Invent. math.

166, 645–675 (2006)

DOI: 10.1007/s00222-006-0011-4

Global well-posedness, scattering and blow-up for

the energy-critical, focusing, non-linear Schrödinger
equation in the radial case

Carlos E. Kenig1 , Frank Merle2

1 Department of Mathematics, University of Chicago, Chicago, IL 60637, USA
2 Département de Mathématiques, Université de Cergy-Pontoise, Pontoise,
95302 Cergy-Pontoise, France

Oblatum 27-II-2006 & 17-VII-2006

Published online: 7 October 2006 – © Springer-Verlag 2006

1. Introduction

In this paper, we consider the Ḣ 1 critical non-linear Schrödinger equation

4
i∂t u + ∆u ± |u| N−2 u = 0 (x, t) ∈ R N × R
u|t=0 = u 0 ∈ Ḣ 1 (R N ).

Here the − sign corresponds to the defocusing problem, while the + sign
corresponds to the focusing problem. The theory of the Cauchy problem
(CP) for this equation was developed in [8] (Cazenave and Weissler). They
show that if u 0  Ḣ 1 ≤ δ, δ small, there exists a unique solution u ∈
C(R; Ḣ 1 (R N )) with the norm u 2(N+2) < ∞ (i.e. the solution scatters in
L x,tN−2
Ḣ (R )). See Sect. 2 of this paper for a review of these results.
1 N

In the defocusing case, Bourgain [5, 6] proved that, for N = 3, 4 and

u 0 radial, this also holds for u 0  Ḣ 1 < +∞, and that for more regular u 0 ,
the solution preserves the smoothness for all time. (Another proof of this
last fact is due to Grillakis [13] for N = 3). Bourgain’s result was then
extended to N ≥ 5 by Tao [26], still under the assumption that u 0 is radial.
Then in [9] (Colliander, Keel, Staffilani, Takaoka and Tao) the result was
obtained for general u 0 , when N = 3. This was extended to N = 4 in [24]
(Ryckman, Visan) and finally to N ≥ 5 in [28] (Visan).


The first author was supported in part by NSF, and the second one in part by CNRS.
Part of this research was carried out during visits of the second author to the University of
Chicago
646 C.E. Kenig, F. Merle

In the focusing case, these results do not hold. In fact, the classical virial
identity (see for example Glassey in [12] and Sect. 5)
  
d2 2N
|x|2
|u 0 (x, t)| 2
dx = 8 |u(t)|2
− |u(t)| N−2
dt 2
  2N
shows that if E(u 0 ) = 12 |u 0 |2 − N−2 2N
|u 0 | N−2 < 0 and |x|u 0 ∈ L 2 (R N ),
the solution must break down in finite time. Moreover,
1
W(x) = W(x, t) =   N−2
|x|2 2
1+ N(N−2)

is in Ḣ 1 (R N ) and solves the elliptic equation

4
∆W + |W| N−2 W = 0,
so that scattering cannot always occur even for global (in time) solutions.
In this paper we initiate the detailed study of the focusing case. We show
(Corollary 5.14):
Theorem 1.1. Assume that E(u 0 ) < E(W ), u 0  Ḣ 1 < W Ḣ 1 , N = 3, 4, 5
and u 0 is radial. Then the solution u with data u 0 at t = 0 is defined for all
time and there exists u 0,+ , u 0,− in Ḣ 1 such that
   
lim u(t) − eit∆ u 0,+  Ḣ 1 = 0, lim u(t) − eit∆ u 0,−  Ḣ 1 = 0.
t→+∞ t→−∞

Antecedents to this kind of result can be found in the L 2 critical case,

in the work of Weinstein [29] and in the H 1 subcritical case in the works
of Beresticky and Cazenave [3], and Zhang [30]. In particular in [3], the
authors use variational ideas and the relationship with the virial identity.
We expect that our arguments will extend to the case of radial data, for
N ≥ 6 using arguments similar to those in the appendix of [26–28] (Tao
and Visan). (It remains an interesting problem to remove the radiality.) The
result is optimal in that clearly the solution W does not scatter. We also show
that for u 0 radial, |x|u 0 ∈ L 2 (R N ), E(u 0 ) < E(W ), but u 0  Ḣ 1 > W Ḣ 1 ,
the solution must break down in finite time.
Our proof introduces a new point of view for these problems. Using
a concentration compactness argument (Sect. 4), we reduce matters to
a rigidity theorem, which we prove in Sect. 5, with the aid of a local-
ized virial identity (in the spirit of Merle [17, 18]). The radiality enters only
at one point, in our proof of the rigidity theorem (see Remark 5.2). We think
that the general strategy of our proof with one extra ingredient should also
apply in the non-radial case. In Sect. 3, we prove some elementary vari-
ational estimates which yield the necessary coercivity for our arguments.
These are automatic in the defocusing case and thus our proof gives an
alternative approach to [5] and [26] for N = 3, 4, 5.
Energy-critical focusing NLS 647

Acknowledgement. We would like to thank the referees for their suggestions and their
careful reading of the manuscript.

2. A review of the Cauchy problem

In this section we will review the Cauchy problem

4
i∂t u + ∆u + |u| N−2 u = 0 (x, t) ∈ R N × R
(CP)
u|t=0 = u 0 ∈ Ḣ 1 (R N )

i.e., the Ḣ 1 critical, focusing, Cauchy problem for NLS. We need two
preliminary results.
Lemma 2.1 (Strichartz estimate [7, 14]). We say that a pair of exponents
(q, r) is admissible if q2 + Nr = N2 and 2 ≤ q, r ≤ ∞. Then, if 2 ≤ r ≤ N−2
2N

(N ≥ 3) (or 2 ≤ r < ∞, N = 2 and 2 ≤ r ≤ ∞, N = 1) we have

i)
 it∆ 
e h  q ≤ C h L 2
L t L rx

ii)
   t 
 +∞   
 i(t−τ)∆
e g(−, τ) dτ  
+ e i(t−τ)∆
g(−, τ) dτ  ≤ C g L q L r
  
−∞ q
L t L rx 0 q
L t L rx
t x

iii)
 
 +∞ 
 e it∆
g(−, τ) dτ  ≤ C g L q L r .
 
−∞ L 2x
t x

Lemma 2.2 (Sobolev embedding). For v ∈ C0∞ (R N+1 ), we have

 
v 2(N+2) 2(N+2) ≤ C x v 2(N+2) 2N(N+2) (N ≥ 3).
N−2 N−2 N−2 N 2 +4
Lt Lx Lt Lx

(Note that 2(N+2)

N−2
= q, 2N(N+2)
N 2 +4
= r is admissible.)
4 N+2
Remark 2.3. Let f(u) = |u| N−2 u, then clearly | f(u)| ≤ |u| N−2 ,
4 4
|∂z f(u)| ≤ C|u| N−2 , |∂ z f(u)| ≤ C|u| N−2 . Moreover, for 3 ≤ N ≤ 6,

|∂z f(u) − ∂z f(v)| 6−N 6−N
≤ C |u − v| · |u| N−2 + |v| N−2 .
|∂ z f(u) − ∂ z f(v)|
Also, note that ( f )(u(x)) = ∂z f(u(x))u(x) + ∂z f(u(x))u(x), so that
4 4
| f(u) − f(v)| ≤ |u − v|{|u| N−2 + |v| N−2 }. Moreover,
x ( f(u(x))) − x ( f(v(x))) = ( f )(u(x))u − ( f )(v(x))v
= ( f )(u(x))u − ( f )(u(x))v
+ { f(u(x))) −  f(v(x))} v,
648 C.E. Kenig, F. Merle

4 6−N
so |x f(u(x)) − x f(v(x))| ≤ C|u| N−2 |u − v| + C|v|{|u| N−2 +
6−N
|v| N−2 }|u − v|.
Remark 2.4. In the estimate ii) in Lemma 2.1, one can actually show: ([14])
ii’)
 
 +∞ 
 e i(t−τ)∆
g(−, τ) dτ  ≤ C g L m  L n ,
  t x
−∞ q
L t L rx

where (q, r), (m, n) are any pair of admissible indices as in i) of Lemma 2.1.
Let us define S(I ), W(I ) norm for an interval I by
vS(I ) = v 2(N+2) 2(N+2) and vW(I ) = v 2(N+2) 2N(N+2) .
N−2 N−2 N−2 N 2 +4
LI Lx LI Lx

Theorem 2.5 (See [8]). Assume u 0 ∈ Ḣ 1 (R N ), t0 ∈ I an interval, and

u 0  Ḣ 1 ≤ A. Then, (for 3 ≤ N ≤ 5) there exists δ = δ(A) such that, if
ei(t−t0 )∆ u 0 S(I ) < δ, there exists a unique solution u to (CP) in I × R N ,
with u ∈ C(I ; Ḣ 1 (R N )),
 
x u  < ∞, uS(I ) ≤ 2δ.
W(I )

Moreover, if u 0,k → u 0 in Ḣ 1 (so that, as we will see, for k large

ei(t−t0 )∆ u 0,k S(I ) < δ) the corresponding solutions u k → u in
C(I ; Ḣ 1 (R N )).
Sketch of proof. Let us assume, without loss of generality, that t0 = 0. (CP)
is equivalent to the integral equation
 t

u(t) = e u 0 +
it∆
ei(t−t )∆ f(u) dt  ,
0
4
where f(u) = |u| N−2 u. We now let Ba,b =  t {vi(t−t
on I × Rn : vS(I ) ≤ a,
 )∆
vW(I ) ≤ b} and Φu 0 (v) = e u 0 + 0 e
it∆
f(v) dt  . We will next
choose δ, a, b so that Φu 0 (v) : Ba,b → Ba,b and is a contraction there: note
that
   
Φu (v) ≤ C A + C x f(v) N+2 2N .
0 W(I )
L 2I L x

This follows, for the first term, by i) (q = 2(N+2)

N−2
, r = 2N(N+2)
N 2 +4
) in Lemma 2.1
and by ii) in Remark 2.4, with the same q, r and m  = 2, n  = N+2 2N
. But
4
x f(u(x)) = ( f )(u(x))x u = O(|u| · |u| N−2 ) so that, using Hölder in-
equality we obtain: (for v ∈ Ba,b)
  4
Φu (v) v
4
≤ + ) · v W(I ) ≤ C A + Ca
N−2
0 W(I )
C A C S(I
N−2 b.
Energy-critical focusing NLS 649

Using Lemma 2.2 for the second term in Φu 0 , and the argument above
together with our assumption on u 0 for the first term, we obtain:
 
Φu (v) 4
≤ δ + Ca N−2 b.
0 S(I )
4
Now choose b = 2AC, and a so that Ca N−2 ≤ 1/2. Then Φu 0 (v)W(I )
≤ b. Next, if δ = a/2, and Ca( N−2 −1) b ≤ 1/2 (possible if N < 6) we obtain
4

Φu 0 (v)S(I ) ≤ a, so that Φu 0 : Ba,b → Ba,b. Next, for the contraction, we

use the same argument in conjunction with Remark 2.3.
   
Φu (v) − Φu (v ) ≤ C x f(v) − x f(v ) N+2 2N
0 0 W(I )
L 2I L x
 
 4 
≤ C |v| N−2 v − v  N+2 2N
L 2I L x
 
 
+ C  v − v |v| N−2 v  N+2
6−N
2N
L 2I L x
 
 6−N

+ C  v − v v N−2 v  N+2 2N .
L 2I L x

4

) v − v W(I ) . For
N−2
The first term is bounded as before by CvS(I
the second and third terms we use Hölder’s inequality to bound them by
6−N 6−N
Cv − v S(I ) v W(I ) vS(I
N−2  N−2
) + v  S(I ) so that
 
Φu (v) − Φu (v ) ≤ Ca N−2 v − v W(I )
4
0 0 W(I )
6−N
+ Ca N−2 bv − v S(I ) .
Lemma 2.2 gives
   
Φu (v) − Φu (v ) ≤ C Φu (v) − Φu (v )
0 0 S(I ) 0 0 W(I )

≤ Ca N−2 v − v W(I ) + Ca N−2 bv − v S(I )

4 6−N

and thus we establish the contraction property (N < 6). We then find
u ∈ Ba,b solving Φu 0 (u) = u. To show that u ∈ C(I  t ; Ḣ
1
), note that
i(t−t  )∆
e u 0 ∈ C(I ; Ḣ ) with norm bounded by A. For the term 0 e
it∆ 1
f(u) dt  ,
 
we use iii) in Lemma 2.1, with (q , r ) = (2, 2N/N + 2). The proof of
Theorem 2.5 is easily concluded from this. (The last continuity statement is
an easy consequence of the fixed point argument, see also Remark 2.17.)
q
Remark 2.6. Using Remark 2.4, it is easy to see that u ∈ L I L rx for any
admissible index pair (q, r).
Remark 2.7. There exists δ such that if u0  Ḣ 1 ≤ 
δ, the conclusion
 of Theo-
rem 2.5 applies to any interval I . In fact, eit∆ u 0  S(I ) ≤ C eit∆ u 0 W(I ) ≤
Cδ, by virtue of Lemma 2.1 i) and the claim follows.
650 C.E. Kenig, F. Merle

Remark 2.8. Given u 0 ∈ Ḣ 1 , there exists (0 ∈) I such that the hypotheses


of Theorem 2.5 is verified on I . This is clear because of eit∆ u 0  S(I ) ≤
   
C eit∆ u 0 W(I ) and the fact that eit∆ u 0 W(R) < ∞ by Lemma 2.1 i).
Remark 2.9 (Energy identity). If u is the solution constructed in Theo-
rem 2.5, we have (with 21∗ = 12 − N1 ) that
  
1 1 ∗
E(u(t)) = |u(t, x)|2 − ∗ |u(t, x)|2 dx
RN 2 2
is constant for t ∈ I . If u 0 ∈ C0∞ (R N ) this follows from a classical integra-
tion by parts, the general general case follows from a limiting argument.
Definition 2.10. Let t0 ∈ I . We say that u ∈ C(I ; Ḣ 1 (R N ))∩{u ∈ W(I )}
is a solution of the (CP) if
 t
i(t−t0 )∆ 
u|t0 = u 0 , and u(t) = e u0 + ei(t−t )∆ f(u) dt  ,
t0
4
with f(u) = |u| u. Note that if u (1) , u (2) are solutions of (CP) on I ,
N−2

u (1) (t0 ) = u (2) (t0 ), then u (1) ≡ u (2) on I × R N . This is because we can
partition I into a finite collection of subintervals I j , so that, with A =
supt∈I maxi=1,2 u (i) (t) Ḣ 1 , the S(I j ) norm of u (i) and the W(I j ) norm of
u (i) are less than a, b, where a, b are obtained in the proof of Theorem 2.5.
If j0 is then such that t0 ∈ I j0 , the uniqueness of the fixed point in the
proof of Theorem 2.5, combined with Remark 2.8 gives an interval  I  t0
so that u (1) (t) = u (2) (t), t ∈  I . A continuation argument now easily gives
u (1) ≡ u (2) , t ∈ I . This allows us to define a maximal interval I(u 0 ) =
(t0 − T− (u 0 ), t0 + T+ (u 0 )), with T± (u 0 ) > 0, where the solution is defined.
If T1 < t0 + T+ (u 0 ), T2 > t0 − T− (u 0 ), T2 < t0 < T1 , then u solves
(CP) in [T2 , T1 ] × R N , so that u ∈ C([T2 , T1 ]; Ḣ 1 (R N )), u ∈ W([T2 , T1 ])
and u ∈ S([T2, T1 ]).
Lemma 2.11 (Standard finite blow-up criterion, see [7]). If T+ (u 0 ) < ∞,
then
uS([t0,t0 +T+ (u 0 )]) = +∞.
A corresponding result holds for T− (u 0 ).
Sketch of proof. Assume T+ (u 0 ) < +∞ and that uS([t0 ,t0 +T+ (u 0 )])
< +∞. Let M = uS([t0,t0 +T+ (u 0 )]) and, for  to be chosen, find N = N()

intervals I j , Nj=1 I j = [t0 , t0 + T+ (u 0 )], such that uS(I j ) ≤ . Our first
step is to show that u L ∞ ([t0 ,t0 +T+ (u 0 )]; Ḣ 1 ) + uW([t0 ,t0 +T+ (u 0 )]) < ∞. We
write the integral equation on each interval I j , to deduce (using the proof
of Theorem 2.5 and iii) in Lemma 2.1) that
4
sup u(t) Ḣ 1 + uW(I j ) ≤ Cu(t j ) Ḣ 1 + C u S(I
N−2
j)
· uW(I j ) ,
t∈I j
Energy-critical focusing NLS 651

where t j is any fixed point in I j . Our desired estimate follows inductively

4
then, by choosing C N−2 ≤ 1/2. Once the first step is done, we then choose
tn ↑ t0 + T+ (u 0 ) and show, using the integral equation once more, that
ei(t−tn )∆ u(tn )S([tn ,t0 +T+ (u 0 )]) ≤ δ/2, for n large. But then, for n large but
fixed, and some 0 > 0, ei(t−tn )∆ u(tn )S([tn ,t0 +T+ (u 0 )]+0 ) ≤ δ. Now, Theo-
rem 2.5 applies and together with Definition 2.10 we reach a contradiction.

Definition 2.12. Let v0 ∈ Ḣ 1 , v(t) = eit∆ v0 and let {tn } be a sequence, with
limn→∞ tn = t ∈ [−∞, +∞]. We say that u(x, t) is a non-linear profile
associated with (v0 , {tn }) if there exists an interval I , with t ∈ I (if t = ±∞,
I = [a, +∞) or (−∞, a]) such that u is a solution of (CP) in I and
lim u(−, tn ) − v(−, tn ) Ḣ 1 = 0.
n→∞

Remark 2.13. There always exists a non-linear profile associated to

(v0 , {tn }). In fact, if t ∈ (−∞, +∞), this is clear by Remark 2.8, with
u 0 = v(x, t). If t = +∞, we solve the integral equation
 +∞

u(t) = e v0 +
it∆
ei(t−t )∆ f(u) dt 
t

in (tn0 , +∞) × R , for n 0 so large that eit∆ v0 S((tn0 ,∞)) ≤ δ, where δ is as

N
 +∞ 
in Theorem 2.5. Then, if n is large u(tn )−v(tn ) = tn ei(t−t )∆ f(u) dt  , and
we have  f(u) L 2 2N/N+2 < ∞, as in the proof of Theorem 2.5. But then,
Lx
(t>tn 0 )
using iii) in Lemma 2.1 we obtain u(tn )−v(tn ) Ḣ 1 ≤ C f(u) L 2 L 2N/N+2 ,
(t>tn ) x
which clearly goes to 0 as n goes infinity. A similar argument applies when
t = −∞.
Note also that if u (1) , u (2) are both non-linear profiles associated to
(v0 , {tn }) in an interval I with t ∈ I , then u (1) ≡ u (2) on I . In fact, if
t ∈ (−∞, +∞), this is clear from the Definition 2.13 and the uniqueness
result in Definition 2.10. If t = +∞, since u (i) W(I ) < ∞, for n ≥ n 0 ,
we have u (i) W(tn ,+∞) ≤  δ, where  δ is as small as we like. By the proof
of Theorem 2.5, we have (with a constant independent of u) that for n  n 0
   
sup u (1) (t) − u (2) (t) L 2 ≤ C u (1) (tn ) − u (2) (tn ) L 2 .
t∈(tn 0 ,tn )

This easily shows that u (1) ≡ u (2) on (tn0 , +∞) and hence on I , as claimed.
The case t = −∞ is similar. Because of this remark, we can always define
a maximal interval I of existence for the non-linear profile associated to
(v0 , {tn }). If t ∈ (−∞, +∞), I = (a, b), I 
I , then supt∈I  u(t) Ḣ 1 <
∞, uS(I  ) < ∞, uW(I  ) < ∞, but if either a or b are finite uS(I ) =
+∞. If t = ±∞, say t = +∞, I = (a, +∞), I  = (α, +∞), α > a, similar
statements can be made. If a > −∞, we can also say uS(I ) = +∞.
652 C.E. Kenig, F. Merle

Theorem 2.14 (Long-time perturbation theory, see also [27]). Let I ⊂ R

be a time interval and let t0 ∈ I . Let u be defined on I × R N (3 ≤ N ≤ 5)
and satisfy supt∈I 
u  Ḣ 1 ≤ A, 
u S(I ) ≤ M for some constants M, A > 0.
Assume that
(i∂t
u + ∆
u + f(
u )) = e (t, x) ∈ I × R N
(in the sense of the appropriate integral equation) and that
 i(t−t )∆ 
2N ≤ , e
u (t0 ) Ḣ 1 ≤ A , e N+2
u 0 −  u (t0 )] S(I ) ≤ .
[u 0 − 
0
L 2I L x

Then, there exists 0 = 0 (M, A, A , N ) such that there exists a solution

of (CP) with u(t0 ) = u 0 in I × R N , for 0 <  < 0 , with uS(I ) ≤
C(M, A, A , N ) and ∀t ∈ I, u(t) − 
u (t) Ḣ 1 ≤ C(A, A , M, N )(A + ).
Proof. We start the proof by showing that  u W(I ) ≤ M, where M  =

M(A, M, N ), for  ≤ 0 . In fact, for η = η(N ) small, to be determined,
split I into γ = γ(M, η) interval I j so that 
u S(I j ) ≤ η. Using the integral
equation, we have
4
u W(I j ) ≤ A + C 
 u  S(I
N−2
j)
u W(I j ) + C e L 2 L 2N/N+2
 ,
x I

4
as in the proof of Theorem 2.5, and the claim follows if Cη N−2 < 1/2. Next,
we write u =  u + w and notice that
i∂t w + ∆w + [ f(
u + w) − f(
u )] = e.
Let I j = [a j , a j+1 ], so that, in order to solve for w we need to solve, in I j ,
the integral equation
 t
i(t−a j )∆ 
w(t) = e w(a j ) + ei(t−t )∆ [ f( u )] dt 
u + w) − f(
aj
 t

+ ei(t−t )∆ e dt  .
aj

The proof of Theorem 2.5 (which holds for 3 ≤ N ≤ 5) now shows

that, for η = η(N ) small enough, and 0 = 0 (N ) small enough, we can
solve the integral equation (assuming t0 = a1 say) in I1 and obtain w
with the bounds wS(I ) ≤ 2, wW(I1 ) ≤ C(A, A ), supt∈I1 w(t) Ḣ 1 ≤
C(A, A )(A + ). We now estimate ei(t−a2 )∆ w(a2 )S(I2 ) , using the inte-
gral equation. Since ei(t−a2 )∆ ei(a2 −t0 )∆ w(t0 ) = ei(t−t0 )∆ w(t0 ), by assumption
ei(t−a2 )∆ ei(a2 −t0 )∆ w(t0 )S(I2 ) ≤ . For the integral term, we use Lemma 2.1,
4
iii) to obtain a bound for its Ḣ 1 norm at a2 by CwS(I
N−2
1)
wS(I1 ) ≤
4
C(2) N−2 C(A, A ). Clearly this procedure can be iterated γ = γ(M, N )
times, provided 0 is small enough, yielding the theorem.
Energy-critical focusing NLS 653

Remark 2.15 (See [7]). If u is a solution of (CP) in I × R N , I = [a, +∞)

(or I = (−∞, a]) there exists u + ∈ Ḣ 1 such that
 
lim u(t) − eit∆ u +  1 = 0. Ḣ
t→+∞
∞ 
To see this, note that  f(u) ∈ W(I ) and hence  t ei(t−t )∆ f(u) dt   Ḣ 1
 t 
→ 0 as t → +∞. Then, u(t) = ei(t−a)∆ u 0 + a ei(t−t )∆ f(u) dt  and hence
 ∞ 
u + = e−ia∆ u 0 + a e−it ∆ f(u) dt  has the desired property. In fact note that
the argument used at the beginning of the proof of Theorem 2.14 shows that
it suffices to assume u to be a solution of (CP) in I  × R N , I 
I , such that
uS(I ) < ∞.
Remark 2.16. We recall that, since we are working in the focusing case,
we have from the argument of Glassey [12] that if |x|2 |u 0 |2 < +∞,
E(u 0 ) < 0, there exists a finite time T such that the solution cannot be
extended for t > T . Clearly, for such a u 0 , the maximal interval of ex-
istence must be finite. (See Definition 2.10.) Note that it is unknown if
limt↑T u(t) Ḣ 1 = +∞ for a general initial data that doesn’t exist for all
time.
Remark 2.17. Theorem 2.14 also yields the following continuity fact, which
will be used later: let  u 0 ∈ Ḣ 1 ,  u 0  Ḣ 1 ≤ A, and let  u be the solution
of (CP), with maximal inteval of existence (T− ( u 0 ), T+ (u 0 )) (see Defin-
ition 2.10). Let u 0,n →  u 0 in Ḣ 1 , and let u n be the corresponding solution
of (CP), with maximal interval of existence (T− (u 0,n ), T+ (u 0,n )). Then,
T− (u 0 ) ≥ lim n→+∞ T− (u 0,n ), T+ ( u 0 ) ≤ lim n→+∞ T+ (u 0,n ) and for each
t ∈ (T− ( u 0 ), T+ (
u 0 )), u n (t) →  u (t) in Ḣ 1 .
Indeed, let I ⊂⊂ (T− ( u 0 ), T+ (u 0 )), so that Supt∈I  u (t) Ḣ 1 = A <
+∞,  u (t)S(I ) = M < +∞. We will show that, for n large, u n exists on
I and ∀t ∈ I, u n (t) −  u (t) Ḣ 1 ≤ C(M, A, N )u 0,n −  u 0  Ḣ 1 . This clearly
yields the remark. To show this, apply Theorem 2.14, with u = u n , u 0 =
u 0,n . Then, if 0 = 0 (M, A, 2A, N ) and n is so large that u 0,n − u 0  Ḣ 1 ≤ 0
and eit∆ [u 0,n −  u 0 ]S(I ) ≤ 0 , using the uniqueness of the solutions we
obtained in Definition 2.10, the claim follows.

3. Some variational estimates

Let
1
W(x) = W(x, t) =   N−2 ,
|x|2 2
1+ N(N−2)

be a stationary solution of (CP). That is, W solves the non-linear elliptic

equation
4
(3.1) ∆W + |W| N−2 W = 0.
654 C.E. Kenig, F. Merle

Moreover, W ≥ 0 and it is radially symmetric and decreasing. Note that

W ∈ Ḣ 1 , but W need not belong to L 2 (R N ). By invariances of the equation,
for θ0 ∈ [−π, π], λ0 > 0, x0 ∈ R N ,
N−2
Wθ0 ,x0 ,λ0 (x) = eiθ0 λ0 2 W(λ0 (x − x0 ))
is still a solution. By the work of Aubin [1], Talenti [25] we have the
following characterization of W:
(3.2) ∀u ∈ Ḣ 1 , u L 2∗ ≤ C N u L 2 ;
moreover,
(3.3)
If u L 2∗ = C N u L 2 , u  = 0, then ∃(θ0 , λ0 , x0 ) such that u = Wθ0 ,x0 ,λ0 ,
where C N is the best constant of the Sobolev inequality
 ∗
in dimension N.
The equation (3.1) gives |W|2 = |W|2 . Also, (3.3) yields
  ∗ N−2/N   N−2
C 2N |W|2 = |W|2 , so that C 2N |W|2 = |W|2 N .
Hence,
  
1 1 1 1 1
|W| = N and E(W ) =
2
− ∗ |W|2 = .
CN 2 2 N C NN
Lemma 3.4. Assume that
u2L 2 < W2L 2 .
Assume moreover that E(u) ≤ (1 − δ0 )E(W ) where δ0 > 0. Then, there
exists δ = δ(δ0 , N ) > 0 such that
 
(3.5) |u|2 ≤ (1 − δ) |W|2
 
2∗
(3.6) |u| − |u|
2
≥δ |u|2

(3.7) E(u) ≥ 0.
∗
C2 2∗
Proof. Consider the function f 1 (y) = 12 y − 2N∗ y 2 , and let y = u2L 2 .
Because of (3.2), f 1 (y) ≤ E(u) ≤ (1 − δ0 )E(W ) = (1 − δ0 ) N1 C1N . Note that
∗ N
C2 2∗
f 1 (0) = 0, f 1 (y) = 12 − 2N y 2 −1 , so that f 1 (y) = 0 if and only if y = yC ,

where yC = C1N = |W|2 . Note also that f 1 (yC ) = NC1 N = E(W ). But
N N
then, since 0 < y < yC and f 1 (y) ≤ (1 − δ0 ) f 1 (yC ) and f 1 is nonnegative
and strictly increasing between 0 and yC , f 1 (yC )  = 0, we have 0 < f 1 (y)
and y ≤ (1 − δ) |W|2 . Thus (3.5) and (3.7) hold. To show (3.6), con-
∗ N
sider the function g1 (y) = y − C 2N y N−2 . Because of (3.2) we have that
Energy-critical focusing NLS 655

 ∗  ∗  2∗ /2
|u|2 −|u|2 ≥ |u|2 −C 2N |u|2 = g1 (y). Note that g1 (y) = 0
if and only if y = 0 or y = yC and that g1 (0) = 1, g1 (yC ) = − N−2
 2
. We
then have, for 0 < y < yC , g1 (y) ≥ C min{y, (yC − y)}, and so, since
1
0 ≤ y < (1 − δ)yC by (3.5), (3.6) follows. Note that δ  δ0 2 .

Note that the relevance of (3.6) comes from the virial identity (see
introduction).
 
Corollary 3.8. Assume that u ∈ Ḣ 1 and that |u|2 < |W|2 . Then
E(u) ≥ 0.

Proof. If E(u) ≥ E(W ) = 1

NC NN
, this is obvious. If E(u) < E(W ), the
claim follows from (3.7).

Theorem 3.9 (Energy trapping). Let u be a solution of the (CP), with

t0 = 0, u|t=0 = u 0 such that for δ0 > 0
 
|u 0 | < |W|2 and E(u 0 ) < (1 − δ0 )E(W ).
2

Let I  0 be the maximal interval of existence given by Definition 2.10. Let

δ = δ(δ0 , N ) be as in Lemma 3.4. Then, for each t ∈ I , we have
 
(3.10) |u(t)| ≤ (1 − δ) |W|2
2

 
∗
(3.11) |u(t)|2 − |u(t)|2 ≥ δ |u(t)|2

(3.12) E(u(t)) ≥ 0.

Proof. By Remark 2.9, E(u(t)) = E(u 0 ), t ∈ I and the Theorem follows

directly from Lemma 3.4 and a continuity argument.

Corollary 3.13.
  as in Theorem 3.9. Then for all t ∈ I we have
Let u, u 0 be
E(u(t))  |u(t)|2  |u 0 |2 , with comparability constants which
depend only on δ0 .

Proof. E(u(t)) ≤ |u(t)|2 , but by (3.11) we have
   
1 1 1 2∗
E(u(t)) ≥ − |u(t)| + ∗
2
|u(t)| − |u(t)|
2
2 2∗ 2

≥ Cδ |u(t)|2 ,

 equivalence follows. For the second one note that E(u(t)) =

so the first
E(u 0 )  |u 0 |2 , by the first equivalence when t = 0.
656 C.E. Kenig, F. Merle

Remark 3.14. Assume that u 0 ∈ Ḣ 1 and that |x|u 0 ∈ L 2 (R N ). Assume that

 
E(u 0 ) < E(W ), but |u 0 | >
2
|W|2 .

If we choose δ0 so that  E(u 0 )2 < (1 − δ0 )E(W

 ), arguing as in Lemma 3.4
we
 can conclude that |u 0 | > (1 + δ)
 |W| 2
, δ = δ(δ0 , N ). But then,
2∗ ∗
|u 0 | − |u 0 | = 2 E(u 0 ) − N−2
2 2
|u 0 | ≤ 2∗ E(W ) − (N−2)
2 2
CN
1
−
N
−2δ
2δ 1
(N−2) C NN
= Since E(u(t)) = E(u 0 ), a continuity argument shows
(N−2)C NN
.

that for all t ∈ I , the maximal interval of existence, we have |u(t)|2 ≥
   ∗ −2δ
(1 + δ) |W|2 and |u(t)|2 − |u(t)|2 ≤ (N−2)C N . But, the virial
2  N

identity ([12]) shows that, if |x|u 0 ∈ L 2 (R N ) then dtd 2 |x|2 |u 0 (x, t)|2 dx =
 ∗ −16δ
8{ |u(t)|2 −|u(t)|2 } ≤ (N−2)C N . This shows that I must be finite, i.e., the
N
maximal interval of existence is finite. This argument is the critical analoge
of the H 1 subcritical result in [3].
Note that in the case where u 0 ∈ Ḣ 1 and u 0 ∈ L 2 (R N ), the same result
holds. Indeed, one can use a local version of the virial identity (See Sect. 5
for such a version) and the extra conservation
 law of the L 2 norm in time
2 −8δ
to control correction terms to obtain dtd 2 φ(|x|)|u 0 (x, t)|2 dx ≤ (N−2)C N,
N
where φ is a regular and compactetly supported function (See for example
Ogawa and Tsutsumi [22]).

4. Existence and compactness of a critical element

Let us consider the statement:

 
(SC) For all u 0 ∈ Ḣ 1 (R N ), with |u 0 |2 < |W|2 and E(u 0 ) < E(W ),
if u is the corresponding solution to the (CP), with maximal inter-
val of existence I (see Definition 2.10), then I = (−∞, +∞) and
uS((−∞,+∞)) < ∞.

 We say that (SC)(u 0) holds if for this particular u 0 , with |u 0 |2 <
|W|2 and E(u 0 ) < E(W ) and u the corresponding solution to the
(CP), with maximal interval of existence I , we have I = (−∞, +∞) and
uS((−∞,+∞)) < ∞.
Note that, because of Remark 2.7, if u 0  L 2 ≤ δ, (SC)(u 0) holds.
Thus, in light of Corollary 3.13, there exists η0 > 0 such that, if u 0 is as
in (SC) and E(u 0 ) < η0 , then (SC)(u 0) holds. Moreover, for any u 0 as in
(SC), E(u 0 ) ≥ 0, in light of Theorem 3.9. Thus, there exists a number E C ,
with η0 ≤ E C ≤ E(W ), such that, if u 0 is as in (SC) and E(u 0 ) < E C ,
(SC)(u 0) holds and E C is optimal with this property. For the rest of this
Energy-critical focusing NLS 657

section we will assume that E C < E(W ). We now prove that there exits
a critical element u 0,C at the critical level of energy E C so that (SC)(u 0,C )
does not hold and from the minimality, this element has a compactness
property up to the symetries of this equation. This is in fact a general
principle which follows from the concentration compactness ideas. More
precisely,
Proposition 4.1. There exists u 0,C in Ḣ 1 , with
 
E(u 0,C ) = E C < E(W ), |u 0,C | < |W|2
2

such that, if u C is the solution of (CP) with data u 0,C , and maximal interval
of existence I , 0 ∈ I̊ , then u C S(I ) = +∞.
Proposition 4.2. Assume u C is as in Proposition 4.1 and that (say)
u C S(I+ ) = +∞, where I+ = (0, +∞) ∩ I . Then there exists x(t) ∈ R N
and λ(t) ∈ R+ , for t ∈ I+ , such that
  
1 x − x(t)
K = v(x, t) : v(x, t) = u C , t
λ(t)(N−2)/2 λ(t)
has the property that K is compact in Ḣ 1 . A corresponding conclusion is
reached if u C S(I− ) = +∞, where I− = (−∞, 0) ∩ I .
The main tools that we will need in order to prove Propositions 4.1 and
4.2 are the following lemmas.

Lemma 4.3 (Concentration compactness). Let {v0,n } ∈ Ḣ 1 , |v0,n |2
≤ A. Assume that eit∆ v0,n  L 2(N+2)/N−2 ≥ δ > 0, where δ = δ(N ) is as in
Theorem 2.5. Then there exists a sequence {V0, j }∞ 1
j=1 in Ḣ , a subsequence of
{v0,n } (which we still call {v0,n }) and a triple (λ j,n ; x j,n ; t j,n ) ∈ R+ ×R N ×R,
with
λ j,n λ j  ,n |t j,n − t j  ,n | |x j,n − x j  ,n |
+ + + →∞
λ j  ,n λ j,n λ2j,n λ j,n

as n → ∞ for j  = j  (we say that (λ j,n ; x j,n ; t j,n ) is orthogonal if this

property is verified) such that
(4.4) V0,1  Ḣ 1 ≥ α0 (A) > 0.

If V jl (x, t) = eit∆ V0, j , then, given 0 > 0, there exists J = J(0 ) and
 
 J
1 x − x −t
(4.5) {wn }∞
j,n j,n
n=1 ∈ Ḣ , so that v0,n = , 2 + wn
1
Vl
(N−2)/2 j
λ
j=1 j,n
λ j,n λ j,n
with eit∆ wn S((−∞,+∞)) ≤ 0 , for n large
658 C.E. Kenig, F. Merle

 J 
 
(4.6) |v0,n | =
2
|V0, j | +
2
|wn |2 + o(1) as n → ∞
j=1


J
(4.7) E(v0,n ) = E V jl − t j,n /λ2j,n + E(wn ) + o(1) as n → ∞.
j=1

Remark 4.8. Lemma 4.3 is due to Keraani [15]. It is based on the “refined
Sobolev inequality” (N = 3)

h L 6 (R3 ) ≤ C h1/3

L 2 (R3 )
h2/3
Ḃ 0
,
2,∞

0
where Ḃ2,∞ is the standard Besov space [4, 11]. (4.4) is a consequence
of the proof of Corollary 1.9 in [15], (here, we use the hypothesis
eit∆ v0,n  L 2(N+2)/N−2 ≥ δ > 0) while (4.7) follows from the orthogonal-
ity of (λ j,n ; x j,n ; t j,n ) as in the proof of (4.6). The rest of the lemma is
contained in the proof of Theorem 1.6 in [15]. See also [2, 10, 16, 21].
 
Lemma 4.9. Let {z 0,n } ∈ Ḣ 1 , with |z 0,n |2 < |W|2 and E(z 0,n ) → E C
and with eit∆ z 0,n S((−∞,+∞)) ≥ δ, with δ as in Theorem 2.5. Let {V0, j } be
as in Lemma 4.3. Assume that one of the two hypothesis
(4.10) lim E V1l − t1,n /λ21,n < EC
n→∞

or after passing to a subsequence, we have that, with sn = −t1,n /λ21,n ,

E(V1l (sn )) → E C , and sn → s∗ ∈ [−∞, +∞], and if U1 is the non-linear
profile (see Definition 2.12 and Remark 2.13) associated to (V0,1, {sn }) we
have that the maximal interval of existence of U1 is I = (−∞, +∞) and
U1 S((−∞,+∞)) < ∞ and

(4.11) lim E V1l − t1,n /λ21,n = EC .

n→∞

Then (after passing to a subsequence), for n large, if z n is the solution of

(CP) with data at t = 0 equal to z 0,n , then (SC)(z 0,n ) holds.
Let us first assume the validity of Lemma 4.9 and use it (together with
Lemma 4.3) to establish Propositions 4.1 and 4.2.
Proof of Proposition 4.1. By the definition of EC , and the assumption 
that E C < E(W ), we can find u 0,n ∈ Ḣ 1 , with |u 0,n |2 < |W|2 ,
E(u 0,n ) → E C , and such that if u n is the solution of (CP) with data at t = 0,
u 0,n and maximal interval of existence In = (−T− (u 0,n ), T+ (u 0,n )), then
eit∆ u 0,n S((−∞,+∞)) ≥ δ = δ(N ) > 0, where δ is as in Theorem 2.5 and
u n S(I N ) = +∞. (Here we are also using Lemma 2.1 and Theorem 2.5.)
Note also that, since E C < E(W ), there exists δ0 > 0 so that, for all n,
we have E(u 0,n ) ≤ (1 − δ0 )E(W ). Because of Theorem 3.9, we can find
Energy-critical focusing NLS 659

 
δ so that |u n (t)|2 ≤ (1 − δ) |W|2 for all t ∈ In , all n. Apply now
Lemma 4.3 for 0 > 0 and Lemma 4.9. We then have, for J = J(0 ), that
 
J
1 x − x j,n −t j,n
(4.12) u 0,n = Vl
(N−2)/2 j
, 2 + wn ,
j=1 λ j,n
λ j,n λ j,n
 J 
 
(4.13) |u 0,n | =
2
|V0, j | +2
|wn |2 + o(1),
j=1
  

J
−t j,n
(4.14) E(u 0,n ) = E V jl + E(wn ) + o(1).
j=1
λ2j,n

Note that because of (4.13) we have, for all  n large, that |wn |2 ≤
(1−δ/2) |W|2 and |V0, j |2 ≤ (1−δ/2) |W|2 . From Corollary 3.8
it now follows that E(V jl (−t j,n /λ2jn )) ≥ 0 and E(wn ) ≥ 0. From this
and (4.14) it follows that E(V1l (−t1,n /λ21,n )) ≤ E(u 0,n ) + o(1) and hence
limn→∞ E(V1l (−t1,n /λ21,n )) ≤ E C . If the left-hand side is strictly less than
E C , Lemma 4.9 gives us a contradiction with the choice of u 0,n , for n large
(after passing to a subsequence). Hence, the left-hand side must equal E C .
Let then U1 be the non-linear profile associated to (V1l , {sn }), with
sn = −t1,n /λ21,n (after passing to a subsequence). We first note that we must
have J = 1. This is because (4.14) and E(u 0,n ) → E C , E(V1l (−sn )) → E C
now imply that E(wn ) → 0 and E(V jl (−t j,n /λ2j,n )) → 0, j = 2, ..., J.
Using (3.6) and the argument in the proof of Corollary 3.13, we have
J
|V jl (−t j,n /λ2j,n )|2 + |wn |2 → 0. We then have, since
 j=2 l 
|V j (−t j,n /λ2j,n )|2 = |V0, j |2 that V0, j = 0, j = 2, . . . , J and
 x−x
|wn |2 → 0. Hence (4.12) becomes u 0,n = (N−2)/2 1
V1l λ1,n1,n , sn + wn .
λ1,n
(N−2)/2
Let v0,n = λ1,n u 0,n (λ1,n (x +x1,n )) and note that scaling gives us that v0,n
(N−2)/2
verifies the same hypothesis as u 0,n . Moreover,  wn = λ1,n wn (λ1,n (x +

x1,n )) still verifies | wn | → 0. Thus
2


v0,n = V1 (sn ) + 
l
wn , |wn |2 → 0.

Let us return to U1 , the non-linear profile associated to (V0,1, {sn })

and let I1 = (T− (U1 ), T+ (U1 )) be its maximal interval of existence (see
Remark 2.13). Note
 that, by definition of non-linear profile, we have
|U1 (sn )|2 = |V1l (sn )|2 + o(1) and E(U1 (sn )) = E(V  1 (sn ))l + o(1).
l

1 (sn )) = E C + o(1) and that |V1 (sn )|2 =

l
Note that2 in this
 case E(V
|V0,1| = |u 0,n | + o(1) < |W| for n large by Theorem 3.9.
2 2

Let’s fix s ∈ I1 . Then E(U1 (sn )) = E(U1 (s)), so that

E(U1 (s)) = E C .
660 C.E. Kenig, F. Merle

 
Moreover,
 |U
 1 (sn )|2
< |W|2 for n large and hence by (3.10)
|U1 (s)| < |W| . If U1 S(I1 ) < +∞, Lemma 2.11 gives us that
2 2

I1 = (−∞, +∞) and we then obtain a contradiction from Lemma 4.9.

Thus,

U1 S(I1 ) = +∞

and we then set u C = U1 (after a translation in time to make s = 0).

Proof of Proposition 4.2. We argue by contradiction. For brevity of notation,

let us set u(x, t) = u C (x, t). If not, there exists η0 > 0 and a sequence {tn }∞
n=1 ,
tn ≥ 0 such that, for all λ0 ∈ R+ , x0 ∈ R N , we have
   
 1 − 
 x x 
, tn − u(x, tn ) ≥ η0 , for n  = n  .
0
(4.15)  (N−2)/2 u
λ λ 0  1
0 Ḣ x

Note that (after passing to a subsequence, so that tn → t ∈ [0, T+ (u 0 )]),

we must have t = T+ (u 0 ), in view of the continuity of the flow in Ḣ 1 , as
guaranteed by Theorem 2.5. Note that, in view of Theorem 2.5 we must
also have eit∆ u(tn )S((0,+∞)) ≥ δ.
Let us apply Lemma 4.3 to v0,n = u(tn ) with 0 > 0. We  next prove
that J = 1. In fact, if lim n→∞ E(V1l (−t1,n /λ21,n )) < E C , since |u(t)|2 ≤

(1 − δ) |W|2 by Theorem 3.9, for all t ∈ I+ and E(u(t)) = E(u 0 ) =
E C < E(W ), by Lemma 4.9 we obtain a contradiction. Hence, we must
have lim n→∞ E(V1l (−t1,n /λ21,n )) = E C . The argument
 used in the proof of
Proposition 4.1 now applies and gives J = 1, |wn |2 → 0. Thus, we
have
  
1 x − x1,n −t1,n
(4.16) u(tn ) = (N−2)/2 V1 l
, 2 + wn , |wn |2 → 0.
λ1,n λ 1,n λ 1,n

−t1,n
Our next step is to show that sn = λ21,n
must be bounded. To see this note
that
 
−(N−2)/2 x − x1,n t − t1,n
eit∆ u(tn ) = λ1,n V1l , + eit∆ wn .
λ1,n (λ1,n )2

Assume that t1,n /λ21,n ≤ −C0 , C0 a large positive constant. Then, since
eit∆ wn S((−∞,+∞)) < δ/2 for n large, and
  
 −(N−2)/2 l x − x1,n t − t1,n   
λ V1 ,  ≤  V1l (y, s)S((C0,+∞)) ≤ δ/2,
 1,n λ1,n (λ1,n )2 
S((0,+∞))

for C0 large, we get a contradiction.

Energy-critical focusing NLS 661

t1,n
If, on the other hand, λ21,n
≥ C0 , for a large positive constant C0 , n large,
we have
  
 −(N−2)/2 l x − x1,n t − t1,n   
λ V1 ,  ≤ V1l (y, s) S((−∞,−C0))
 1,n λ (λ )2 
1,n 1,n S((−∞,0))
≤ δ/2,

for C0 large. Hence, eit∆ u(tn )S((−∞,0)) ≤ δ, for n large and hence, Theo-
rem 2.5 now gives uS((−∞,tn )) ≤ 2δ, which, since tn → T+ (u 0 ) gives us
a contradiction. Thus |t1,n /λ21,n | ≤ C0 and after passing to a subsequence,

t1,n /λ21,n → t0 ∈ (−∞, +∞).

But then, since (4.15) and (4.16) imply that, for n  = n  large (independently
of λ0 , x0 ) we have
 x−x0 
1 1 λ
− x 1,n
Vl 0
, −t1,n /(λ1,n )2
(λ0 )(N−2)/2 (λ1,n )(N−2)/2 1 λ1,n
 
1 l x − x 1,n 
− V , −t1,n /(λ1,n ) 2
≥ η0 /2
(λ1,n )(N−2)/2 1 λ1,n 1 Ḣ
or
 (N−2)/2  
λ1,n yλ1,n
V1l +
xn,n − 
x0 , −t1,n /(λ1,n )2
λ1,n λ0 λ0 λ1,n

− V1l y, −t1,n /λ21,n ≥ η0 /2,

Ḣ 1

xn,n is a suitable point in R N and λ0 , 

where  x0 are arbitrary. But if we choose
λ0 = λ1,n /λ1,n , 
x0 = xn,n , we reach a contradiction since −t1,n /(λ1,n )2 →
−t0 and −t1,n /(λ1,n )2 → −t0 .
Thus, to complete the proofs of Propositions 4.1 and 4.2 we only need
to provide the proof of Lemma 4.9.
Proof
  Let us assume first that (4.11) holds and set A =
of Lemma 4.9.
|W|2 , A = |W|2 , M = U1 S((−∞,+∞)). Arguing (for some
0 > 0 in Lemma 4.3) as in the proof of Proposition 4.1, we see that
lim n→∞2E(V1 (−t1,n /λ1,n )) = E C and E C < E(W ), imply that J = 1,
l 2

|wn | → 0. Moreover, if
(N−2)/2 (N−2)/2
v0,n = λ1,n z 0,n (λ1,n (x + x1,n )), 
wn = λ1,n wn (λ1,n (x + x1,n )),
t1,n
sn = − 2 ,
λ1,n
662 C.E. Kenig, F. Merle


we have | wn |2 → 0 and v0,n = V1l (sn ) + 
wn , while eit∆ v0,n S((−∞,+∞))
  
≥ δ, |v0,n | < |W| , E(v0,n ) → E C . Note now that |V
2 2 1l (sn ) −
U1 (sn )|2 = o(1) by definition of non-linear profile. We then have


v0,n = U1 (sn ) + wn , 
2
wn → 0.

Moreover, 2 as in the proof of Proposition 4.1, E(U1 (0)) = E C and

|U1 (t)| < |W|2 for all t. We now apply Theorem 2.14, with
0 < 0 (M, A, A , N ) and n large, with  u = U1 , e ≡ 0, t0 = 0, u 0 = v0,n .
This case now follows.
Assume next that (4.10) holds. The first claim is that for j ≥ 2 we
also have lim n→∞ E(V jl (−t j,n /λ2j,n )) < E C . In fact, after passing to a sub-
sequence, assume limn→∞ E(V1l (−t1,n /λ1,n )) < E C . Because of (4.6) we
have
 J 
|z 0,n | ≥
2
|V0, j |2 + o(1),
j=1

and since E C < E(W ), for n large we  have 2E(z 0,n ) ≤ (1− δ0 )E(W ),
by Lemma
 3.4, |z 0,n |2
≤ (1 − δ) |W|  and hence |V 0, j |2
≤
(1 − δ) |W|2 . Similarly, |wn |2 ≤ (1 − δ) |W|2 . By Corollory 3.8,
we have E(V jl (−t j,n /λ2j,n )) ≥ 0, E(wn ) ≥ 0. Also, from (4.4) and the
proof of Corollary 3.13, we have, for n large, that E(V1l (−t1,n /λ21,n )) ≥

C |V0,1 |2 ≥ cα0 = α0 > 0, so that, from (4.7) we obtain, for n large


J
E(z 0,n ) ≥ α0 + E V jl − t j,n /λ2j,n + o(1),
j=2

so that the claim follows from E(z 0,n ) → E C .

We next claim that (after passing to a subsequence so that, for each j,
limn E(V jl (−t j,n /λ2j,n )) exists and limn (−t j,n /λ2j,n ) = s j ∈ [−∞, +∞] ex-
ists) if U j is the non-linear profile associated to (V jl , {−t j,n /λ2j,n }), then U j
verifies (SC). In fact, by definition of non-linear profile,  E(U j ) < E C , since
limn E(V j (−t j,n /λ j,n )) < E C . Moreover, since |V jl (−t j,n /λ2j,n )|2 ≤
l 2

(1 − δ) |W|2 , by the definition ofnon-linear profile  and 2Theorem 3.9, if
t ∈ I j , the maximal interval for U j , |U j (t)| < |W| so that, by the
2

definition of E C our claim follows. Note that the argument in the proof of
Theorem 2.14 also gives that U j W((−∞,+∞)) < +∞.
Our final claim is that there exists j0 so that, for j ≥ j0 we have

 2(N+2)/(N−2)
(4.17)  
U j S((−∞,+∞)) ≤ C |V0, j |2
N+2/N−2
.
Energy-critical focusing NLS 663

 
In fact, from (4.6), for fixed J we see that (choosing n large) Jj=1 |V0, j |2
  
≤ |z 0,n |2 +o(1) ≤ 2 |W|2 . Thus, for j ≥ j0 , we have |V0, j |2 ≤  δ,

where δ is so small that e V0, j S((−∞,+∞)) ≤ δ, with δ as in Theorem 2.5.
it∆

From Remark 2.13 it then follows that U j S((−∞,+∞)) ≤ 2δ, and using
the integral equation in Remark 2.13, that U j (0) Ḣ 1 ≤ CV0, j  Ḣ 1 and
U j W((−∞,+∞)) ≤ CV0, j  Ḣ 1 , which gives (4.17).
For 0 > 0, to be chosen, define now
J(0 )
 
 1 x − x j,n t − t j,n
(4.18) Hn,0 = (N−2)/2
Uj , 2 .
j=1 λ j,n λ j,n λ j,n

We then have:

(4.19) Hn,0 S((−∞,+∞)) ≤ C0 ,

uniformly in 0 , for n ≥ n(0 ). In fact,

⎡  ⎤ 2(N+2)
 J(0 )

N−2
  1 x − x j,n t − t j,n ⎦
 Hn, 2(N+2)/(N−2) = ⎣ Uj , 2
(N−2)/2 λ j,n
0 S((−∞,+∞))
j=1 λ j,n
λ j,n

J(0 )  
  2(N+2)
 x − x j,n t − t j,n
1
N−2

≤ (N−2)/2
U j ,
j=1 λ j,n λ j,n λ2j,n
 
  1 x − x j,n t − t j,n
+ C J(0) (N−2)/2 j
U , 2
 λ j,n λ j,n λ j,n
j = j
  N+6
x − x j  ,n t − t j  ,n
N−2
1
(N−2)/2
U j , 2 = I + II.
λ j ,n λ j ,n λ j  ,n

For n large, II → 0, by the orthogonality of (λ j,n ; x j,n ; t j,n ) (see Ker-

aani [15], Lemma 2.7, (2.95), (2.96), etc.) Hence, for n large we have
II ≤ I. But (with j0 as in (4.17)),

 J(0 )
j0
 2(N+2)/(N−2)   2(N+2)/(N−2)
I≤  
U j S((−∞,+∞)) + U j 
S((−∞,+∞))
j=1 j= j0

  2(N+2)/(N−2)
j0 J(0 ) 

≤  
U j S((−∞,+∞)) + C |V0, j |2
N+2/(N−2)
≤ C0 /2
j=1 j= j0

because of (4.6).
664 C.E. Kenig, F. Merle

For 0 > 0, to be chosen, define

(4.20)
J(0 )
  4
4  1 x − x j,n t − t j,n
N−2

Rn,0 = |Hn,0 | N−2 Hn,0 − U

(N−2)/2 j
, 2
j=1 λ j,n
λ j,n λ j,n
 
1 x − x j,n t − t j,n
× (N−2)/2 U j , 2 .
λ j,n λ j,n λ j,n

We then have

(4.21) For n = n(0 ) large, Rn,0  L 2 L 2N/N+2

x
→ 0 as n → ∞.
t

This follows from the orthogonality of (λ j,n ; x j,n ; t j,n ), the fact that
U j S((−∞,+∞)) < ∞, U j W((−∞,+∞)) < ∞, and arguments of Ker-
aani [15] (see in particular (2.95), (2.96)).
We now will apply Theorem 2.14. Let  u = Hn,0 , e = Rn,0 , where 0
 0) 1 l x−x j,n −t j,n
is still to be determined. Recall that z 0,n = J(
j=1 (N−2)/2 V j λ j,n
, λ2
λ j,n j,n

+ wn , where eit∆ wn S((−∞,+∞)) ≤ 0 . By the definition of non-linear

profile, we now have

(4.22) z 0,n (x) = Hn,0 (x, 0) + 

wn (x),

where, for n large eit∆  wn S((−∞,+∞)) ≤ 20 .

Notice also that, because of the orthogonality of (λ j,n  ; x j,n ; t j,n ), 2for
n = n(0 ) large, we have (using also Corollary 3.13), that |Hn,0 (t)| ≤
 J(0)   t−t j,n   0)   J(0) 
2 j=1 E 0 U j λ2 ≤ 4C J( j=1 |V0, j |2 , and j=1 |V0, j |2
  j,n 
≤ |z 0,n |2 |z0,n |2 + o(1) ≤ 2 |W| 
2
. Let now M = C0 , with C0 as

in (4.19), A = C |W| , A = A + |W|2 , 0 < 0 (M, A, A , N )/2,
2 

where 0 (M, A, A , N ) is as in Theorem 2.14. Fix 0 and choose n so large

that Rn,0  L 2 L 2N/N+2 < 0 and so that all the above properties hold. Then
t x
Theorem 2.14 gives the conclusion of Lemma 4.9 in the case when (4.10)
holds.

Remark 4.23. Assume that {z 0,n } in Lemma 4.3 are all radial. Then V0, j , wn
can be chosen to be radial and we can choose x j,n ≡ 0. This follows directly
from Keraani’s proof [15]. If we then define (SC) and E C by restricting only
to radial functions, we obtain a u C as in Proposition 4.1 which is radial, and
we can establish Proposition 4.2 with x(t) ≡ 0.
Energy-critical focusing NLS 665

5. Rigidity theorem

In this section we will prove the following:

Theorem 5.1. Assume that u 0 ∈ Ḣ 1 is such that

 
E(u 0 ) < E(W ), |u 0 | <
2
|W|2 .

Let u be the solution of (CP) with u|t=0 = u 0 , with maximal interval of

existence (−T− (u 0 ), T+ (u 0 )) (see Definition 2.10). Assume that there exists
λ(t) > 0, for t ∈ [0, T+ (u 0 )), with the property that

   
1 x
K = v(x, t) = u ,t : t ∈ [0, T+ (u 0 ))
λ(t)(N−2)/2 λ(t)

is such that K is compact in Ḣ 1 . Then T+ (u 0 ) = +∞, u 0 ≡ 0.

Remark 5.2. We conjecture that Theorem 5.1 remains true if v(x, t) =

1
λ(t)(N−2)/2
u x−x(t)
λ(t)
, t , with x(t) ∈ R N , t ∈ [0, T+ (u 0 )). In other words, for
“energy subcritical” initial data, compactness up to the invariances of the
equation, for solutions, is only true for u ≡ 0.

We start out with a special case of the strengthened form of Theorem 5.1,
namely:

Proposition 5.3. Assume that u, v, λ(t), x(t) are as in Remark 5.2, that
|x(t)| ≤ C0 and that λ(t) ≥ A0 > 0. Then the conclusion of Theo-
rem 5.1 holds. Moreover, if T+ (u 0 ) < +∞, the hypothesis |x(t)| ≤ C0
is not needed.

Remark 5.4. Because of the continuity of u(t) in Ḣ 1 , it is clear that in

proving Proposition 5.3 we can assume that λ(t), x(t) ∈ C ∞ ([0, T+ (u 0 )))
and that λ(t) > 0 for each t ≥ 0. Indeed, first by the compactness of
K and the theory of (CP), we construct piecewise contant (with small
jumps) λ1 (t), x1 (t) such that the corresponding set K 1 is included in K̃ 1 =
{w(t) solution of (CP) with initial data in K , t ∈ [0, t0 ]}, t0 small. Then
we can contruct regular λ2 (t), x2 (t) such that K 2 is included in the precom-
− (N−2)
pact set λ0 2 w((x − x0 )λ−1 0 ), for w ∈ K̃ 1 , 1/2 ≤ λ0 ≤ 2, |x 0 | ≤ 1 .
The continuity of λ(t), x(t) will not be used in our proof.
666 C.E. Kenig, F. Merle

In the next lemma we will collect some useful facts:

Lemma 5.5. Let u, v be as in Proposition 5.3.
i) Let δ0 > 0 be such that E(u 0 ) ≤ (1 − δ0 )E(W ). Then for all t ∈
[0, T+ (u 0 )), we have
 
|u(t)| ≤ (1 − δ) |W|2
2

 
2∗
|u(t)| − |u(t)| ≥ δ |u(t)|2
2

 
C1,δ0 |u 0 | ≤ E(u 0 ) ≤ C2 |u 0 |2
2

E(u(t)) = E(u 0 )
  
C1,δ0 |u 0 |2 ≤ |u(t)|2 ≤ C2 |u 0 |2 .

ii)
 
|v(t)| ≤ C22
|W|2


v(t)2L 2∗ ≤ C3 |W|2 .
x

iii) For all x0 ∈ R N

 
|v(x, t)|2
≤ C4 |W|2 .
|x − x0 |2

vi) For each  > 0, there exists R(0 ) > 0, such that, for 0 ≤ t < T+ (u 0 ),
we have

∗ |v|2
|v|2 + |v|2 + ≤ 0 .
|x|≥R(0 ) |x|2

Proof. i) follows from Theorem 3.9 and Corollary 3.13. ii) follows from
i) by Sobolev embedding, while iii) follows from i) by Hardy’s inequality.
iv) follows (using Sobolev embedding and the Hardy inequality) from the
compactness of K.

The next lemma is a localized virial identity, in the spirit of Merle [17],
Lemma 3.6.
Energy-critical focusing NLS 667

Lemma 5.6. Let ϕ ∈ C0∞ (R N ), t ∈ [0, T+ (u 0 )). Then:

i)
 
d
|u|2 ϕ dx = 2Im uuϕ dx
dt
ii)
  
d2
|u| ϕ dx = 4
2
Re ∂ xl ∂ x j ϕ · ∂ xl u · ∂ x j u
dt 2

l, j

4 ∗
− ∆ ϕ |u| −
2 2
∆ϕ |u|2 .
N
The proof of Lemma 5.6 is standard, see [17] and Glassey [12].
Proof of Proposition 5.3. The proof splits in two cases, the finite time
blow-up case for u and the infinite time of existence for u.
Case 1: T+ (u 0 ) < ∞. (In this case we don’t need the assumption |x(t)| < C0
or the energy constraints on u, only supt∈[0,T+(u 0 )) |u(t)|2 < ∞ is needed.
Note that this rules out the existence of self similar solutions in Ḣ 1 , i.e.
solutions for which λ(t) ∼ (T − t)−1/2 .)
Note first that λ(t) → ∞ as t → T+ (u 0 ). If not, there exists ti ↑ T+ (u 0 ),
i)
with λ(ti ) → λ0 < +∞. Let vi (x) = λ(t )(N−2)/2
1
u x−x(t
λ(ti )
, ti and let v(x) ∈
i
Ḣ 1 be such that vi → v in Ḣ 1 (from the compactness of K ). Hence,
x(ti ) (N−2)/2
u x − λ(t i)
, ti = λ(ti )(N−2)/2 vi (λ(ti )x) → λ0 v(λ0 x) in Ḣ 1 (since
λ(ti ) ≥ A0 , λ0 ≥ A0 ). Let now h(x, t) be the solution of (CP), given
(N−2)/2
by Remark 2.8 with data λ0 v(λ0 x) at time T+ (u 0 ), in an interval
(T+ (u 0 ) − δ, T+ (u 0 ) + δ), with hS((T+(u 0 )−δ,T+ (u 0 )+δ)) < ∞. Let h i (x, t)
x(ti )
be the solution with data at T+ (u 0 ) equal to u x − λ(t i)
, ti . Then, the (CP)
theory guarantees that
sup h i S((T+(u 0 )− δ ,T+ (u 0 )+ δ )) < ∞.
2 2
i
x(ti )
But, u x − λ(t i)
, t + ti − T+ (u 0 ) = h i (x, t), contradicting Lemma 2.11,
since T+ (u 0 ) < ∞.
Let
∗
us prove now a decay result for u from the concentration properties
in L 2 of u at T+ (u 0 ). Let us now fix ϕ ∈ C0∞ (R N ), ϕ radial, ϕ ≡ 1 for
|x| ≤ 1, ϕ ≡ 0 for |x| ≥ 2 and set ϕ R (x) = ϕ(x/R). Define

(5.7) y R (t) = |u(x, t)|2 ϕ R (x) dx, t ∈ [0, T+ (u 0 )).

We then have:

(5.8) |yR | ≤ CN |W|2 .
668 C.E. Kenig, F. Merle

In fact, by Lemma 4.6, i)


2
|yR | ≤ Im uuϕ(x/R) dx
R
 1/2  1/2 
|u|2
≤ CN |u|2
· ≤ C N |W|2 ,
|x|2
by ii) in Lemma 5.5.
We also have:
 
(5.9) For all R > 0, |u(x, t)|2 dx → 0 as t → T+ (u 0 ).
|x|<R

In fact, u(y, t) = λ(t)(N−2)/2 v(λ(t)y + x(t), t), so that

 
−2
|u(x, t)| dx = λ(t)
2
|v(y + x(t), t)|2 dy
|x|<R |y|<Rλ(t)

= λ(t)−2 |v(z, t)|2 dz
B(x(t),Rλ(t))

−2
= λ(t) |v(z, t)|2 dz
B(x(t),Rλ(t))∩B(0,Rλ(t))

+ λ(t)−2 |v(z, t)|2 dz
B(x(t),Rλ(t))\B(0,Rλ(t))
= A + B.
2 
By Hölder’s inequality, A ≤ λ(t)−2 (Rλ(t)) N N v2L 2∗ ≤ 2 R2 C3 |W|2 ,
which is small with .
2
B ≤ λ(t)−2 (Rλ(t)) N N v2L 2∗ (|x|≥Rλ(t))
= R2 v2L 2∗ (|x|≥Rλ(t)) → 0 as t → T+ (u 0 ),
by iv) in Lemma 5.5, since λ(t) → +∞ as t → T+ (u 0 ).
From (5.9) and (5.8), we have:
 
y R (0) ≤ y R (T+ (u 0 )) + C N T+ (u 0 ) |W| = C N T+ (u 0 ) |W|2 .
2

Thus, letting R → +∞ we obtain

u 0 ∈ L 2 (R N ).

Arguing as before, |y R (t) − y R(T+ (u 0 ))| ≤ C N (T+ (u 0 ) − t) |W|2 so
that y R (t) ≤ C N (T+ (u 0 ) − t) |W|2 . Letting R → ∞, we see that
u(t)2L 2 ≤ C N (T+ (u 0 ) − t) |W|2 and so by the conservation of the
L 2 norm u(T+ (u 0 )) L 2 = u 0  L 2 = 0. But then u ≡ 0, contradicting
T+ (u 0 ) < ∞.
Energy-critical focusing NLS 669

Case 2: T+ (u 0 ) = +∞.
In this case we assume, in addition, that |x(t)| ≤ C0 . We first note that
(5.10) For each  > 0, there exists R() > 0 such that,
for all t ∈ [0, ∞), we have:

∗ |u|2
|u|2 + |u|2 + 2 ≤ .
|x|>R() |x|

In fact, u(y, t) = λ(t)(N−2)/2 v(λ(t)y + x(t), t), so that

 
|u(y, t)|2 dy = λ(t) N |v(λ(t)y + x(t), t)|2 dy
|y|>R()
|y|>R()
= |v(z + x(t), t)|2 dz
|z|>R()λ(t)
≤ |v(z + x(t), t)|2 dz
|z|≥R() A0

≤ |v(α, t)|2 dα,
|α|≥R() A0 −C0

and the statement for this term now follows from Lemma 5.5 iv). The other
terms are handled similarly.
(5.11) There exists R0 > 0 such that, for all t ∈ [0, +∞), we have
  
2∗
8 |u| − 8
2
|u| ≥ Cδ0 |u 0 |2 .
|x|≤R0 |x|≤R0
  ∗
In fact,
 (3.11) combined with Lemma 5.5 i) yields 8 |u| 
2
− 8 |u|2 ≥
δ0 |u 0 |2 . Now combine this with (5.10), with  = 0 |u 0 |2 to obtain
C
(5.11).
To prove Case 2, we choose ϕ ∈ C0∞ (R N ), radial,
 with ϕ(x) = |x|2 for
|x| ≤ 1, ϕ(x) ≡ 0 for |x| ≥ 2. Define z R (t) = |u(x, t)|2 R2 ϕ Rx dx. We
then have:


for t > 0, |z R (t)| ≤ C N,δ0 |u 0 |2 R2


for R large enough, t > 0, z R ≥ C N,δ0 |u 0 |2 .

In fact, from Lemma 5.6, i),

 x 
 |x|
z R (t) ≤ 2R Im uuϕ dx ≤ C N R |u| |u|
R 0≤|x|≤2R |x|
 1/2  1/2 
|u|2
≤ CN R 2
|u| 2
≤ CN R 2
|u 0 |2 ,
|x|2
670 C.E. Kenig, F. Merle

because of Lemma 5.5 i), while from Lemma 5.6, ii),

  x   x  |u|2

z R (t) = 4 Re ∂xl ∂x j ϕ ∂xl u · ∂x j u − ∆2 ϕ
l, j
R R R2
  
4 2∗ 2∗
− ∆ϕ |u| ≥ 8 |u| − |u|
2
N |x|≤R
   
|u|2 2∗
− CN |u| + 2 + |u|
2
≥ C N,δ0 |u 0 |2
R≤|x|≤2R |x|
for R large, in view of (5.11) and (5.10). 
If we now integrate in t, we have z R(t) − z R (0) ≥ C N,δ0 t |u 0 |2 , but
we alsohave |z R (t) − z R (0)| ≤ 2C N R2 |u 0 |2 , a contradiction for t large,
unless |u 0 |2 = 0.
Proof of Theorem 5.1. (See [19] for similar proof) Assume that u 0  ≡ 0 so
that |u 0 |2 > 0 and because of Lemma 5.5 i) (which is still valid here),
E(u 0 ) ≥ C1,δ0 |u 0 |2 and hence E(u 0 ) > 0. Because of Proposition 5.3,
we only need to treat the case where there exists {tn }∞
n=1 , tn ≥ 0, tn ↑ T+ (u 0 ),
so that
λ(tn ) → 0.

(If tn → t0 ∈ [0, T+ (u 0 )), we obtain for all R > 0, |x|≥R |v(t0 )|2∗ = 0 but

|v(t0 )|2 > 0). After possibly redefining {tn }∞
n=1 we can assume that

λ(tn ) ≤ 2inft∈[0,tn ] λ(t)

and from our hypothesis
 
1 x
wn (x) = u , tn → w0 in Ḣ 1 .
λ(tn )(N−2)/2 λ(tn )

By Theorem
 3.9 we have
 E(W ) > E(w 0 ) = E(u 0 ) > 0, |u(t)|2 ≤
(1 − δ) |W| so that |w0 | < |W| . Thus w0  ≡ 0. Let us now
2 2 2

consider solutions of (CP), wn (x, τ), w0 (x, τ) with data wn (−, 0), w0 (−, 0)
at τ = 0, defined in maximal intervals τ ∈ (−T− (wn ), 0], τ ∈ (−T− (w0 ), 0]
respectively.
Since wn → w0 in Ḣ 1 , lim n→∞ T− (wn ) ≥ T− (w0 ) and

for each τ ∈ (−T− (w0 ), 0], wn (x, τ) → w0 (x, τ) in Ḣ 1 .

(See Remark 2.17)

Note that by uniqueness in (CP) (see Definition 2.10), for 0 ≤ tn + τ/λ(tn )2 ,

wn (x, τ) = λ(tn )(N−2)/2
1
u λ(txn ) , tn + λ(tτn )2 . Remark that lim n→∞ τn =
lim n→∞ tn λ(tn )2 ≥ T− (w0 ) and thus for all τ ∈ (−T− (w0 ), 0] for n large,
Energy-critical focusing NLS 671

0 ≤ tn + τ/λ(tn )2 ≤ tn . Indeed,if τn → τ0 < T− (w0 ), then wn (x, −τn ) =

1
u ( x ) → w0 (x, −τ0 ) in Ḣ 1 with λ(tn ) → 0 which is a contra-
λ(tn )(N−2)/2 0 λ(tn )
diction from u 0  ≡ 0, w0  ≡ 0.
Fix now τ ∈ (−T− (w0 ), 0], for n sufficiently large v(x, tn + τ/λ(tn )2 ),
λ(tn + τ/λ(tn )2 ) are defined and we have

(5.12) v(x, tn + τ/λ(tn )2 )

 
1 x
= u , tn + τ/λ(tn )2
λ(tn + τ/λ(tn )2 )(N−2)/2 λ(tn + τ/λ(tn )2 )
 
1 x
= wn ,τ ,
 N−2
λn (τ) 2 
λn (τ)

with

 λ(tn + τ/λ(tn )2 ) 1
(5.13) λn (τ) = ≥
λ(tn ) 2

(because of the fact λ(tn ) ≤ 2in f t∈[0,tn ] λ(t).) One can assume after passing
to a subsequence that  λn (tn + τ/λ(tn )2 ) →  λ0 (τ) with 12 ≤ 
λ0 (τ) ≤ +∞
and v(x, tn + τ/λ(tn ) ) → v0 (x, τ) in Ḣ , as n → ∞. Remark that 
2 1
λ0 (τ) <
+∞. If not, we will have  (N−2)/2 1
w0  x , τ → v0 (x, τ) which implies
λn (τ) λn (τ)
w0 (x, τ) = 0 which contradicts E(w0 ) = E(u 0 ) > 0. Thus  λ0 (τ) < +∞
and v0 (x, τ) =  (N−2)/21
w0  x , τ where v0 (τ) ∈ K . We thus obtain
λ0 (τ) λ0 (τ)
a contradiction from Proposition 5.3. Note that the same proof applies in
the nonradial situation with the extra parameter x(tn ).
 
Corollary 5.14. Assume that E(u 0 ) < E(W ), |u 0 |2 < |W|2 and u 0
is radial. Then the solution u of the Cauchy problem (CP) with data u 0 at
t = 0 has time interval of existence I = (−∞, +∞), uS((−∞,+∞)) < +∞
and there exists u 0,+ , u 0,− in Ḣ 1 such that
   
lim u(t) − eit∆ u 0,+  Ḣ 1 = 0, lim u(t) − eit∆ u 0,−  Ḣ 1 = 0.
t→+∞ t→−∞

Moreover, if we define δ0 so that E(u 0 ) ≤ (1 − δ0 )E(W ), there exists

a function M(δ0 ) so that

uS((−∞,+∞)) ≤ M(δ0 ).

Proof. From the integral equation in Theorem 2.5, it is clear that u(t) is
radial for each t ∈ I . Using Remark (4.23) and Theorem 5.1 we obtain (SC)
or I = (−∞, +∞), uS((−∞,+∞)) < +∞. Now Remark 2.15 finishes the
proof of the first statement.
672 C.E. Kenig, F. Merle

For the last statement, let

  
Dδ0 = u 0 ∈ Ḣ radial,
1
|u 0 | < |W|2 and
2


E(u 0 ) ≤ (1 − δ0 )E(W )

M(δ0 ) = supu∈Dδ0 u S((−∞,+∞)) .

We need to show M(δ0 ) < +∞. If not there is a sequence u 0,n in Dδ0
and the corresponding solutions u n such that u n S((−∞,+∞)) → +∞ as
n → +∞. Note that we can assume that eit∆ u 0,n S((−∞,+∞)) ≥ δ, with δ
as in Theorem 2.5. Arguing as in the proof of Proposition 4.1, we would
conclude that first J = 1 in the decomposition given in Lemma 4.3 and
then since U1 S((−∞,+∞)) < +∞ we reach a contradiction (See also [15],
Corollary 1.14).
Remark 5.15. Note that Corollary 5.14 is sharp. In fact, W(x) is radial
and clearly WS((−∞,+∞)) = +∞. Moreover,
 Remark
 3.14 shows that if
u 0 ∈ H 1 radial, E(u 0 ) < E(W ), but |u 0 |2 > |W|2 , we have that I ,
the maximal interval of existence, is finite.
Let us remark that we have, in fact, proved a slightly stronger result:
Corollary 5.16. Let u 0 ∈ Ḣ 1 be radially symetric
 and assume that for all
t ∈ (−T− (u 0 ), T+ (u 0 )) we have |u(t)|2 ≤ |W|2 − δ0 , for δ0 > 0.
Then the solution u of the Cauchy problem (CP) with data u 0 at t = 0 has
time interval of existence I = (−∞, +∞), uS((−∞,+∞)) < +∞.
Proof. Note first that if E(u 0 ) < E(W ), Corollary 5.14 yields the result, so
that we can assume that E(u 0 ) ≥ E(W ). Observe that the end of the proof
of Lemma 3.4 gives us that, for t ∈ (−T− (u 0 ), T+ (u 0 )),
  
∗
(5.17) |u(t)|2 − |u(t)|2 ≥ Cδ0 |u(t)|2

and that energy conservation, the assumption that E(u 0 ) ≥ E(W ) yields

inft∈(−T− (u 0 ),T+ (u 0 )) |u(t)|2 ≥ C.

From the Remark 2.7, ifδo is close to |W|2 , our conclusion holds.
 0 ≤2δc < |W| , so that if for t ∈ (−T− (u 0 ), T+ (u 0 )),
2
We can then find
|u(t)| ≤ |W| − δ0 , δ0 > δc , our desired conclusion holds and δc is
2

optimal with this property. We assume δc > 0 and reach a contradiction by

establishing the analogs of Propositions 4.1, 4.2 and using (5.17). In order
to do so, we just need to modify
 the statement
 of Lemma 4.9, by replacing
(4.10), say, by for t ∈ I , |U1 (t)|2 ≤ |W|2 − δc , where U1 (t) is the
non-linear profile in Lemma 4.9 and I its maximal interval of existence as in
Remark 2.13. (4.11) is replaced analogously. The proofs are then identical
to those given in Sect. 4 and that of Corollary 5.14.
Energy-critical focusing NLS 673

As a consequence of Corollaries 5.14 and 5.16, we obtain the following

concentration phenomenon for all radial type II finite blow-up solutions.
Here by type II finite blow-up solution, we mean a solution u whose maximal
interval of existence I is finite and for which there is a C, such that for all
t ∈ I , |u(t)|2 < C. On the other hand, type I finite blow-up solution is
such that what the time of existence is finite but the Ḣ 1 norm blows up.
Corollary 5.18. Let u 0 ∈ Ḣ 1 (no size restrictions) be radially
 symetric and
assume that T+ (u 0 ) < +∞, and that ∀t ∈ [0, T+ (u 0 )), |u(t)|2 ≤ C0 .
Then, for all R > 0, we have
 
2
lim t→T+ (u 0 ) |u(t)| ≥
2
|W|2
|x|≤R N
 
lim t→T+ (u 0 ) |u(t)| ≥ |W|2 .
2
|x|≤R

Proof. Consider tn → T+ (u 0 ) and apply Lemma 4.3 to the sequence u(tn ).

Arguing in an analogous manner to the proof of Theorem 2.14, we must
have λ j,n → +∞ for some j and the corresponding non-linear profile U j
has U j S((0,T+(U j )) = +∞. If the first inequality does not hold, we can find
a sequence {tn } as before and R0 > 0, η0 > 0 so that
 
2
|u(tn )| ≤
2
|W|2 − η0 .
|x|≤R0 N
In addition,we must have (since λ j,n → +∞) that
   
2 2 2
U j − t j,n /λ j,n ≤
2
|W| − η0 <
2
|W| < |W|2 .
2
N N

Thus E(U j ) < E(W ) = N1 |W|2 and Corollary 5.14 gives a contradic-
tion.
 inequality doesnot hold, we can find R0 > 0, η0 > 0 so that
If the second
for all t ∈ I , |x|≤R0 |u(t)|2 ≤ |W|2 −η0 . By the argument at the begin-
ing of the proof of case 1 of Proposition 5.13, we must have −t j,n /λ2j,n < C.
 
Thus, we obtain, for t > M, that |U j (t)|2 ≤ |W|2 − η0 , so that
Corollary 5.16 concludes the proof.
Remark 5.19. Note that we have not yet shown that u 0 as in Lemma 5.18
exist, but we expect that this is the case. We also expect to have data u 0 for
which type I blow-up occurs.
Remark 5.20. In the case N ≥ 4, consider now u 0 ∈ H 1 radial as in
Corollary 5.18 (but not type II), then using the L 2 conservation
 and energy
laws, estimates as in [20] yield for
 any sequence tn such that |u(tn )|2 →
+∞ that for all R > 0, we have |x|≤R |u(tn )|2 → +∞ which leads to the
same conclusions as in Corollary 5.18. Note that when N = 3, one expects
674 C.E. Kenig, F. Merle

that the conclusion in this remark is false in light of examples analogous

to the ones constructed by Raphael in [23] which give a radial solution
blowing up exactly on a sphere.

References
1. Aubin, T.: Équations différentielles non linéaires et problème de Yamabe concernant
la courbure scalaire. J. Math. Pures Appl., IX. Sér. 55, 269–296 (1976)
2. Bahouri, H., Gérard, P.: High frequency approximation of solutions to critical nonlinear
wave equations. Am. J. Math. 121, 131–175 (1999)
3. Berestycki, H., Cazenave, T.: Instabilité des états stationnaires dans les équations de
Schrödinger et de Klein–Gordon non linéaires. C. R. Acad. Sci., Paris, Sér. I, Math.
293, 489–492 (1981)
4. Bergh, J., Lofstrom, J.: Interpolation Spaces. An Introduction. Grundlehren der Math-
ematischen Wissenschaften, No. 223. Berlin, New York: Springer 1976
5. Bourgain, J.: Global well-posedness of defocusing critical nonlinear Schrödinger equa-
tion in the radial case. J. Am. Math. Soc. 12, 145–171 (1999)
6. Bourgain, J.: New global well-posedness results for nonlinear Schrödinger equations.
AMS Colloquium Publications, 46, 1999
7. Cazenave, T.: Semilinear Schrödinger Equations. Courant Lecture Notes in Mathe-
matics, vol. 10. New York: New York University Courant Institute of Mathematical
Sciences 2003
8. Cazenave, T., Weissler, F.B.: The Cauchy problem for the critical nonlinear Schrödinger
equation in Hs . Nonlinear Anal., Theory Methods Appl. 14, 807–836 (1990)
9. Colliander, J., Keel, M., Staffilani, G., Takaoke, H., Tao, T.: Global well-posedness and
scattering for the energy-critical nonlinear Schrödinger equation in R3 . To appear in
Ann. Math.
10. Gérard, P.: Description du défaut de compacité de l’injection de Sobolev. ESAIM
Control Optim. Calc. Var. 3, 213–233 (1998)
11. Gerard, P., Meyer, Y., Oru, F.: Inégalités de Sobolev précisées, Séminaire sur les Équa-
tions aux Dérivées Partielles, 1996–1997, Exp. No. IV, 11. École Polytech. 1997
12. Glassey, R.T.: On the blowing up of solutions to the Cauchy problem for nonlinear
Schrödinger equations. J. Math. Phys. 18, 1794–1797 (1977)
13. Grillakis, M.G.: On nonlinear Schrödinger equations. Commun. Partial Differ. Equa-
tions 25, 1827–1844 (2000)
14. Keel, M., Tao, T.: Endpoint Strichartz estimates. Am. J. Math. 120, 955–980 (1998)
15. Keraani, S.: On the defect of compactness for the Strichartz estimates of the Schrödinger
equations. J. Differ. Equations 175, 353–392 (2001)
16. Keraani, S.: On the blow up phenomenon of the critical Schrödinger equation. J. Funct.
Anal. 235, 171–192 (2006)
17. Merle, F.: Determination of blow-up solutions with minimal mass for nonlinear
Schrödinger equations with critical power. Duke Math. J. 69, 427–454 (1993)
18. Merle, F.: On uniqueness and continuation properties after blow-up time of self-similar
solutions of nonlinear Schrödinger equation with critical exponent and critical mass.
Comm. Pure Appl. Math. 45, 203–254 (1992)
19. Merle, F.: Existence of blow-up solutions in the energy space for the critical generalized
KdV equation. J. Am. Math. Soc. 14, 555–578 (2001)
20. Merle, F., Tsutsumi, Y.: L2 concentration of blow-up solutions for the nonlinear
Schrödinger equation with critical power nonlinearity. J. Differ. Equations 84, 205–
214 (1990)
21. Merle, F., Vega, L.: Compactness at blow-up time for L2 solutions of the critical
nonlinear Schrödinger equation in 2D. Int. Math. Res. Not. 8, 399–425 (1998)
22. Ogawa, T., Tsutsumi, Y.: Blow-up of H1 solution for the nonlinear Schrödinger equation.
J. Differ. Equations 92, 317–330 (1991)
Energy-critical focusing NLS 675

23. Raphael, P.: Existence and stability of a solution blowing-up on a sphere for a L2
supercritical nonlinear Schrödinger equation. Duke Math. J. 134(2), 199–258 (2006)
24. Ryckman, E., Visan, M.: Global well-posedness and scattering for the defocusing
energy-critical nonlinear Schrödinger equation in R1+4 . To appear in Amer. J. Math.
Preprint, http://arkiv.org/abs/math.AP/0501462
25. Talenti, G.: Best constant in Sobolev inequality. Ann. Mat. Pura Appl. 110, 353–372
(1976)
26. Tao, T.: Global well-posedness and scattering for the higher-dimensional energy-critical
nonlinear Schrödinger equation for radial data. New York J. Math. 11, 57–80 (2005)
(electronic)
27. Tao, T., Visan, M.: Stability of energy-critical nonlinear Schrödinger equations in high
dimensions. Electron. J. Differ. Equ. 118, 28 (2005) (electronic)
28. Visan, M.: The defocusing energy-critical nonlinear Schrödinger equation in higher
dimensions. Preprint, http://arkiv.org/abs/math.AP/0508298
29. Weinstein, M.: Nonlinear Schrödinger equations and sharp interpolation estimates.
Commun. Math. Phys. 87, 567–576 (1982/83)
30. Zhang, J.: Sharp conditions of global existence for nonlinear Schrödinger and Klein–
Gordon equations. Nonlinear Anal., Theory Methods Appl. 48, 191–207 (2002)