
I
0
such that |u(t
0
)|
L
2n
n2
x
<
1
. Normalize N(t
0
) = 1. Since u is a MEBS, by Lemma
2.9, we may assume that
|e
(+i)(tt
0
)
u(t
0
)|
L
2(n+2)
n2
t,x
([t
0
,+)R
n
)
1.
Dene P
lo
= P
<c(
0
)
, P
hi
= P
>C(
0
)
, P
me
= 1 P
lo
P
hi
, Strichartz estimate and
Proposition 4.1 imply that for any t > t
0
,
|e
(+i)(tt
0
)
P
lo
u(t
0
)|
L
2(n+2)
n2
t,x
([t
0
,+)R
n
)
+|e
(+i)(tt
0
)
P
hi
u(t
0
)|
L
2(n+2)
n2
t,x
([t
0
,+)R
n
)

0
.
Hence
|e
(+i)(tt
0
)
P
me
u(t
0
)|
L
2(n+2)
n2
t,x
([t
0
,+)R
n
)
1.
On the other hand, by the Strichartz estimate,
|e
(+i)(tt
0
)
P
me
u(t
0
)|
L
2(n+2)
n
t,x
([t
0
,+)R
n
)
|P
me
u(t
0
)|
L
2
x
C(
0
).
The above two estimates together with Holders inequality implies that
|e
(+i)(tt
0
)
P
me
u(t
0
)|
L

t,x
([t
0
,+)R
n
)
c(
0
).
Hence, there exist a time t
1
> t
0
, and a point x
1
R
n
such that
[e
(+i)(t
1
t
0
)
P
me
u(t
0
)(x
1
)[ c(
0
). (4.4)
Choosing time t
2
which is symmetric to t
1
with respect to t
0
, i.e., t
0
t
2
= t
1

t
0
. Dene f(t
2
) := P
me

x
1
, where
x
1
is the Dirac mass at x
1
. Dene f(t) :=
e
(+i)(tt
2
)
f(t
2
) for any t > t
2
. Then we can easily check that for any t > t
2
,
1 p , there holds |f(t)|
L
p
x
C(
0
)'t t
2
`
n
p

n
2
. Hence
c(
0
) ['e
(+i)(t
0
t
2
)
P
me
u(t
0
),
x
1
`[ = ['u(t
0
), f(t
0
)`[
|f(t
0
)|
L
2n
n+2
x
|u(t
0
)|
L
2n
n2
x

1
C(
0
)'t
1
t
0
`, (4.5)
23
which means that both t
1
and t
2
are far from t
0
. Hence
|f|
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
([t
0
,)R
n
)
C(
0
)|' t
2
`

n2
n+2
|
L
2(n+2)
n2
t
([t
0
,))
C(
0
)[t
2
t
0
[

n2
2(n+2)
C(
0
)
n2
2(n+2)
1
. (4.6)
Then, following the same step as in Visan [18], we can show that
|u|
L
2(n+2)
n2
t,x
([t
0
,+)R
n
)
C(
0
,
1
).
Since [t
0
, +) contains the interval I
1
, this contradicts (1.16).
We now state an interpolation lemma which is useful in proving the spatial
concentration of energy:
Lemma 4.3 (Corollary 4.2, [21] ) Suppose that 1 p
0
< p < , < s
1
< s <
s
0
< , 0 < < 1 and
1
p
=

p
0
+
1

, s = s
0
+ (1 )s
1
,
then we have
|u|

H
s
p
(R
n
)
C|u|

B
s
0
p
0
,p
0
(R
n
)
|u|
1

B
s
1
,
(R
n
)
.
Using the ideas in Bourgain [3] and Lemma 4.3, we can prove the following:
Proposition 4.4 (Spatial concentration of energy at each time) For any minimal
energy blowup solution u to (1.3)(or only suppose |u|
L
2n
n2
x

1
, |u|

H
1
x
1), and
for any t I
0
, there exists a dyadic number M(t) and a position x
0
(t) R
n
such
that

n/2
1
M(t)
n2
2
[P
M(t)
u(t, x
0
(t))[ M(t)
n2
2
.
Furthermore, there holds the inverse Sobolev inequality:
M(t)
n2
2
pn+kp
|P
M(t)
()
k/2
u(t, x)|
p
L
p
(|xx
0
(t)|C(
1
)/M(t))

np/2
1
M(t)
n2
2
pn+kp
, k = 0, 1 (4.7)
for all 1 < p < . In particular, we have

|xx
0
(t)|C(
1
)/M(t)
[P
M(t)
u(t, x)[
2
dx
n
1
. (4.8)
24
Proof. We only prove the second inequality in (4.7), by Proposition 4.2 and Lemma
4.3, we have for any t I
0
,

1
|u|
L
2n
n2
x
|u|
n2
n

H
1
x
|u|
2
n

n2
2
,
|u|
2
n

n2
2
,
. (4.9)
It follows that there exists a dyadic number M(t) and x
0
(t) R
n
such that
[P
M(t)
u(t, x
0
(t))[ M(t)
n2
2

n
2
1
.
Recall that P
N
= T
1
(/N)T. Dene () = (/2) + () + (2), then is a
smooth function with compact support in [1/4, 4]. Thus
M(t)
n2
2

n
2
1
[P
M(t)
u(t, x
0
)[
= [T
1
((/M(t))[[
k
) P
M(t)
()
k/2
u(t, x
0
)[
= M(t)
nk

(T
1
([[
k
))(M(t)(x
0
x))P
M(t)
()
k/2
u(t, x)dx

M(t)
nk

|xx
0
|C(
1
)/M(t)
(T
1
([[
k
))(M(t)(x
0
x))P
M(t)
()
k/2
u(t, x)dx

+ M(t)
nk

|xx
0
|>C(
1
)/M(t)
(T
1
([[
k
))(M(t)(x
0
x))P
M(t)
()
k/2
u(t, x)dx

M(t)
n
p
k
(|P
M(t)
()
k/2
u|
L
p
(|xx
0
|C(
1
)/M(t))
|T
1
([[
k
)|
L
p

+|P
M(t)
()
k/2
u|
L
2|T
1
([[
k
)|
L
2
(|xx
0
|>C(
1
)/M(t))
). (4.10)
Since is a rapidly decreasing function, we can choose C(
1
) suciently large such
that the second term in the last inequality is smaller than M(t)
n2
2

n
p
+k

n
2
1
/2, which
implies that
|P
M(t)
()
k/2
u|
L
p
(|xx
0
|C(
1
)/M(t))
M(t)
n2
2

n
p
+k

n
2
1
/2,
thus the conclusion follows.
Remark 4.5 M(t) in Proposition 4.4 is equivalent to N(t) in Proposition 4.1. More
precisely,
M(t) [c(
1
)N(t), C(
1
)N(t)].
Proof. By Proposition 4.4, for any t I
0
, there exists a dyadic number M(t) such
that
|P
M(t)
u(t, x)| 2n
n2

n/2
1
, (4.11)
25
While Proposition 4.1 tells us that there exists dyadic number N(t) satisfying
|P
c(
1
)N(t)
u(t)| 2n
n2
+|P
C(
1
)N(t)
u(t)| 2n
n2

100n
1
. (4.12)
Then
M(t) [c(
1
)N(t), C(
1
)N(t)]. (4.13)
Hence, for the MEBS, we can regard M(t) in Proposition 4.4 as to be identical with
N(t) in Propostion 4.1.
Corollary 4.6 Let I
0
be dened as in (1.17), then
[I
0
[
1
5
N
2
min
. (4.14)
Proof. By Proposition 4.4, we know that
N
2
min
N(t)
2
C(
1
)

|xx
0
(t)|C(
1
)/N(t)
[P
N
u[
2(n+2)
n2
dx.
Integrating it on I
0
, and together with (1.16), we get
[I
0
[N
2
min
C(
1
)

I
0

|xx
0
(t)|C(
1
)/N(t)
[P
N
u[
2(n+2)
n2
dx C(
1
)
1
5
,
which is the result, as desired.
Remark 4.7 If we normalize N
min
= 1, this corollary tells us that the length of I
0
on which the MEBS is dened is not too long, in fact it can be controlled by C
1
5
.
4.2 N(t) takes nitely many values
Lemma 4.8 Let u be a MEBS. Let I = [t
1
, t
2
] be any subinterval of I
0
with
[I[N(t
1
)
2

n/2
1
/4C, |u|
n+2
n2
L
2(n+2)
n2
t,x
(IR
n
)

n/2
1
/4C.
Then we have
c(
1
)N(t
1
) N(t) C(
1
)N(t
1
), t I.
Proof. By Proposition 4.4 we see that for any t = t
1
, there exists N(t
1
) such that
N(t
1
)

n2
2
[P
N(t
1
)
u(t
1
)[
n/2
1
. (4.15)
26
On the other hand, for any t
2
t
1
,
N(t
1
)

n2
2
[P
N(t
1
)
u(t
1
) P
N(t
1
)
u(t
2
)[
N(t
1
)

n2
2
+
n
2
|P
N(t
1
)
[u(t
1
) u(t
2
)]|
2
N(t
1
)

t
2
t
1
|P
N(t
1
)

u()|
2
d
N(t
1
)

t
2
t
1
|P
N(t
1
)
u()|
2
+|P
N(t
1
)
[u[
4
n2
u|
2
d
N(t
1
)
2
[t
1
t
2
[|u|
L

([t
1
,t
2
],L
2
)
+ N(t
1
)

t
2
t
1
|u()|
n+2
n2
2(n+2)
n2
d
CN(t
1
)
2
[t
1
t
2
[ + CN(t
1
)[t
1
t
2
[
1
2
|u|
n+2
n2
L
2(n+2)
n2
t,x
([t
1
,t
2
]R
n
)
.
Therefore, we obtain
N(t
1
)

n2
2
[P
N(t
1
)
u(t
1
) P
N(t
1
)
u(t
2
)[
n/2
1
/2, t
1
, t
2
I. (4.16)
(4.15) and (4.16) yield
N(t
1
)

n2
2
[P
N(t
1
)
u(t
2
)[
n/2
1
/2.
Following the same step as in Proposition 4.4, we can show that for p =
2n
n2
,

|xx
0
(t)|C(
1
)/N(t
1
)
[P
N(t
1
)
u(t
2
, x)[
p
dx
np/2
1
N(t
1
)
n2
2
pn
,
which gives that
|P
N(t
1
)
u(t
2
)| 2n
n2

n/2
1
. (4.17)
On the other hand, Proposition 4.1 tells us that
|P
c(
1
)N(t
2
)
u(t
2
)| 2n
n2
+|P
C(
1
)N(t
2
)
u(t
2
)| 2n
n2

100n
1
. (4.18)
By (4.17) and (4.18), we get
N(t
1
) [c(
1
)N(t
2
), C(
1
)N(t
2
)],
which is the result, as desired.
Corollary 4.9 Let I
j

J
j=1
be a pairwise disjoint decomposition of I
0
with I
j
=
[t
j
, t
j+1
] and
[I
j
[N(t
j
)
2

n/2
1
/4C, |u|
n+2
n2
L
2(n+2)
n2
t,x
(I
j
R
n
)

n/2
1
/4C.
Then we have N(t) N(t
j
) for any t I
j
, i.e., N(t) is a step function.
27
Proof. By Lemma 4.8, we can regard N(t) as
N(t) N(t
j
), t [t
j
, t
j+1
].
So, the result follows.
Corollary 4.10 We have
N
min
:= min
tI
0
N(t) > 0.
and N(t)
1
is a Lebesgue integrable function on I
0
.
Proof. Since N(t) takes nitely many values N(t
j
) (1 j J) and each N(t
j
) is
a dyadic number, we have the result.
4.3 Energy is almost conserved as , a 0
By Theorem 2.7, we see that the energy has a dissipation for the solutions of the
CGL. However, for the MEBS u, we show that the energy E(u(t)) tends to E
c
as
, a 0. More precisely, we have
Lemma 4.11 Let u be a minimal energy blowup solution, then

I
0

R
n
[u(, x)[
2
dxd + a

I
0

R
n
[u(, x)[
2(n+2)
n2
dxd
4
.
Proof. Let I
0
= [0, t
0
]. Since u is a solution of Eq. (1.3), we see from Theorem 2.7
that
E(u(t
0
)) +

t
0
0

R
n
[u(, x)[
2
dxd + a

t
0
0

R
n
[u(, x)[
2(n+2)
n2
dxd E
c
.
If the conclusion of Lemma 4.11 does not hold, then
E(u(t
0
)) E
c

4
.
Therefore the induction hypothesis Lemma 1.5 tells us that
|u|
L
2(n+2)
n2
t,x
(I
1
R
n
)
M(E
c

4
),
which contradicts the denition of minimal energy blowup solution on interval I
1
.
28
4.4 Morawetz estimate
Proposition 4.12 Suppose n > 6. Let u be a minimal energy blowup solution to
(1.3) and N

< c(
2
)N
min
, N
min
:= inf
tI
0
N(t), suppose furthermore that 0 < a, <

6
, then

I
0

R
n

R
n
[P
N

u(t, y)[
2
[P
N

u(t, x)[
2
[x y[
3
dxdydt
+

I
0

R
n

R
n
[P
N

u(t, y)[
2
[P
N

u(t, x)[
2n
n2
[x y[
dxdydt
1
N
3

. (4.19)
To prove this proposition, we rst recall a known statement for general Schrodinger
equation:
i
t
+ = ^. (4.20)
Proposition 4.13 (Interaction Morawetz inequality for general NLS, [18]) Let be
a solution to (4.20), then satises
(n 1)(n 3)

I
0

R
n

R
n
[(t, y)[
2
[(t, x)[
2
[x y[
3
dxdydt
+ 2

I
0

R
n

R
n
[(t, y)[
2
x y
[x y[
^,
p
(t, x)dxdydt
4||
3
L

t
L
2
x
||
L

H
1
x
+ 4

I
0

R
n

R
n
[^,
m
(t, y)[[(t, x)[[(t, x)[dxdydt, (4.21)
where we denote the mass bracket f, g
m
:= Im(f g) and the momentum bracket
f, g
p
:= Re(f g g

f).
It is easy to see that the CGL equation (1.3) can be rewritten as iu
t
+ u =
iu+(1ai)[u[
4
n2
u := [u[
4
n2
u+^

. We can regard ^

as an additional nonlinear
term for the NLS, then we need to compute the new terms ^

, u
p
and ^

, u
m
.
On the other hand, the rst term ||
L

t
L
2
x
in the right hand side of (4.21) is not
scaling invariant and it can become very large after scaling, so we should throw away
the low frequency part of when we apply Proposition 4.13. Let us now take up
the main business of this section:
Proof of Proposition 4.12. Writing the left handside of (4.19) as L(N

, u), we
claim that we only need to consider the case N

= 1, i.e., L(1, u)
1
for any
minimal energy blowup solution u to (1.3) on I
0
. In fact, if u is a minimal energy
29
blowup solution to (1.3) on I
0
, then by the scaling invariance of L
2(n+2)
n2
t,x
norm and
energy norm, u

n2
2
u(
t

2
,
x

) is a minimal energy blowup solution to (1.3) on

2
I
0
. So there holds L(1, u
N

)
1
, that is
L(1, u
N

) =

N
2

I
0

R
n

R
n
[P
1
u
N

(t, y)[
2
[P
1
u
N

(t, x)[
2
[x y[
3
dxdydt
+

N
2

I
0

R
n

R
n
[P
1
u
N

(t, y)[
2
[P
1
u
N

(t, x)[
2n
n2
[x y[
dxdydt
1
, (4.22)
in which
P
1
u
N

(t, x) = N
1
n
2

(P
N

u)

t
N
2

,
x
N

. (4.23)
By (4.22), we obtain the result, as desired. So our main task is to show the following
L(1, u) =

I
0

R
n

R
n
[P
1
u(t, y)[
2
[P
1
u(t, x)[
2
[x y[
3
dxdydt
+

I
0

R
n

R
n
[P
1
u(t, y)[
2
[P
1
u(t, x)[
2n
n2
[x y[
dxdydt
1
. (4.24)
To do this, we dene u
hi
= P
1
u, u
lo
= P
<1
u. By assumption 1 = N

< c(
2
)N
min
,
we have 1 < c(
2
)N(t) for any t I
0
. Choosing c(
2
) small enough, say, 1 <
c(
2
)N(t) <
2
c(
2
)N(t), c(
2
) is another small number depends on
2
, then com-
bining Proposition 4.1 with Sobolev embedding, we have
|u

1
2
|
L

H
1
x
(I
0
R
n
)
+|u

1
2
|
L

t
L
2n
n2
x
(I
0
R
n
)

2
. (4.25)
This in particular implies
|u
lo
|
L

H
1
x
(I
0
R
n
)
+|u
lo
|
L

t
L
2n
n2
x
(I
0
R
n
)

2
. (4.26)
(4.25) and Bernsteins inequality yield
|u
hi
|
L

t
L
2
x
(I
0
R
n
)
|P
1
1
2
u|
L

t
L
2
x
(I
0
R
n
)
+|P
>
1
2
u|
L

t
L
2
x
(I
0
R
n
)
|P
1
1
2
u|
L

H
1
x
(I
0
R
n
)
+
2
|P
>
1
2
u|
L

H
1
x
(I
0
R
n
)

2
. (4.27)
Our goal is to prove (4.24), which implies particularly that

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
hi
(t, x)[
2
[x y[
3
dxdydt
1
, (4.28)
30
which can be rewritten as
|[u
hi
[
2
|
L
2
t

n3
2
x
(I
0
R
n
)

1
2
1
. (4.29)
By the standard continuity argument, it suces to prove (4.29) under the bootstrap
hypothesis:
|[u
hi
[
2
|
L
2
t

n3
2
x
(I
0
R
n
)
(C
0

1
)
1
2
, (4.30)
where C
0
is a large constant depending on energy but not on any
i
. We claim that
(4.30) implies
|[[

n3
4
u
hi
|
L
4
t,x
(I
0
R
n
)
(C
0

1
)
1
4
, (4.31)
as can be seen by taking f = u
hi
in the following lemma:
Lemma 4.14 (Lemma 5.6, [18])
|[[

n3
4
f|
L
4
x
|[[

n3
2
[f[
2
|
1
2
2
,
To prove Proposition 4.12, we rst prove a weaken form:
Proposition 4.15 Under the same assumption as Proposition 4.12, we have

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
hi
(t, x)[
2
[x y[
3
dxdydt +

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
hi
(t, x)[
2n
n2
[x y[
dxdydt

2
+
2

I
0

R
n
[u
hi
(t, x)[[P
hi
([u[
4
n2
u [u
hi
[
4
n2
u
hi
[u
lo
[
4
n2
u
lo
)(t, x)[dxdt (4.32)
+
2

I
0

R
n
[u
hi
(t, x)[[P
lo
([u
hi
[
4
n2
u
hi
)(t, x)[dxdt (4.33)
+
2

I
0

R
n
[u
hi
(t, x)[[P
hi
([u
lo
[
4
n2
u
lo
)(t, x)[dxdt (4.34)
+
2
2

I
0

R
n
[u
lo
(t, x)[[u
lo
(t, x)[
4
n2
[u
hi
(t, x)[dxdt (4.35)
+
2
2

I
0

R
n
[u
lo
(t, x)[[u
hi
(t, x)[
n+2
n2
dxdt (4.36)
+
2
2

I
0

R
n
[P
lo
([u[
4
n2
u)(t, x)[[u
hi
(t, x)[dxdt (4.37)
+

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
lo
(t, x)[
n+2
n2
[u
hi
(t, x)[
[x y[
dxdydt (4.38)
31
+

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
lo
(t, x)[[u
hi
(t, x)[
n+2
n2
[x y[
dxdydt (4.39)
+

I
0

R
n

R
n
[u
hi
(t, y)[
2
[P
lo
([u
hi
[
4
n2
u
hi
)(t, x)[[u
hi
(t, x)[
[x y[
dxdydt. (4.40)
Proof. We apply Proposition 4.13 with = u
hi
, ^ = P
hi
([u[
4
n2
u) + iu
hi

aiP
hi
([u[
4
n2
u) := ^
1
+^
2
+^
3
, then we obtain
(n 1)(n 3)

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
hi
(t, x)[
2
[x y[
3
dxdydt
+ 2

I
0

R
n

R
n
[u
hi
(t, y)[
2
x y
[x y[
^
1
+^
2
+^
3
, u
hi

p
dxdydt
4|u
hi
|
3
L

t
L
2
x
|u
hi
|
L

H
1
x
+ 4

I
0

R
n

R
n
[^
1
+^
2
+^
3
, u
hi

m
(t, y)[[u
hi
(t, x)[[u
hi
(t, x)[dxdydt. (4.41)
By energy conservation and (4.27),
4|u
hi
|
3
L

t
L
2
x
|u
hi
|
L

H
1
x

3
2
.
Visan [18] has shown that

I
0

R
n

R
n
[^
1
, u
hi

m
(t, y)[[u
hi
(t, x)[[u
hi
(t, x)[dxdydt (4.32) + (4.33) + (4.34),

I
0

R
n

R
n
[u
hi
(t, y)[
2
x y
[x y[
[^
1
, u
hi

p
(t, x)[dxdydt

I
0

R
n

R
n
[u
hi
(t, y)[
2
[u
hi
(t, x)[
2n
n2
[x y[
dxdydt + (4.35) + . . . + (4.40).
So, it suces to consider the corresponding estimates for ^
2
and ^
3
. We rst begin
with the momentum bracket term:
^
2
, u
hi

p
= Re(iu
hi
u
hi
u
hi
iu
hi
)
= Im[([u
hi
[
2
+ u
hi
u
hi
) 2u
hi
u
hi
], (4.42)
and
^
3
, u
hi

p
= ai[u[
4
n2
u, u
hi

p
aiP
lo
([u[
4
n2
u), u
hi

p
= ai[u[
4
n2
u, u
p
ai[u
lo
[
4
n2
u
lo
, u
lo

p
32
ai([u[
4
n2
u [u
lo
[
4
n2
u
lo
), u
lo

p
aiP
lo
([u[
4
n2
u), u
hi

p
= ai([u[
4
n2
u [u
lo
[
4
n2
u
lo
), u
lo

p
aiP
lo
([u[
4
n2
u), u
hi

p
:= I
1
+ I
2
, (4.43)
where in the last equality, we use the fact that i[u[
4
n2
u, u
p
= 0. This can be
seen in the following:
i[u[
4
n2
u, u
p
= Re(i[u[
4
n2
u u u(i[u[
4
n2
u))
= Im[[u[
4
n2
u u u([u[
4
n2
u)] = 0.
Now we handle the mass bracket term,
^
2
, u
hi

m
= Im(iu
hi
u
hi
) = Re(u
hi
u
hi
), (4.44)
and
^
3
, u
hi

m
= aReP
hi
([u[
4
n2
u)u
hi
= aRe[P
hi
([u[
4
n2
u)u
hi
([u
hi
[
4
n2
u
hi
)u
hi
] a[u
hi
[
2n
n2
= aRe[P
hi
([u[
4
n2
u [u
hi
[
4
n2
u
hi
[u
lo
[
4
n2
u
lo
)u
hi
]
+ aReP
lo
([u
hi
[
4
n2
u
hi
)u
hi
aReP
hi
([u
lo
[
4
n2
u
lo
)u
hi
a[u
hi
[
2n
n2
. (4.45)
Then (4.42) corresponds to

I
0

R
n

R
n
[u
hi
(t, y)[
2
x y
[x y[
^
2
, u
hi

p
(t, x)dxdydt
= (n 1)

I
0

R
n

R
n
[u
hi
(t, y)[
2
1
[x y[
Im([u
hi
[
2
+ u
hi
u
hi
)dxdydt
+ 2

I
0

R
n

R
n
[u
hi
(t, y)[
2
x y
[x y[
Im(u
hi
u
hi
(t, x))dxdydt = II
1
+ II
2
.
We rst estimate II
1
, using the Hardy-Littlewood-Sobolev inequality and Holder

s
inequality,
II
1
= (n 1)

I
0

R
n

1
[ [
[u
hi
[
2

(x)Im([u
hi
[
2
+ u
hi
u
hi
)dxdt

I
0

1
[ [
[u
hi
[
2

L
p
x
(|[u
hi
[
2
|
L
p

x
+|u
hi
u
hi
|
L
p

x
)dt

I
0
|[u
hi
[
2
|
L
q
x
(|u
hi
|
2
L
2p

x
+|u
hi
|
L
2p

2p

x
|u
hi
|
L
2
x
)dt
33

I
0
|u
hi
|
2
L
2q
x
(|u
hi
|
2
L
2p

x
+|u
hi
|
L
2p

2p

x
|u
hi
|
L
2
x
)dt, (4.46)
where p is chosen large enough(at least p > n), and
1
q
=
1
p
+
n1
n
. Using interpolation,
|u
hi
|
L
2q
x
|u
hi
|

1
L
2
x
|u
hi
|
1
1
L
2
x
,
|u
hi
|
L
2p

x
|u
hi
|

2
L
2
x
|u
hi
|
1
2
L
2
x
,
|u
hi
|
L
2p

2p

x
|u
hi
|

3
L
2
x
|u
hi
|
1
3
L
2
x
,
where 0 <
i
< 1, i = 1, 2, 3. For convenience, we choose p =
3
2
n, then q =
3n
3n1
,
1
=
5
6
,
2
=
2
3
,
3
=
1
3
. Bernsteins inequality tells us that |u
hi
|
2
|u
hi
|
2

|u
hi
|
2
. Replace the above estimates into (4.46), and applying the energy estimate
in Theorem 2.7, |u|
2
L
2
t,x
E(u
0
), we get
II
1

I
0
|u
hi
|
5
3
L
2
x
|u
hi
|
1
3
L
2
x
(|u
hi
|
4
3
L
2
x
|u
hi
|
2
3
L
2
x
+|u
hi
|
1
3
L
2
x
|u
hi
|
2
3
L
2
x
|u
hi
|
L
2
x
)dt,

5
3
2

I
0
|u
hi
|
2
2
dt
5
3
2
. (4.47)
The second term can be bounded by
II
2

I
0
|u
hi
|
L
2
x
|u
hi
|
L
2
x
|u
hi
|
2
L
2
y
dt

2
2

I
0
|u
hi
|
2
2
dt
2
2
.
In the sequel, we deal with the contribution of the momentum bracket (4.43).
We denote by = that we omit the conjugate symbols in the equality. Writing
f, g
p
=(fg) + fg, then
I
1
= ai([u[
4
n2
u [u
lo
[
4
n2
u
lo
), u
lo

p
= a[i([u[
4
n2
u [u
lo
[
4
n2
u
lo
)u
lo
] + a[i([u[
4
n2
u [u
lo
[
4
n2
u
lo
)u
lo
]. (4.48)
Integrating by parts, we nd that the rst term in (4.48) corresponds to a scalar
multiple of

I
0

R
n

R
n
[u
hi
(t, y)[
2
[[u[
4
n2
u [u
lo
[
4
n2
u
lo
[(t, x)[u
lo
(t, x)[
[x y[
dxdydt (4.38) + (4.39).
The second term corresponds to

I
0

R
n

R
n
[u
hi
(t, y)[
2
[[u[
4
n2
u [u
lo
[
4
n2
u
lo
[(t, x)[u
lo
(t, x)[dxdydt (4.35) + (4.36).
34
Now we estimate I
2
in (4.43), when the derivative is taken on P
lo
([u[
4
n2
u), we
estimate it by a

I
0

R
n

R
n
[u
hi
(t, y)[
2
[P
lo
([u[
4
n2
u)[[u
hi
(t, x)[dxdydt (4.37), while
when the derivative falls on u
hi
, it is a bad term, so we integrate by parts, then we
obtain
a

I
0

R
n

R
n
[u
hi
(t, y)[
2
[P
lo
([u[
4
n2
u)(t, x)[[u
hi
(t, x)[
+
[u
hi
(t, y)[
2
[P
lo
([u[
4
n2
u)(t, x)[[u
hi
(t, x)[
[x y[
dxdydt (4.37) + + (4.40).
Now let us deal with the mass bracket as in (4.41). (4.44) corresponds to
C

I
0

R
n

R
n
[Reu
hi
u
hi
[(t, y)[u
hi
(t, x)[[u
hi
(t, x)[dxdydt
|u
hi
|
L

t
L
2
x
|u
hi
|
L

t
L
2
y

I
0
|u
hi
|
L
2
x
|u
hi
|
L
2
y
dt
2
2
. (4.49)
(4.45) corresponds to

I
0

R
n

R
n
[^
3
, u
hi

m
(t, y)[[u
hi
(t, x)[[u
hi
(t, x)[dxdydt
(4.32) + (4.33) + (4.34) + a

I
0

R
n

R
n
[u
hi
[
2n
n2
(t, y)[u
hi
(t, x)[[u
hi
(t, x)[dxdydt
(4.32) + (4.33) + (4.34) + a

I
0
|u
hi
|
L
2
x
|u
hi
|
L
2
x
|u
hi
|
2n
n2
L
2n
n2
y
dt,
(4.32) + (4.33) + (4.34) + a
2

I
0
|u
hi
|
2n
n2
L
2n
n2
y
dt,
in which
a

I
0
|u
hi
|
2n
n2
L
2n
n2
y
dt
a|u
hi
|
L
2
x,t
(I
0
R
n
)
|u
hi
|
n+2
n2
L
2(n+2)
n2
t,x
(I
0
R
n
)
a
1/2
[I
0
[
1/2

2
(a
n2
2(n+2)
|u
hi
|
L
2(n+2)
n2
t,x
(I
0
R
n
)
)
n+2
n2

1/2
6

1/2
5

2

1/2
4

2
, (4.50)
where in the last second inequality we apply Corollary 4.6 and Lemma 4.11. Up to
now, we nish the proof of Proposition 4.15.
Now let us develop estimates on the low and high-frequency parts to u, which
we will use to bound the error terms in Proposition 4.15.
35
Proposition 4.16 The low-frequency part of u satises
|u
lo
|

S
1
(I
0
R
n
)
C
1

4
(n2)
2
2
. (4.51)
The high-frequency part of u can be split into a good and a bad term, i.e., u
hi
= g+b,
they satisfy the estimates
|g|

S
0
(I
0
R
n
)
C
1

2
n2
2
, (4.52)
|g|

S
1
(I
0
R
n
)
C
1
, (4.53)
|[[

2
n2
b|
L
2
t
L
2n(n2)
n
2
3n2
x
(I
0
R
n
)
C
1

1
4
1
. (4.54)
Proof. We dene g and b to be the unique solutions to the following initial value
problems respectively

(i
t
+ (1 i)))g = G + P
hi
F(u
lo
) + P
hi
(F(u
lo
) F(g) F(u
lo
)),
g(t
0
) = u
hi
(t
0
).
and

(i
t
+ (1 i)))b = B + P
hi
(F(u) F(u
lo
+ g)),
b(t
0
) = 0.
where F(z) = (1 ai)[z[
4
n2
z, G+B = P
hi
(F(g)), The aim of dening b(t
0
) here is
to make use of the improved Strichartz estimate (Proposition 2.4). Then following
the same way as in [18], using bootstrap argument, we can prove our Proposition
4.16.
With Proposition 4.16 at hand, we can bound the error terms in the right hand
side of Proposition 4.15, we note here that u
lo
is a good term, we apply the estimation
(4.51) directly, while u
hi
is a bad term, we split it into g and b, making use of (4.52)
(4.54), we can also treat it, since the argument is the same as in [18], we omit it. So,
we get that (4.32)(4.40) can all be bounded by
1
, then Proposition 4.12 is proven.
By Proposition 4.16 and scaling, there holds
Corollary 4.17 Suppose n > 6, u is a minimal energy blowup solution to (1.3),
N

< c(
2
)N
min
. Then we can split P
N

u = g + b satisfying:
|g|

S
0
(I
0
R
n
)

2
n2
2
N
1

, (4.55)
|g|

S
1
(I
0
R
n
)
1, (4.56)
36
|[[

2
n2
b|
L
2
t
L
2n(n2)
n
2
3n2
x
(I
0
R
n
)

1
4
1
N

3
2

. (4.57)
Upon scaling, we can prove
|b|
L
2n
n2
t
L
2n
2
(n+1)(n2)
x
(I
0
R
n
)
(
1
4
1
N

3
2

)
n2
n
(4.58)
|b|
L
2(n+2)
n2
t
L
2n(n+2)
(n2)(n+3)
x
(I
0
R
n
)
(
1
4
1
N

3
2

)
n2
n+2
(4.59)
|P
N

u|
L
3
t
L
6n
3n4
x
(I
0
R
n
)

1
6
1
N
1

. (4.60)
4.5 Contradiction argument
Proposition 4.18 Let u be a minimal energy blowup solution to (1.3), 0 < a,

6
. Then for any t I
0
, we have
N(t) C(
4
)N
min
.
Proof. N(t) is a step function, therefore there exists t
min
I
0
such that N(t
min
) =
N
min
. By Bernstein inequality and energy conservation, we have
|P
c(
0
)N
min
<<C(
0
)N
min
u(t
min
)|
L
2
x
c(
0
)N
1
min
, (4.61)
|P
>C(
0
)N
min
u(t
min
)|
L
2
x
c(
0
)N
1
min
.
Suppose for contradiction that there exists t
evac
I
0
such that N(t
evac
) C(
4
)N
min
,
we may assume t
min
< t
evac
(since the case t
evac
< t
min
is more easier). By Proposi-
tion 4.1, for any
4

0
and for every t I
0
, |P
<c()N(t)
u(t)|

H
1
x
. Therefore,
if we take C(
4
) large enough, we have
|P
<
1
4
N
min
u(t
evac
)|

H
1
x

4
. (4.62)
Dene u
l
= P
<
10n
3
N
min
u, u
h
= P

10n
3
N
min
u, P
h
= P

10n
3
N
min
, then from (4.61), we
know
|u
h
(t
min
)|
L
2
x

1
N
1
min
. (4.63)
If we could show
|u
h
(t
evac
)|
L
2
x

1
2

1
N
1
min
, (4.64)
noticing that the Bernstein inequality tells us
|P
>C(
1
)N
min
u
h
(t
evac
)|
L
2
x
c(
1
)N
1
min
. (4.65)
37
then (4.64) and (4.65) together with the triangle inequality imply
|P
C(
1
)N
min
u
h
(t
evac
)|
L
2
x

1
4

1
N
1
min
.
Again by Bernstein, we have
|P
C(
1
)N
min
u(t
evac
)|

H
1
x

10n
3
N
min
|P
C(
1
)N
min
u
h
(t
evac
)|
L
2
x
c(
1
,
3
),
which would contradict with (4.62) if we choose
4
small enough.
So now we only need to prove (4.64). We use the standard continuity method,
suppose that there exists a time t

satisfying t
min
t

t
evac
and
inf
t
min
tt

|u
h
|
L
2
x

1
2

1
N
1
min
. (4.66)
We will show that this can be bootstrapped to
inf
t
min
tt

|u
h
|
L
2
x

3
4

1
N
1
min
. (4.67)
Then t

[t
min
, t
evac
] : (4.66) holds is both open and closed in [t
min
, t
evac
], then
(4.64) holds. Now we show that (4.66) implies (4.67). Dene L(t) =

R
n
[u
h
(t, x)[
2
dx.
By (4.63), we have L(t
min
)
2
1
N
2
min
. If we could show that

t
min
[
t
L(t)[
1
100

2
1
N
2
min
, then
L(t) L(t
min
)

t
t
min
[
t
L(t)[dt
99
100

2
1
N
2
min
,
which means that (4.67) holds. Note that

t
L(t) = 2

R
n
P
h
([u[
4
n2
u), u
h

m
dx 2a

R
n
Re(P
h
([u[
4
n2
u)u
h
)dx 2

[u
h
[
2
dx
= 2

Im[(P
h
([u[
4
n2
u) [u
h
[
4
n2
u
h
)u
h
]dx
2a

R
n
Re[(P
h
([u[
4
n2
u) [u
h
[
4
n2
u
h
)u
h
]dx
2a

R
n
[u
h
[
2n
n2
dx 2

R
n
[u
h
[
2
dx, (4.68)
then we only need to bound the following three terms respectively:

t
min

R
n
[P
h
([u[
4
n2
u) [u
h
[
4
n2
u
h
[[u
h
[dxdt
1
100

2
1
N
2
min
, (4.69)
a

t
min

R
n
[u
h
[
2n
n2
dxdt
1
100

2
1
N
2
min
, (4.70)
38

t
min

R
n
[u
h
[
2
dxdt
1
100

2
1
N
2
min
. (4.71)
To estimate (4.69), we need to prove the following Stirchartz estimate for low fre-
quencies:
Proposition 4.19 Under the assumptions above, there holds
|P
N
u|

S
1
([t
min
, t
evac
]R
n
)

4
+ max

3
2
3
N
3
2
,

n+2
n2
3
N
n+2
n2
,
for all N
3
N
min
.
To prove Proposition 4.19, we now set up an inverse Strichartz estimate for the low
frequency part of the solution of the CGL (1.3). That is to say, when restricting to
low frequency, we in fact can solve the CGL backward:
Lemma 4.20 Suppose u is a minimal energy blowup solution to (1.3), 0 < a <

6
. Let [t, t
evac
] I
0
. Then for any admissible pairs (q, r) and (q
1
, r
1
), there holds
|P
N
u(s)|

S
1
([t, t
evac
]R
n
)
|P
N
u(t
evac
)|

H
1
x
+|P
N
([u[
4
n2
u)|
L
q

1
s
L
r

1
x
, (4.72)
for any N
3
N
min
.
Proof of Lemma 4.20. Write F = [u[
4
n2
u. Applying the Littlewood-Paley pro-
jections P
N
and P
M
to (1.3), we have

t
P
M
P
N
u ( + i)P
M
P
N
u + (a + i)P
M
P
N
F = 0 (4.73)
holds for any dyadic number M. Then (4.73) can be rewritten as:
P
M
P
N
u(s) = e
(+i)(st
evac
)
P
M
P
N
u(t
evac
)(a+i)

s
t
evac
e
(+i)(s)
P
M
P
N
F()d,
where t s t
evac
. Hence, for any admissible pair (q, r), we have
|P
M
P
N
u|
L
q
s
L
r
x
|e
(+i)(st
evac
)
P
M
P
N
u(t
evac
)|
L
q
s
L
r
x
+

s
t
evac
e
(+i)(s)
P
M
P
N
F()d

L
q
s
L
r
x
. (4.74)
Claim. Let 1 < p < . Suppose that the conditions of Lemma 4.20 are veried.
Then (/2N)e
(t
evac
s)||
2
, (/2N)e
s||
2
, (/2N)e
||
2
M
p
.
39
The Claim is a direct consequence of the Mihlin multiplier estimate and Corollary
4.6. Combining this claim with the Strichartz estimate for the NLS, we get
|e
(+i)(st
evac
)
P
M
P
N
u(t
evac
)|
L
q
s
L
r
x
|P
2N
e
(st
evac
)
P
M
P
N
u(t
evac
)|
L
2
x
|P
M
P
N
u(t
evac
)|

H
1
x
,
and

s
t
evac
e
(+i)(s)
P
M
P
N
F()d

L
q
s
L
r
x

P
2N
e
s

s
t
evac
e
i(s)
P
2N
e

P
M
P
N
F()d

L
q
s
L
r
x

s
t
evac
e
i(s)
P
2N
e

P
M
P
N
F()d

L
q
s
L
r
x
|P
2N
e

P
M
P
N
F()|
L
q

L
r

1
x
|P
M
P
N
F()|
L
q

L
r

1
x
. (4.75)
That is
|P
M
P
N
u|
L
q
s
L
r
x
|P
M
P
N
u(t
evac
)|

H
1
x
+|P
M
P
N
([u[
4
n2
u)|
L
q

1
s
L
r

1
x
.
Squaring and summing this inequality over all M, we obtain (4.72).
With this lemma together with (4.62), following the same step as in Visan [18],
we can prove Proposition 4.19 and then the estimate (4.69).
Now let us bound (4.70), by Corollary 4.6, we get
a|u
h
|
2n
n2
L
2n
n2
t,x
(I
0
R
n
)
a[I
0
[|u
h
|
2n
n2
L

t
L
2n
n2
x

1
100

2
1
N
2
min
. (4.76)
Now we estimate (4.71). By Bernsteins inequality and Lemma 4.11, we have

t
min

R
n
[u
h
[
2
dxdt
20n
3
N
2
min

t
min

R
n
[u
h
[
2
dxdt

20n
3
N
2
min

4

1
100

2
1
N
2
min
,
which is (4.71), as desired.
Now we have enough estimates to complete the contradiction argument.
Lemma 4.21 For any minimal energy blowup solution u to (1.3), we have

I
0
N(t)
1
C(
1
,
2
)N
3
min
. (4.77)
40
Furthermore, by Proposition 4.18, we have
[I
0
[ C(
1
,
2
,
4
)N
2
min
. (4.78)
Proof. By (4.60), we have

I
0

R
n
[P
N

u[
6n
3n4
dx
3n4
2n
dt
1
2
1
N
3

, N

< c(
2
)N
min
. (4.79)
Let N

= c(
2
)N
min
, then

I
0

R
n
[P
N

u[
6n
3n4
dx
3n4
2n
dt C(
1
,
2
)N
3
min
. (4.80)
On the other hand, Bernstein and energy conservation tell us that

|xx(t)|C(
1
)/N(t)
[P
<N

u(t)[
6n
3n4
dx C(
1
)N(t)
n
|P
<N

u(t)|
6n
3n4
L

x
C(
1
)N(t)
n
N(t)
3n(n2)
3n4
c(
2
)|P
<N

u(t)|
6n
3n4
L
2n
n2
x
c(
2
)N(t)

2n
3n4
. (4.81)
By Proposition 4.4, we have

|xx(t)|C(
1
)/N(t)
[u(t)[
6n
3n4
dx c(
1
)N(t)

2n
3n4
. (4.82)
Hence the triangle inequality oers
c(
1
)N(t)

2n
3n4

|xx(t)|C(
1
)/N(t)
[P
N

u(t, x)[
6n
3n4
dx. (4.83)
Integrating over I
0
and comparing with (4.80), we get (4.77).
Proposition 4.22 Suppose u is a minimal energy blowup solution, then
|u|
L
2(n+2)
n2
t,x
(I
0
R
n
)
C(
0
,
1
,
2
,
4
). (4.84)
Proof. Let = (
0
, N
max
) > 0 be chosen later. Divide I
0
into O([I
0
[/) subinter-
vals I
1
, . . . , I
J
with [I
j
[ . Let t
j
= min I
j
. Since N(t
j
) N
max
, by Proposition
4.1
|P
C(
0
)N
max
u(t
j
)|

H
1
x

0
. (4.85)
Let u(t) = e
(+i)(tt
j
)
P
<C(
0
)N
max
u(t
j
), then the above bound becomes
|u(t
j
) u(t
j
)|

H
1
x

0
, (4.86)
41
which implies on each time-space interval I
j
R
n
,

N
|e
(+i)(tt
j
)
P
N
(u(t
j
) u(t
j
))|
2
L
2(n+2)
n2
t
L
2n(n+2)
n
2
+4
x
(I
j
R
n
)
1
2

0
. (4.87)
Bernstein, Sobolev inequality, together with energy conservation yield
| u(t)|
L
2(n+2)
n2
x
C(
0
)N
n2
n+2
max
| u(t
j
)|
L
2n
n2
x
C(
0
)N
n2
n+2
max
,
hence
| u(t)|
L
2(n+2)
n2
t,x
C(
0
)N
n2
n+2
max

n2
2(n+2)
. (4.88)
Similarly,
|([ u(t)[
4
n2
u(t))|
L
2n
n+2
x
| u(t)|
L
2n
n2
x
| u(t)|
4
n2
L
2n
n2
x
C(
0
)N
max
| u(t)|
L
2
x
| u(t)|
4
n2

H
1
x
C(
0
)N
max
,
which means that
|([ u(t)[
4
n2
u(t))|
L
2
t
L
2n
n+2
x
(I
j
R
n
)
C(
0
)N
max

1
2
.
With the above preparation, by Lemma 2.9 with e = (1 ai)[ u[
4
n2
u, we nd that
|u|
L
2(n+2)
n2
t,x
(I
j
R
n
)
1.
Adding them together, we get
|u|
L
2(n+2)
n2
t,x
(I
0
R
n
)
C(
0
)N
2
max
[I
0
[ C(
0
, ...,
4
).

This proposition contradict with the denition of minimal energy blowup solution,
hence Theorem 1.3 is proven.
5 Proof of Theorem 1.2
Suppose u, v are solutions to (1.3) and (1.4) respectively, let w = u v, then w
satises the following equation:
w
t
iw + i([u[
4
n2
u [v[
4
n2
v) u + a[u[
4
n2
u = 0, w
0
= u
0
v
0
. (5.1)
42
Applying the Strichartz estimate for NLS on any time-space interval I R
n
, we
obtain
|w|
L

t
L
2
x
(IR
n
)
C|w(min I)|
2
+ C|u|
L
1
t
L
2
x
+ aC|u|
4
n2
L
2(n+2)
n2
t,x
|u|
L
2

t,x
+ C

|u|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)
+|v|
4
n2
L
2(n+2)
n2
t,x
(IR
n
)

|w|
L
2

t,x
. (5.2)
Owing to Theorem 1.3 and 2.8, for any T > 0, we can divide [0, T] into K (de-
pends on E(u
0
) and E(v
0
)) mutually disjoint intervals I
k
such that [0, T] =

K
k=1
I
k
satisfying
C

|u|
4
n2
L
2(n+2)
n2
t,x
(I
k
R
n
)
+|v|
4
n2
L
2(n+2)
n2
t,x
(I
k
R
n
)

1
2
, 1 k K. (5.3)
Then on each I
k
R
n
, there holds
|w|
L

t
L
2
x
(I
k
R
n
)
2C|u v|
L

t
L
2
x
(I
k1
R
n
)
+ 2C([I
k
[)
1
2
(
1
2
|u|
L
2
t,x
(I
k
R
n
)
) + 2Ca|u|
L
2

t,x
(I
k
R
n
)
(2C)
k
|u
0
v
0
|
2
+ 2C([I
k
[)
1
2
(
1
2
|u|
L
2
t,x
(I
k
R
n
)
)
+ (2C)
2
([I
k1
[)
1
2
(
1
2
|u|
L
2
t,x
(I
k1
R
n
)
)
+ + (2C)
k
([I
1
[)
1
2
(
1
2
|u|
L
2
t,x
(I
1
R
n
)
) + 2Ca|u|
L
2

t,x
(I
k
R
n
)
+ (2C)
2
a|u|
L
2

t,x
(I
k1
R
n
)
+ + (2C)
k
a|u|
L
2

t,x
(I
1
R
n
)
(2C)
k
(|w
0
|
2
+

jk
([I
j
[)
1
2
(
1
2
|u|
L
2
t,x
(I
j
R
n
)
)
+ a(sup
jk
|u|
L
2

t,x
(I
j
R
n
)
)(1 + 2C + + (2C)
k
)
(2C)
k+1

|w
0
|
2
+

jk
([I
j
[)
1
2
E(u
0
) + a sup
jk
|u|
L
2

t,x
(I
j
R
n
)

. (5.4)
In view of Corollary 2.3 and inequality (5.3), one can easily check that
sup
jk
|u|
L
2

t,x
(I
j
R
n
)
2C(E(u
0
))|u
0
|
2
. (5.5)
Summarizing the above discussion, we obtain that for any T > 0,
|w|
L

t
L
2
x
([0,T]R
n
)
(2C)
K+2
(|w
0
|
2
+ (T)
1/2
+ a|u
0
|
2
), (5.6)
where K = K(E(u
0
), E(v
0
)) is a large constant that is independent of and a. This
implies the desired results.
43
Acknowledgment. The authors would like to thank Professor Carlos E. Kenig
for his suggestion on this problem. This work is supported in part by the National
Science Foundation of China, grant 10571004; and the 973 Project Foundation of
China, grant 2006CB805902.
References
[1] P. Bechouche, A.J ungel, Inviscid limits of the complex Ginzburg-Landau equation, Commun.
Math.Phys., 214 (2000), 201-226.
[2] J. Bergh and J. Lofstrom, Interpolation Spaces, SpringerVerlag, 1976.
[3] J. Bourgain, The global solutions of nonlinear Schrodinger equations, Amer. Math. Soc., 1999.
[4] T. Cazenave and F. B. Weissler, The Cauchy problem for the critical nonlinear Schrodinger
equation in H
s
, Nonlinear Anal. TMA, 14 (1990), 807836.
[5] J. Colliander, M. Keel, G. Stalani, H. Takaoka and T. Tao, Global well-posedness and
scattering for the energy-critical nonlinear Schrodinger equation in R
3
, Ann. of Math., 166
(2007), 1-100.
[6] D. Foschi, Inhomogeneous Strichartz estimates, J.Hyperbolic Dier. Equ. 2 (2005), 1-24.
[7] J. Ginibre and G. Velo, The Cauchy problem in local spaces for the complex GinzburgLandau
equation II, Contraction Methods, Commun. Math. Phys., 187 (1997), 4579.
[8] V. Ginzburg and L. Landau, On the theory of superconductivity, Zh. Eksp. Fiz., 20 (1950),
1064 English transl. in: Men of Physics: L. D. Landau, Vol. L, D. Ter Haar(ed.), New York:
Pergamon Press, 1965, 546-568.
[9] M. Grillakis, On nonlinear Schrodinger equation, Comm.Partial Dierential Equations, 25
(2000), 1827-1844.
[10] M. Gross and P. Hohenberg, Pattern formation outside of equilibrium, Rev. of Mordern Phys.,
65 (1993), 851-1089.
[11] E. Gross, Hydradynamics of a superuid condensate, J. Math. Phys., 4 (1963), 195-207.
[12] T. Kato, On nonlinear Schrodinger equations, Ann.Inst.H.Poincare Phys. Theor. 46 (1987),
113-129.
[13] M. Keel and T. Tao, Endpoint Strichartz estimates, Amer. J. of Math., 120 (1998), 955-980.
[14] S. Machihara, Y. Nakamura, The inviscid limit for the complex Ginzburg-Landau equation,
JMAA, 281 (2003), 552-564.
[15] E. Ryckman, M. Visan, Global well posedness and scattering for the defocusing energy-critical
nonlinear Schrodinger equation in R
1+4
, Amer. J. of Math., 129 (2007), 1-60.
[16] C. Sulem, P.L. Sulem, The nonlinear Schrodinger equation: Self-Focusing and Wave Collapse,
Springer-Verlag, Applied Mathematical Sciences, 139 (1999).
44
[17] T. Tao, M.Visan, Stability of energy-critical nonlinear Schrodinger equations in high dimen-
sions, Electron. J. Di. Eqns., 118 (2005), 1-28.
[18] M. Visan, The defocusing energy-critical nonlinear Schrodinger equation in higher dimentions,
Duke Math. J., 138 (2007), no.2, 281-374.
[19] B. Wang, The limit behavoir of solutions for the complex GinzburgLandau equation, Com-
mun. Pure Appl. Math., 55 (2002), 481-508.
[20] B. Wang, Large time behavior for critical and subcritical complex GinzburgLandau equations
in H
1
, Science China(A), 46 (2003), no.1, 64-74.
[21] B. Wang, Concentration phenomenon for the L
2
super critical nonlinear Schrodinger equation
in energy spaces, Commun.Contemp.Math., 8 (2006), 309-330.
[22] J.H. Wu, The inviscid limit of the complex Ginzburg-Landau equation, J.Dierential Equa-
tions 142 (1998), no.2, 413-433.
45