LECTURE NOTES I: ON LOCAL AND GLOBAL THEORY FOR

NONLINEAR SCHRÖDINGER EQUATION

NATAŠA PAVLOVIĆ AND NIKOLAOS TZIRAKIS

Abstract. The notes serve as an introduction to the analysis of dispersive

partial differential equations. They are organized as follows:
• Part I focuses on basic theory for local and global analysis of the semi-
linear Schrödinger equation.
• Part II concentrates on basic local and global theory for the Korteveg de
Vries equation.
• Part III gives a review of some recents results on a derivation of nonlinear
dispersive equations from quantum many body systems.

Dislaimer. The notes are prepared as a study tool for participants of the
MSRI summer school “Dispersive Partial Differential Equations”, June 16-27,
2014. We tried to include many of the relevant references. However it is
inevitable that we had to make sacrifices in the choice of the material that is
included in the notes. As a consequence, there are many important works that
we could not present in the notes.

1. What is a dispersive PDE

Informally speaking, a partial differential equation (PDE) is characterized as

dispersive if, when no boundary conditions are imposed, its wave solutions spread
out in space as they evolve in time. As an example consider the linear homogeneous
Schrödinger equation on the real line
iut + uxx = 0, (1.1)
for a complex valued function u = u(x, t) with x ∈ R and t ∈ R. If we try to find a
solution in the form of a simple wave
u(x, t) = Aei(kx−ωt) ,
we see that it satisfies the equation if and only if
ω = k2 . (1.2)
The relation (1.2) is called the dispersive relation corresponding to the equation
(1.1). It shows that the frequency is a real valued function of the wave number.
If we denote the phase velocity by v = ωk , we can write the solution as u(x, t) =
Aeik(x−v(k)t) and notice that the wave travels with velocity k. Thus the wave
The work of N.P. is supported in part by NSF grant DMS-1101192.
The work of N.T. is supported in part by University of Illinois Research Board Grant RB-14054.
Both authors are thankful to the MSRI staff for all help in organizing the workshop.
1
2 N. PAVLOVIĆ AND N. TZIRAKIS

propagates in such a way that large wave numbers travel faster than smaller ones1.
If we add nonlinear effects and study for example
iut + uxx + |u|p−1 u = 0,
we will see that even the existence of solutions over small times requires delicate
techniques.

Going back to the linear homogenous equation (1.1), let us now consider
Z
u0 (x) = û0 (k)eikx dk.
R
For each fixed k the wave solution becomes
2
u(x, t) = û0 (k)eik(x−kt) = û0 (k)eikx e−ik t .
Summing over k (integrating) we obtain the solution to our problem
Z
2
u(x, t) = û0 (k)eikx−ik t dk.
R

Since |û(k, t)| = |û0 (k)| we have that ku(t)kL2 = ku0 kL2 . Thus the conservation
of the L2 norm (mass conservation or total probability) and the fact that high
frequencies travel faster, leads to the conclusion that not only the solution will
disperse into separate waves but that its amplitude will decay over time. This is
not anymore the case for solutions over compact domains. The dispersion is limited
and for the nonlinear dispersive problems we notice a migration from low to high
frequencies. This fact is captured by zooming more closely in the Sobolev norm
Z 1/2
2 2s
kukH s = |û(k)| (1 + |k|) dk

and observing that it actually grows over time.

Another characterization of dispersive equations comes from the observation that
the space-time Fourier transform (we usually denote by (ξ, τ ) the dual variables of
(x, t)) of their solutions are supported on hyper-surfaces that have non vanishing
Gaussian curvature. For example taking the Fourier transform of the solution of
the linear homogeneous Schrödinger equation
iut + ∆u = 0,
for x ∈ R and t ≥ 0, we obtain that u(ξ, τ ) is supported2 on τ = |ξ|2 .
n

In dispersive equations there is usually a competition between dispersion that

over time smooths out the initial data (in terms of extra regularity and/or in terms
of extra integrability) and the nonlinearity that can cause concentration, blow-up or
even ill-posedness in the Hadamard sense. We focus our attention on the following
two dispersive equations:

1Trying a wave solution of the same form to the heat equation u − u

t xx = 0, we obtain that
the ω is complex valued and the wave solution decays exponential in time. On the other hand the
transport equation ut − ux = 0 and the one dimensional wave equation utt = uxx have traveling
waves with constant velocity.
2In this light the linear wave equation in dimension higher than two is dispersive as the solution
is supported on the cone τ = |ξ|.
 ON LOCAL AND GLOBAL THEORY FOR NLS 3

• Nonlinear Schrödinger (NLS) equation given by

iut + ∆u + f (u) = 0,
where u : Rn × R → C.

• The Korteweg-de Vries equation (KdV) given by

ut + uxxx + uux = 0,
where u : M × R → R with M ∈ {R, T}

as two prime examples. However the methods that are reviewed in these notes apply
equally well to other dispersive PDE. The competition mentioned above comes to
light in a variety of ways. On one hand, we have the case of the NLS (4.2) of
defocusing type with a polynomial nonlinearity of high enough power. In this case
the global energy solution that we will obtain satisfy additional decay estimates that
over time weaken the nonlinear effects. It is then possible to compare the dynamics
of the NLS with the linear problem and show that as t → ∞ the nonlinearity
“disappears” and the solution approaches the free solution. On the other hand, we
have the case of the KdV equation. There the dispersion and the nonlinearity are
balanced in such away that solitary waves (global traveling wave solutions) exist
for all times. These traveling waves are smooth solutions that prevent the equation
from scattering even on the real line. Many different phenomena intertwine with
dispersion but in these notes we can develop and partially answer only the most
basic of questions. For more details the reader can consult [2, 4, 32, 44, 46].
To analyze further the properties of dispersive PDE and outline some recent
developments we start with a concrete example.

2. The semi-linear Schrödinger equation.

Consider the semi-linear Schrödinger equation (NLS) in arbitrary dimensions

iut + ∆u + λ|u|p−1 u = 0, x ∈ Rn , t ∈ R, λ ± 1,

(2.1)
u(x, 0) = u0 (x) ∈ H s (Rn ).
for any 1 < p < ∞. Here H s (Rn ) denotes the s Sobolev space, which is a Banach
space that contains all functions that along with their distributional s-derivatives
belong to L2 (Rn ). This norm is equivalent (through the basic properties of the
Fourier transform) to
Z  12
kf kH s (Rn ) = (1 + |ξ|)2s |fb(ξ)|2 dξ < ∞.
Rn
When λ = −1 the equation is called defocusing and when λ = 1 it is called focusing.
NLS is a basic dispersive model that appears in nonlinear optics and water wave
theory, and it can be derived from quantum many body systems as we shall see in
Part III of the notes.
Before we outline basic properties and questions of interest concerning solutions
to (2.1), we review symmetries of the equation.
4 N. PAVLOVIĆ AND N. TZIRAKIS

2.1. Symmetries of the equation. One of the questions that we shall consider
is the following: for what values of s ∈ R one can expect reasonable solutions? The
symmetries of the equation (2.1) can be very helpful in addressing this question.

(1) A symmetry that we shall often mention is the scaling symmetry, that
can be formulated as follows. Let λ > 0. If u is a solution to (2.1) then
2 x t 2 x
uλ (x, t) = λ− p−1 u( , 2 ), uλ0 = λ− p−1 u0 ( ),
λ λ λ
is a solution to the same equation. If we compute kuλ0 kḢ s we see that

kuλ0 kḢ s = λsc −s ku0 kḢ s

where sc = n2 − p−12
. It is then clear that as λ → ∞:
(a) If s > sc (sub-critical case) the norm of the initial data can be made
small while at the same time the time interval is made longer. This
is the best possible scenario for local well-posedness. Notice that uλ
lives on [0, λ2 T ].
(b) If s = sc (critical case) the norm of the initial data is invariant while
the time interval gets longer. There is still hope in this case, but it
turns out that to provide globally defined solutions one has to work
very hard.
(c) If s < sc (super-critical case) the norms grow as the time interval is
made longer. Scaling works against us in this case; we cannot expect
even locally defined strong solutions, at least in deterministic sense.
(2) Then we have the Galilean Invariance: If u is a solution to (2.1) then
2
eix·v e−it|v| u(x − 2vt, t)

is a solution to the same equation with data eix·v u0 (x).

(3) Other symmetries:
(a) There is also time reversal symmetry. We can thus consider solu-
tions in [0, T ] instead of [−T, T ].
(b) Spatial rotation symmetry which leads to the property that if we
start with radial initial data then we obtain a radially symmetric so-
lution.
(c) Time translation invariance that leads for smooth solutions to the
conservation of energy
Z Z
1 λ
E(u)(t) = |∇u(t)|2 dx − |u(t)|p+1 dx = E(u0 ). (2.2)
2 p+1
(d) Phase rotation symmetry eiθ u that leads to mass conservation

ku(t)kL2 = ku0 kL2 . (2.3)

(e) Space translation invariance that leads to the conservation of the

momentum
Z
p~(t) = = ū∇udx = p~(0). (2.4)
Rn
 ON LOCAL AND GLOBAL THEORY FOR NLS 5

(4) In the case that p = 1+ n4 , we also have the pseudo-conformal symmetry

where if u is a solution to (2.1) then for t 6= 0
1 x 1 i|x|2
n u( , )e 4t
|t| 2 t t
is also a solution. This leads to the pseudo-conformal conservation law
8t2 λ
Z
K(t) = k(x + 2it∇)uk2L2 − |u|p+1 dx = kxu0 k2L2 .
p + 1 Rn

2.2. Questions of interest and relevant notation. We will study NLS and
related equations via considering questions

• of local-in-time nature (local existence of solutions, uniqueness, regularity),

• of global-in-time nature (existence of solutions for large times, finite time
blow-up, scattering).

The standard treatment of the subject is presented in the books of Cazenave [4]
and Tao [46], among others. We will refer to these books, especially the first one,
throughout the notes.

We start by listing some questions of interest:

1. Consider X a Banach space. Starting with initial data u0 ∈ H s (Rn ), we say

that the solution exists locally-in-time, if there exists T > 0 and a subset X of
Ct0 Hxs ([0, T ] × Rn ) such that there exists a unique solution to (1). Note that if
u(x, t) is a solution to (1) then −u(−x, t) is also a solution. Thus we can extend
any solution in Ct0 Hxs ([0, T ] × Rn ) to a solution in Ct0 Hxs ([−T, T ] × Rn ). We also
demand that there is continuity with respect to the initial data in the appropriate
topology.

2. If T can be taken to be arbitrarily large then we say that we have a global

solution.

3. Assume u0 ∈ H s (Rn ) and consider a local solution. If there is a T ? such

that
lim? ku(t)kH s = ∞,
t→T
we say that the solution blows up in finite time. At this point, we can mention
a statement of the so called “blow-up alternative” which is usually proved along
with the local theory. More precisely, the blow-up alternative is a statement that
characterizes the finite time of blow-up, which for example can be done along the
following lines: if (0, T ∗ ) is the maximum interval of existence, then if T ∗ < ∞, we
have limt→T ∗ ku(t)kH s = ∞. Analogous statements can be made for (−T ∗ , 0).

4. As a Corollary to the blow-up alternative one obtains globally defined solu-

tions if there is an a priori bound of the H s norms for all times. Such an a priori
bound is of the form:
sup ku(t)kH s < ∞,
t∈R
6 N. PAVLOVIĆ AND N. TZIRAKIS

and it usually comes from the conservation laws of the equation. For (2.1) this
is usually the case for s = 0, 1. An important comment is in order. Our notion
of global solutions in the point 2. described above does not require that ku(t)kH s
remains uniformly bounded in time. As we said unless s = 0, 1, it is not a triviality
to obtain such a uniform bound. In case that we have quantum scattering, these
uniform bounds are byproducts of the control we obtain on our solutions at infinity.

5. If u0 ∈ H s (Rn ) and we have a well defined local solution, then for each (0, T )
we have that u(t) ∈ Hxs (Rn ). Persistence of regularity refers to the fact that if
we consider u0 ∈ H s1 (Rn ) with s1 > s, then u ∈ X ⊂ Ct0 Hxs1 ([0, T1 ] × Rn ), with
T1 = T . Notice that any H s1 solution is in particular an H s solution and thus
(0, T1 ) ⊂ (0, T ). Persistence of regularity affirms that T1 = T and thus u cannot
blow-up in H s1 before it blows-up in H s both backward and forward in time.

6. Scattering is usually the most difficult problem of the ones mentioned above.
Assume that we have a globally defined solution (which is true for arbitrary large
data in the defocusing case). The problem then is divided into an easier (existence
of the wave operator) and a harder (asymptotic completeness) problem. We will see
shortly that the Lp norms of linear solutions decay in time. This time decay is sug-
gestive that for large values of p the nonlinearity can become negligible as t → ±∞.
Thus we expect that u can be approximated by the solution of the linear equa-
tion. We have to add here that this theory is highly nontrivial for large data. For
small data we can have global solutions and scattering even in the focusing problem.

7. A solution that will satisfy (at least locally) most of these properties will be
called a strong solution. We will give a more precise definition later in the notes.
This is a distinction that is useful as one can usually derive through compactness ar-
guments weak solutions that are not unique. The equipment of the derived (strong)
solutions with the aforementioned properties is of importance. For example the fact
that local H 1 solutions satisfy the energy conservation law is a byproduct not only
of the local-in-time existence but also of the regularity and the continuity with re-
spect to the initial data properties.

8. To make the exposition easier we mainly consider H s solutions where s is an in-

teger. From a mathematical point of view one can investigate solutions that evolve
from rougher and rougher initial data (and thus belong to larger classes of spaces).

3. Local Well-Posedness

When trying to establish existence of local (in time) solutions, an important

step consists of constructing the aforementioned Banach space X. This process
is delicate (the exception being the construction of smooth solutions that is done
classically) and is built upon certain estimates that the linear solution satisfies.
First we recall those estimates.

3.1. Fundamental solution, Dispersive and Strichartz estimates. Recall

(from an undergraduate or graduate PDE course) that we can obtain the solution
to the linear problem by utilizing the Fourier transform. Then for smooth initial
 ON LOCAL AND GLOBAL THEORY FOR NLS 7

data (say in the Schwartz class S(Rn )) the solution of the linear homogeneous
equation is given as the convolution of the data with the tempered distribution
1 i|x|2
i 4t
Kt (x) = n e .
(4πit) 2
Thus we can write the solution as:
Z
1 |x−y|2
u(x, t) = U (t)u0 (x) = eit∆ u0 (x) = Kt ?u0 (x) = n ei 4t u0 (y)dy. (3.1)
(4πit) 2 Rn
Another fact from our undergraduate (or graduate) machinery is Duhamel’s prin-
ciple:
Let I be any time interval and suppose that u ∈ Ct1 S(I × Rn ) and that F ∈
Ct0 S(I × Rn ). Then u solves

x ∈ Rn ,

iut + ∆u = F, t ∈ R,
n (3.2)
u(x, t0 ) = u(t0 ) ∈ S(R )
if and only if
Z t
u(x, t) = ei(t−t0 )∆ u(t0 ) − i ei(t−s)∆ F (s)ds. (3.3)
0

Definition 3.1. Let I be a time interval which contains zero, u0 := u(x, 0) ∈

H s (Rn ) and
F ∈ C(H s (Rn ); H s−2 (Rn )).
We say that
u ∈ C(I; H s (Rn )) ∩ C 1 (I; H s−2 (Rn ))
is a strong solution of (3.2) on I, if it satisfies the equation for all t ∈ I in the
sense of H s−2 (thus as a distribution for low values of s) and u(0) = u0 .
Remark 3.2. By a little semigroup theory this definition of a strong solution is
equivalent to saying that for all t ∈ I, u satisfies (3.3).

Now we state the basic dispersive estimate for solutions to the homogeneous
equation (3.2), with F = 0. From the formula (3.1) we see that:
1
kukL∞
x
≤ n ku0 kL1 .
(4|t|π) 2
2
it|ξ|2
In addition the solution satisfies that û(ξ, t) = e−4π û0 (ξ), which together with
Plancherel’s theorem implies that
ku(t)kL2x = ku0 kL2x .
Riesz-Thorin interpolation Lemma then implies that for any p ≥ 2 and t 6= 0 we
have that
1
ku(t)kLpx ≤ 1 1 ku0 kLp ,
0 (3.4)
(4|t|π)n( 2 − p )
where p0 is the dual exponent of p satisfying p1 + p10 = 1.

Fortunately, the basic dispersive estimates (3.4) can be extended by duality

(using a T T ? argument) to obtain very useful Strichartz estimates, [4, 16, 27, 41].
In order to state Strichartz estimates, first, we recall the definition of an admissible
pair of exponents.
8 N. PAVLOVIĆ AND N. TZIRAKIS

Definition 3.3. Let n ≥ 1. We call a pair (q, r) of exponents admissible if

2 ≤ q, r ≤ ∞
are such that
2 n n
+ = (3.5)
q r 2
and (q, r, n) 6= (2, ∞, 2).

Now we can state the Strichartz estimates:

Theorem 3.4. Let n ≥ 1. Then for any admissible exponents (q, r) and (q̃, r̃) we
have the following estimates:

• The homogeneous estimate:

keit∆ u0 kLqt Lrx (R×Rn ) . ku0 kL2 , (3.6)

• The dual estimate:

Z
k e−it∆ F (·, t) dtkL2x (Rn ) ≤ kF kLq̃ Lr̃ (R×Rn ) (3.7)
t x
R

• The non-homogeneous estimate:

Z t
k ei(t−s)∆ F (·, s) dskLqt Lrx (R×Rn ) . kF kLq̃0 Lr̃0 (R×Rn ) , (3.8)
t x
0

1 1 1 1
where q̃ + q̃ 0 = 1 and r̃ + r̃ 0 = 1.
Remark 3.5. Actually the proof of Strichartz estimates R t implies more. In partic-
ular, the operator eit∆ u0 (x) belongs to C(R, L2x ) and 0 U (t − s)F (s)ds belongs to
¯ L2x ) where t ∈ I is any interval of R.
C(I,

In the lectures and during TA sessions, we will discuss the non-endpoint case.
The end-point case was proved in [27].

3.2. Notion of local well-posedness. We are now ready to give a precise defi-
nition of what we mean by local well-posedness of the initial value problem (IVP)
(2.1).
Definition 3.6. We say that the IVP (2.1) is locally well-posed (lwp) and admits
a strong solution in H s (Rn ) if for any ball B in the space H s (Rn ), there exists a
finite time T and a Banach space X ⊂ L∞ s n
t Hx ([0, T ] × R ) such that for any initial
data u0 ∈ B there exists a unique solution u ∈ X ⊂ Ct0 Hxs ([0, T ] × Rn ) to the
integral equation
Z t
u(x, t) = U (t)u0 + iλ U (t − s)|u|p−1 u(s)ds.
0
Furthermore the map u0 → u(t) is continuous as a map from H s (Rn ) into Ct0 Hxs ([0, T ]×
Rn ). If uniqueness holds in the whole space Ct0 Hxs ([0, T ] × Rn ) then we say that the
lwp is unconditional.
 ON LOCAL AND GLOBAL THEORY FOR NLS 9

In what follows we assume that p − 1 = 2k. This implies that the nonlinearity
is sufficiently smooth to perform all the calculations in a straightforward way.

3.3. Well-posedness for smooth solutions. We start with the H s well-posedness

theory, with an integer s > n2 . For more general statements see [25].
Theorem 3.7. Let s > n2 be an integer. For every u0 ∈ H s (Rn ) there exists T ∗ > 0
and a unique maximal solution u ∈ C((0, T ∗ ); H s (Rn )) that satisfies (2.1) and in
addition satisfies the following properties:
i) If T ∗ < ∞ then ku(t)kH s → ∞ as t → ∞. Moreover lim supt→T ∗ ku(t)kL∞ = ∞.
ii) u depends continuously on the initial data in the following sense. If un,0 → u0
in H s and if un is the corresponding maximal solution with initial data un,0 , then
un → u in L∞ ((0, T ); H s (Rn )) for every interval [0, T ] ⊂ [0, T ∗ ).
iii) In addition, the solution u satisfies conservation of energy (2.2) and conserva-
tion of mass (2.3).
Remark 3.8. A comment about uniqueness. Suppose that one proves existence
and uniqueness in C([−T, T ]; XM ) where XM , M = M (ku0 kX ), T = T (M ), is a
fixed ball in the space X. One can then easily extend the uniqueness to the whole
space X by shrinking time by a fixed amount. Indeed, shrinking time to T 0 we get
existence and uniqueness in a larger ball XM 0 . Now assume that there are two
different solutions one staying in the ball XM and one separating after hitting the
boundary at some time |t| < T 0 . This is already a contradiction by the uniqueness
in XM 0 .

3.3.1. Preliminaries. To prove Theorem 3.7 we need the following two lemmata:
Lemma 3.9. Gronwall’s inequality: Let T > 0, k ∈ L1 (0, T ) with k ≥ 0 a.e. and
two constants C1 , C2 ≥ 0. If ψ ≥ 0, a.e in L1 (0, T ), such that kψ ∈ L1 (0, T )
satisfies
Z t
ψ(t) ≤ C1 + C2 k(s)ψ(s)ds
0
for a.e. t ∈ (0, T ) then,
Z t

ψ(t) ≤ C1 exp C2 k(s)ds .
0

Proof. For a proof, see e.g. Evans [15]. 

Lemma 3.10. Let g(u) = ±|u|2k u and consider and s, l ≥ 0, integers with l ≤ s
and s > n2 . Then
kg(u)kH s . kuk2k+1
Hs , (3.9)
kuk2k kvk2k


kg(u) − g(v)kL2 . Hs + Hs ku − vkL2 , (3.10)
kg (l) (u) − g (l) (v)kL∞ . kuk2k−l 2k−l


H s + kvkH s ku − vkH s , (3.11)
2k 2k


kg(u) − g(v)kH s . kukH s + kvkH s ku − vkH s . (3.12)

n
Proof. To prove (3.9) we use the algebra property of H s for s > 2 and the fact
that kukH s = kūkH s .
10 N. PAVLOVIĆ AND N. TZIRAKIS

To prove (3.10) and (3.11) note that since g is smooth we have that
|g(u) − g(v)| . |u|2k + |v|2k |u − v|,

|g (l) (u) − g (l) (v)| . |u|2k−l + |v|2k−l |u − v|.

Then
kg(u) − g(v)kL2 . kuk2k 2k 2k 2k



L∞ + kvkL∞ ku − vkL2 . kukH s + kvkH s ku − vkL2 ,
2k−l 2k−l
kg (l) (u)−g (l) (v)kL∞ . kukL ku−vkL∞ . kuk2k−l 2k−l



∞ +kvkL∞ H s +kvkH s ku−vkL2 ,
s ∞
where we used the fact that H embeds in L .

To prove (3.12) notice that the L2 part of the left hand side follows from (3.10).
For the derivative part consider a multi-index α with |α| = s. Then Dα u is the
Qk
sum (over k ∈ {1, 2, ..., s}) of terms of the form g (k) (u) j=1 Dβj u where |βj | ≥ 1
Pk
and |α| = |β1 | + ... + |βk |. Now let pj = |β2sj | such that j=1 p1j = 12 . We have by
Hölder’s inequality
k
Y k
Y
(k) βj (k)
kg (u) D ukL2 . kg (u)k L∞ kDβj ukLpj .
j=1 j=1

By complex interpolation (or Gagliardo-Nirenberg inequality) we obtain

|βj | |βj |
1−
kDβj ukLpj . kukHss kukL∞ s

and thus
k
Y
kg (k) (u) Dβj ukL2 . kg (k) (u)kL∞ kukH s kukk−1 2k+1
L∞ . kukH s
j=1

where in the last inequality we used (3.11). Thus we obtain

kDα ukL2 . kuk2k+1
Hs . (3.13)
α
Again notice that the term D (g(u) − g(v)) is the sum of terms of the form
k
Y k
Y k k
(k) βj (k) βj
 (k) (k)
Y βj (k)
Y
g (u) D u−g (v) D v = g (u)−g (v) D u+g (v) Dβj wj
j=1 j=1 j=1 j=1

where wj ’s are equal to u or v except one that is equal to u − v. The second of the
left hand side is estimated as in the proof of (3.13). For the first the same trick
applies but now to estimate kg (k) (u) − g (k) (v)kL∞ we use (3.12). 

3.3.2. A proof of Theorem 3.7. Now we present a proof of Theorem 3.7.

Existence and Uniqueness. We construct solutions by a fixed point argument.

Given M, T > 0 to be chosen later, we set I = (0, T ) and consider the space
E = {u ∈ L∞ (I; H s (Rn )) : kukL∞ (I;H s ) ≤ M },
equipped with the distance
d(u, v) = ku − vkL∞ (I;L2 ) .
We note that (E, d) is a complete metric space.
 ON LOCAL AND GLOBAL THEORY FOR NLS 11

Now based on the equation (2.1), with λ = −1, in the integral form, we introduce
the mapping Φ as follows:
Z t
Φ(u)(t) = eit∆ u0 − i ei(t−τ )∆ |u|2k u(τ ) dτ =: eit∆ u0 + H(u)(t).
0

By Lemma 3.10, Minkowski’s inequality and the fact that eit∆ is an isometry in
H s we have that
kΦ(u)(t)kH s . ku0 kH s + T kg(u)kL∞ (I;H s ) ≤ ku0 kH s + T C(M )M,

where we used the notation g(u) = ±|u|2k u as in Lemma 3.10. Furthermore using
Lemma 3.10 again we have
kΦ(u)(t) − Φ(v)(t)kL2 . T C(M )ku − vkL∞ (I;L2 ) . (3.14)

Therefore we see that if M = 2ku0 kH s and T C(M ) < 21 , then Φ is a contraction

of (E, d) and thus has a unique fixed point. Uniqueness in the full space follows by
the remark above or alternatively by the remark and Gronwall’s Lemma.

Blow-up alternative. Let u0 ∈ H s and define

T ∗ = sup{T > 0 : there exists a solution on [0, T ]}. (3.15)
Now let T ∗ < ∞ and assume that there exists a sequence tj → T ∗ such that
ku(tj )kH s ≤ M . In particular for k such that tk is close to T ∗ we have that
ku(tk )kH s ≤ M . Now we solve our problem with initial data u(tk ) and we extend
our solution to the interval [tk , tk + T (M )]. But if we pick k such that
tk + T (M ) > T ∗
we then contradict the definition of T ∗ . Thus limt→T ∗ ku(t)kH s = ∞ if T ∗ < ∞.

We now show that if T ∗ < ∞ then lim supt→T ∗ ku(t)kL∞ = ∞. Indeed suppose
that lim supt→T ∗ ku(t)kL∞ < ∞. Since u ∈ C([0, T ∗ ); H s ) we have that
M = sup ku(t)kL∞ < ∞
0≤t<T ∗

where we used the fact that H s embeds in L∞ . By Duhamel’s formula and Lemma
3.10 we have that
Z t
ku(t)kH s ≤ ku0 kH s + C(M ) ku(τ )kH s dτ.
0
∗
By Gronwall’s lemma we have that ku(t)kH s ≤ ku0 kH s eT C(M )
for all 0 ≤ t < T ∗ .
But this contradicts the blow-up of ku(t)kH s at T ∗ .

Continuous dependence. Let u0 ∈ H s and consider u0,n ⊂ H s such that un,0 → u0

in H s as n → ∞. Since for n sufficiently large we have that ku0,n kH s ≤ 2ku0 kH s
by the local theory there exists T = T (ku0 kH s ) such that u and un are defined on
[0, T ] for n ≥ N and
kukL∞ ((0,T );H s ) + sup kun kL∞ ((0,T );H s ) ≤ 6ku0 kH s .
n≥N
12 N. PAVLOVIĆ AND N. TZIRAKIS

Now note that un (t)−u(t) = eit∆ (un,0 −u0 )+H(un )(t)−H(u)(t). If we use Lemma
3.10 we see that for all t ∈ (0, T ) and n sufficiently large, there exists C such that
Z t
kun (t) − u(t)kH s ≤ kun,0 − u0 kH s + C kun (τ ) − u(τ )kH s dτ.
0
By Gronwall’s lemma we see that un → u in H s as n → ∞. Iterating this property
to cover any compact subset of (0, T ∗ ) we finish the proof.

As a final note we remark that if we solve the equation, starting from u0 and
u(t1 ) over the intervals [0, t1 ]
and [t1 , t2 ] respectively, by continuous dependence, to
prove that C [0, T ]; H s (Rn ) , it is enough to consider the difference u(t1 ) − u0 in
the H s norm. Since
Z t1
u(t1 ) − u0 = (eit1 ∆ − 1)u0 − i ei(t1 −τ ) g(u)(τ ) dτ,
0
it∆
using again Lemma 3.10 and the fact that e u0 (x) ∈ C(R; H s ) we have
ku(t1 ) − u0 kH s . k(eit1 ∆ − 1)u0 kH s + |t1 |kuk2k+1
L∞ ((0,t1 );H s )
which finishes the proof.

Conservation laws: Since we develop the H 1 theory below we implicitly have s ≥ 2.

We have at hand a solution that satisfies the equation in the classical sense for high
enough s (in general in the H s−2 sense with s ≥ 2 and thus in particular u satisfies
the equation at least in the L2 sense. All integrations below then can be justified
in the Hilbert space L2 ). To obtain the conservation of mass we can multiply the
equation by iū, integrate and then take the real part. To obtain the conservation
of energy we multiply the equation by ūt , take the real part and then integrate.

3.4. Local well-posedness in the H 1 sub-critical case. For more details we

refer to [4, 25, 26].
4
Theorem 3.11. Let 1 < p < 1 + n−2 , if n ≥ 3 and 1 < p < ∞, if n = 1, 2. For
every u0 ∈ H (R ) there exists a unique strong H 1 solution of (2.1) defined on the
1 n

maximal interval (0, T ? ). Moreover

u ∈ Lγloc ((0, T ? ); Wx1,ρ (Rn ))
for every admissible pair (γ, ρ). In addition
lim ku(t)kH 1 = ∞
t→T ?
if T ? < ∞, and u depends continuously on u0 in the following sense: There exists
T > 0 depending on ku0 kH 1 such that if u0,n → u0 in H 1 and un (t) is the cor-
responding solution of (2.1), then un (t) is defined on [0, T ] for n sufficiently large
and
un (t) → u(t) in C([0, T ]; H 1 ) (3.16)
?
for every compact interval [0, T ] of (0, T ). Finally we have that
Z Z
1 λ
E(u)(t) = |∇u(t)|2 dx − |u(t)|p+1 dx = E(u0 )
2 p+1
and
M (u)(t) = ku(t)kL2 = ku0 kL2 = M (u0 ).
 ON LOCAL AND GLOBAL THEORY FOR NLS 13

We note that W 1,ρ is the Sobolev space of Lρ functions with weak derivatives in
ρ
L of order one.

Proof. First we establish:

Existence and Uniqueness. In order to define the space on which we shall apply the
fixed point argument, we pick r to be r := p + 1. Fix M, T > 0 to be chosen later
and let q be such that the pair (q, r) is admissible.3 Consider the set
E = {u ∈ L∞ 1 n q
t Hx ([0, T ] × R ) ∩ L ((0, T ); W
1,r
(Rn )) : (3.17)
kukL∞ 1 ≤ M and kuk q
t ((0,T );H ) L Wx1,r ≤ M }. (3.18)
t

equipped with the distance

d(u, v) = ku − vkLq ((0,T );Lr (Rn )) + ku − vkL∞ ((0,T );L2 (Rn )) .
It can be shown that (E, d) is a complete metric space.
We write the solution map via Duhamel’s formula as follows:
Z t
it∆
Φ(u)(t) = e u0 − i ei(t−τ )∆ |u|p−1 u(τ ) dτ. (3.19)
0

Now we provide a few estimates that we shall use in order to justify that the
mapping Φ is a contraction on (E, d). Notice that for r = p + 1 we have
k|u|p−1 ukLrx0 . kukpLrx
and thus by Hölder
p−1
k|u|p−1 ukLqt Lrx0 . kukL ∞ Lr kukLq Lr .
t x t x
(3.20)
However by Sobolev embedding we have that
kukLrx . kukH 1 ,
which together with (3.20) implies that:
p−1
k|u|p−1 ukLqt Lrx0 . kukL∞ H 1 kukLq Lr .
t x
(3.21)
t x

Similarly, since the nonlinearity is smooth,

p−1
k∇(|u|p−1 u)kLqt Lrx0 . kukL ∞ H 1 k∇ukLq Lr .
t x
(3.22)
t x

Now we combine (3.21) and (3.22) to obtain for u ∈ E:

p−1
k|u|p−1 ukLq W 1,r0 . kukL∞ H 1 kukLq W 1,r
x
(3.23)
t x t x t

Furthermore, applying Hölder’s inequality in time, followed by an application of

(3.23) gives:
q−q 0
k|u|p−1 ukLq0 W 1,r0 . T q0 q k|u|p−1 ukLq W 1,r0
t x t x
q−q 0
p−1
.T q0 q kukL∞ H 1 kukLq W 1,r .
x
(3.24)
t x t

3Since the admissibility condition reads 2 + n = n , and r = p + 1, we have that q = 4(p+1) .

q r 2 n(p−1)
14 N. PAVLOVIĆ AND N. TZIRAKIS

Now we are ready to show that Φ is a contraction on (E, d). Using Duhamel’s
formula (3.19) and Strichartz estimates we obtain:
kΦ(u)(t)kLq Wx1,r . keit∆ u0 kLq Wx1,r + k|u|p−1 ukLq0 W 1,r0
t t t x
q−q 0
. ku0 kH 1 + T q0 q kukp−1
L∞ 1 kukLq W 1,r , (3.25)
t Hx t x

where to obtain (3.25) we used (3.24). Also by Duhamel’s formula (3.19), Strichartz
estimates and (3.24) we have:
q−q 0
kΦ(u)(t)kL∞ 1 . ku0 kH 1 + T
t Hx
q0 q kukp−1
L∞ H 1 kukL Wx .
q 1,r (3.26)
t x t

Hence (3.25) and (3.26) imply:

q−q 0
kΦ(u)(t)kLq Wx1,r + kΦ(u)(t)kL∞ 1 ≤ Cku0 kH 1 + CT
t Hx
q0 q M p−1 kukLq Wx1,r . (3.27)
t t

Now we set M = 2Cku0 kH 1 and then choose T small enough such that
q−q 0 1
CT q0 q M p−1 ≤ .
2
We note that such choice of T is indeed possible thanks to the fact that for p <
4
1 + n−2 we have that q > 2 and thus q > q 0 . For such T ∼ T (ku0 kH 1 ) we have that
kΦ(u)(t)kE ≤ M whenever u ∈ E and thus Φ : E → E. In a similar way, one can
obtain the following estimate on the difference:
kΦ(u)(t) − Φ(v)(t)kLq Wx1,r + kΦ(u)(t) − Φ(v)(t)kL∞ 2
t Lx
t

provides a unique solution u ∈ E. Notice that by the above estimates and the
Strichartz estimates we have that u ∈ Ct0 ((0, T ); H 1 (Rn )).

To extend uniqueness in the full space we assume that we have another solution v
and consider an interval [0, δ] with δ < T . Then as before
α p−1 p−1
ku(t) − v(t)kLq Wx1,r + ku(t) − v(t)kL∞ 1 ≤ Cδ (kuk ∞
δ Hx L H 1 + kvkL∞ H 1 )ku − vkL Wx
q 1,r
δ T x T x δ

But if we set
K = max(kukL∞ 1 + kvkL∞ H 1 ) < ∞
T Hx T x

then for δ small enough we obtain

1
ku(t)−v(t)kLq Wx1,r +ku(t)−v(t)kL∞ 1 ≤
δ Hx
(ku(t)−v(t)kLq Wx1,r +ku(t)−v(t)kL∞ 1)
δ Hx
δ 2 δ

which forces u = v on [0, δ]. To cover the whole [0, T ] we iterate the previous argu-
ment Tδ times.

Membership in the Strichartz space. The fact that

u ∈ Lγloc ((0, T ∗ ); Wx1,ρ (Rn ))
for every admissible pair (γ, ρ), follows from the Strichartz estimates on any com-
pact interval inside (0, T ∗ ).

Blow-up alternative. The proof is the same as in the smooth case.

Continuous dependence can be obtained via establishing estimates on

kun (t) − u(t)kLq Wx1,r + kun (t) − u(t)kL∞ 1.
t Hx
t
 ON LOCAL AND GLOBAL THEORY FOR NLS 15

We skip details and refer the interested reader to [4].

Conservation laws. The proof of the conservation of mass is similar to the smooth
case but now we use the pairing (ut , u)H 1 −H −1 . A proof of conservation of en-
ergy is more involved since we need more derivatives to make sense of the energy
functional. Details can be found in e.g. [4]. 
Remark 3.12. We pause to give a couple of comments:

(1) Notice that when λ = −1 (defocusing case), the mass and energy conserva-
tion provide a global a priori bound
sup ku(t)kH 1 ≤ CM (u0 ),E(u0 ) .
t∈R
By the blow-up alternative we then have that T ∗ = ∞ and the problem is
globally well-posed (gwp).
(2) Let I = [0, T ]. An inspection of the proof reveals that we can run the lwp
argument in the space S 1 (I × Rn ) with the norm
kukS 1 (I×Rn ) = kukS 0 (I×Rn ) + k∇ukS 0 (I×Rn )
where
kukS 0 (I×Rn ) = sup kukLqt∈I Lrx .
(q,r)−admissible

3.5. Well-posedness for the L2 sub-critical problem. We now state the lwp
and gwp theory for the L2 sub-critical problem. The reader can consult e.g. [49]
for details.
Theorem 3.13. Consider 1 < p < 1 + n4 , n ≥ 1 and an admissible pair (q, r) with
p + 1 < q. Then for every u0 ∈ L2 (Rn ) there exists a unique strong solution of
iut + ∆u + λ|u|p−1 u = 0,

(3.28)
u(x, 0) = u0 (x)
defined on the maximal interval (0, T ? ) such that
u ∈ Ct0 ((0, T ? ); L2 (Rn )) ∩ Lqloc ((0, T ? ); Lr (Rn )).
Moreover
u ∈ Lγloc ((0, T ? ); Lρ (Rn ))
for every admissible pair (γ, ρ). In addition
lim ku(t)kL2 = ∞
t→T ?
if T ? < ∞ and u depends continuously on u0 in the following sense: There exists
T > 0 depending on ku0 kL2 such that if u0,n → u0 in L2 and un (t) is the corre-
sponding solution of (3.28), then un (t) is defined on [0, T ] for n sufficiently large
and
un (t) → u(t) in Lγloc ([0, T ]; Lρ (Rn )) (3.29)
for every admissible pair (γ, ρ) and every compact interval [0, T ] of (0, T ? ). Finally
we have that
M (u)(t) = ku(t)kL2 = ku0 kL2 = M (u0 ) and thus T ? = ∞. (3.30)
16 N. PAVLOVIĆ AND N. TZIRAKIS

Remark 3.14. We give a couple of comments:

(1) Notice that global well-posedness follows immediately.

(2) The equation makes sense in H −2 .

Finally we state the L2 -critical lwp theory when p = 1 + n4 , [5]. We should

4
mention that a similar theory holds for the H 1 critical problem (p = 1 + n−2 ), [5].
For dimensions n = 1, 2 the problem is always energy sub-critical.
Theorem 3.15. Consider p = 1 + n4 , n ≥ 1. Then for every u0 ∈ L2 (Rn ) there
exists a unique strong solution of
 4
iut + ∆u + λ|u| n u = 0,
(3.31)
u(x, 0) = u0 (x)
defined on the maximal interval (0, T ? ) such that
u ∈ Ct0 ((0, T ? ); L2 (Rn )) ∩ Lp+1 ?
loc ((0, T ); L
p+1
(Rn )).
Moreover
u ∈ Lγloc ((0, T ? ); Lρ (Rn ))
for every admissible pair (γ, ρ). In addition if T ? < ∞
lim ku(t)kLqloc ((0,T ? );Lr (Rn )) = ∞
t→T ?

for every admissible pair (q, r) with r ≥ p + 1. u also depends continuously on u0

in the following sense: If u0,n → u0 in L2 and un (t) is the corresponding solution
of (3.31), then un (t) is defined on [0, T ] for n sufficiently large and
un (t) → u(t) in Lq ([0, T ]); Lr (Rn )) (3.32)
for every admissible pair (q, r) and every compact interval [0, T ] of (0, T ? ). Finally
we have that
M (u)(t) = ku(t)kL2 = ku0 kL2 = M (u0 ) f or all t ∈ (0, T ? ). (3.33)
Remark 3.16. Again, we give a few comments:

(1) Notice that the blow-up alternative in this case is not in terms of the L2
norm, which is the conserved quantity of the problem. This is because the
problem is critical and the time of local well-posedness depends not only on
the norm but also on the profile of the initial data. On the other hand if
we have a global Strichartz bound on the solution global well-posedness is
guaranteed by the Theorem. We will see later that this global Strichartz
bound is sufficient for proving scattering also.
(2) It is easy to see that if ku0 kL2 < µ, for µ small enough, then by the
Strichartz estimates
keit∆ u0 kLp+1 Lp+1
x (R×Rn ) < Cµ < η.
t

Thus for sufficiently small initial data T ? = ∞ and after only one iteration
we have global well-posedness for the focusing or defocusing problem. In
addition we have that u ∈ Lqt (R; Lrx (Rn )) for every admissible pair (q, r)
and thus we also have scattering for small data. But this is not true for
large data as the following example shows.
 ON LOCAL AND GLOBAL THEORY FOR NLS 17

Consider λ > 0. We know that there exists nontrivial solutions of the

form
u(x, t) = eiωt φ(x)
where φ is a smooth nonzero solution of
−∆φ + ωφ = |φ|p−1 φ
with ω > 0. But
kφkLrx (Rn ) ≤ M
/ Lqt (R; Lrx (Rn )) for any q < ∞.
for every r ≥ 2 and thus u ∈

Although some recent results have appeared for super-critical equations, the the-
ory has been completed only for the defocusing critical problem and those devel-
opments are recent. More precisely, global energy solutions for the 3d defocusing
energy-critical problem with radially symmetric initial data was obtained in [3].
The radially symmetric assumption was removed in [10]. For n ≥ 4 the problem
was solved in [42, 50]. The defocusing mass-critical problem is now solved in all
dimensions in a series of papers, [11, 12, 13].
To obtain global-in-time solutions for the focusing problems, as we have seen,
one needs to assume a bound on the norm of the data. For the energy-critical
focusing problem one can consult the work [28], where a powerful program that
helped settle many critical problems, has been introduced; for higher dimensions
see e.g. [29]. Results concerning the mass-critical focusing problem are obtained in
[14] in all dimensions.

4. Morawetz type inequalities

To study in more details the local or global solutions of the above problems
we have to revisit the symmetries of the equation. We first write down the local
conservation laws or the conservation laws in differentiable form. The differential
form of the conservation law is more flexible and powerful as it can be localized to
any given region of space-time by integrating against a suitable cut-off function or
contracting against a suitable vector fields. One then does not obtain a conserved
quantity but rather a monotone quantity. Thus from a single conservation law one
can generate a variety of useful estimates. We can also use these formulas to study
the blow-up and concentration problems for the focusing NLS and the scattering
problem for the defocusing NLS.
The question of scattering or in general the question of dispersion of the nonlinear
solution is tied to weather there is some sort of decay in a certain norm, such as
the Lp norm for p > 2. In particular knowing the exact rate of decay of various Lp
norms for the linear solutions, it would be ideal to obtain estimates that establish
similar rates of decay for the nonlinear problem. The decay of the linear solutions
can immediately establish weak quantum scattering in the energy space but to esti-
mate the linear and the nonlinear dynamics in the energy norm we usually looking
for the Lp norm of the nonlinear solution to go to zero as t → ∞.
Strichartz type estimates assure us that certain Lp norms going to zero but only
for the linear part of the solution. For the nonlinear part we need to obtain gen-
eral decay estimates on solutions of defocusing equations. The mass and energy
18 N. PAVLOVIĆ AND N. TZIRAKIS

conservation laws establish the boundedness of the L2 and the H 1 norms but are
insufficient to provide a decay for higher powers of Lebesgue norms. In these notes
we provide a summary of recent results that demonstrate a straightforward method
to obtain such estimates by taking advantage of the momentum conservation law
Z Z
= u∇udx = = u0 ∇u0 dx. (4.1)
Rn Rn

Thus we want to establish a priori estimates for the solutions to the power type
nonlinear Schrödinger equation
iut + ∆u = λ|u|p−1 u, x ∈ Rn , t ∈ R,

(4.2)
u(x, 0) = u0 (x) ∈ H s (Rn )
for any p > 1 and λ ∈ R. Equation (4.2) is the Euler-Lagrange equation for the
Lagrangian density
1 p − 1 p+1
L(u) = − ∆(|u|2 ) + λ |u| .
2 p+1
Space translation invariance leads to momentum conservation
Z
p~(t) = = ū∇udx, (4.3)
Rn

a quantity that has no definite sign. It turns out that one can also use this conser-
vation law in the defocusing case and prove monotonicity formulas that are very
useful in studying the global-in-time properties of the solutions at t = ∞. For most
of these classical results the reader can consult [4], [46].
The study of the problem at infinity is an attempt to describe and classify the
asymptotic behavior-in-time for the global solutions. To handle this issue, one tries
to compare the given nonlinear dynamics with suitably chosen simpler asymptotic
dynamics. For the semilinear problem (4.2), the first obvious candidate for the
simplified asymptotic behavior is the free dynamics generated by the group S(t) =
e−it∆ . The comparison between the two dynamics gives rise to the questions of the
existence of wave operators and of the asymptotic completeness of the solutions.
More precisely, we have:
i) Let v+ (t) = S(t)u+ be the solution of the free equation. Does there exist a
solution u of equation (4.2) which behaves asymptotically as v+ as t → ∞, typically
in the sense that ku(t) − v+ kH 1 → 0, as t → ∞. If this is true, then one can define
the map Ω+ : u+ → u(0). The map is called the wave operator and the problem
of existence of u for given u+ is referred to as the problem of the existence of the
wave operator. The analogous problem arises as t → −∞.
ii) Conversely, given a solution u of (4.2), does there exist an asymptotic state u+
such that v+ (t) = S(t)u+ behaves asymptotically as u(t), in the above sense. If
that is the case for any u with initial data in X for some u+ ∈ X, one says that
asymptotic completeness holds in X.
In effect the existence of wave operators asks for the construction of global so-
lutions that behave asymptotically as the solution of the free Schrödinger equation
while the asymptotic completeness requires all solutions to behave asymptotically
in this manner. It is thus not accidental that asymptotic completeness is a much
harder problem than the existence of the wave operators (except in the case of small
 ON LOCAL AND GLOBAL THEORY FOR NLS 19

data theory which follows from the iterative methods of the local well-posedness
theory).
Asymptotic completeness for large data not only require a repulsive nonlinearity
but also some decay for the nonlinear solutions. A key example of these ideas is
contained in the following generalized virial inequality, [31]:

Z Z
2
(−∆∆a(x))|u(x, t)| dxdt + λ (∆a(x))|u(x, t)|p+1 dxdt ≤ C (4.4)
Rn ×R Rn ×R

where a(x) is a convex function, u is a solution to (4.2), and C a constant that

depends only on the energy and mass bounds.

An inequality of this form, which we will call a one-particle inequality, was first
derived in the context of the Klein-Gordon equation by Morawetz in [34], and then
extended to the NLS equation in [31]. Most of these estimates are referred in the
literature as Morawetz type estimates. The inequality was applied to prove as-
ymptotic completeness first for the nonlinear Klein-Gordon and then for the NLS
equation in [41], and then in [31] for slightly more regular solutions in space di-
mension n ≥ 3. The case of general finite energy solutions for n ≥ 3 was treated
in [18] for the NLS and in [17] for the Hartree equation. The treatment was then
improved to the more difficult case of low dimensions in [35, 36].
The bilinear a priori estimates that we outline here give stronger bounds on the
solutions and in addition simplify the proofs of the results in the papers cited above.
For a detailed summary of the method see [19]. In the original paper by Morawetz,
the weight function that was used was a(x) = |x|. This choice has the advantage
1
that the distribution −∆∆( |x| ) is positive for n ≥ 3. More precisely it is easy to
n−1
compute that ∆a(x) = |x| and that
(
8πδ(x), if n = 3
−∆∆a(x) = (n−1)(n−3)
|x|2 , if n ≥ 4.
In particular, the computation in (4.4) gives the following estimate for n = 3 and
λ positive
|u(x, t)|p+1
Z Z
|u(t, 0)|2 dt + dxdt ≤ C. (4.5)
R R3 ×R |x|
Similar estimates are true in higher dimensions. The second, nonlinear term, or
certain local versions of it, have played central role in the scattering theory for
the nonlinear Schrödinger equation, [3], [18], [22], [31]. The fact that in 3d, the
bi-harmonic operator acting on the weight a(x) produces the δ−measure can be
exploited further. In [9], a quadratic Morawetz inequality was proved by correlating
two nonlinear densities ρ1 (x) = |u(x)|2 and ρ2 (y) = |u(y)|2 and define as a(x, y)
the distanceRbetween x and y in 3d. The authors obtained an a priori estimate
of the form R3 ×R |u(x, t)|4 dx ≤ C for solutions that stay in the energy space. A
frequency localized version of this estimate has been successfully implemented to
remove the radial assumption of Bourgain, [3], and prove global well-posedness and
scattering for the energy-critical (quintic) equation in 3d, [10]. For n ≥ 4 new
20 N. PAVLOVIĆ AND N. TZIRAKIS

quadratic Morawetz estimates were given in [47]. Finally in [6] and in [40] these
estimates were extended to all dimensions.
We should mention that taking as the weight function the distance between two
points in Rn is not the only approach, see [7] for a recent example. Nowadays
it is well understood that the bilinear Morawetz inequalities provide a unified ap-
proach for proving energy scattering for energy sub-critical solutions of the NLS
when p > 1 + n4 (L2 super-critical nonlinearities). This last statement has been
rigorously formalized only recently due to the work of the aforementioned authors,
and a general exposition has been published in [19]. Sub-energy solution scattering
in the same range of powers has been initiated in [9]. For the L2 -critical problem,
scattering is a very hard problem, but the problem has now been resolved in a series
of new papers by B. Dodson, [11, 12, 13]. For mass sub-critical solutions, scattering
even in the energy space is a very hard problem, and is probably false. Nevertheless,
two particle Morawetz estimates have been used for the problem of the existence
(but not uniqueness) of the wave operator for mass subcritical problems, [24]. We
have already mentioned their implementation to the hard problem of energy crit-
ical solutions in [3], [22], and [10]. Recent preprints have used these inequalities
for the mass critical problem, [11], and the energy super-critical problem, [30]. For
a frequency localized one particle Morawetz inequality and its application to the
scattering problem for the mass-critical equation with radial data see [48].

We start with the equation

iut + ∆u = λ|u|p−1 u (4.6)
with p ≥ 1 and λ ∈ R. We use Einstein’s summation convention throughout. Ac-
cording to this convention, when an index variable appears twice in a single term,
once in an upper (superscript) and once in a lower (subscript) position, it implies
that we are summing over all of its possible values. We will also write ∇j u for
∂u n n ∂a(x,y)
∂xj . For a function a(x, y) defined on R × R we define ∇x,j a(x, y) = ∂xj and
similarly for ∇x,k a(x, y).

We define the mass density ρ and the momentum vector p~, by the relations
ρ = |u|2 , pk = =(ū∇k u).
It is well known, [4], that smooth solutions to the semilinear Schrödinger equation
satisfy mass and momentum conservation. The local conservation of mass reads
p = ∂t ρ + 2∇j pj = 0
∂t ρ + 2div~ (4.7)
and the local momentum conservation is
 1 p − 1 p+1

∂t pj + ∇k δkj − ∆ρ + λ |u| + σkj = 0 (4.8)
2 p+1
where the symmetric tensor σjk is given by
σjk = 2<(∇j u∇k u).

Notice that the term λ p−1

p+1 |u|
p+1
is the only nonlinear term that appears in the
expression. One can express the local conservation laws purely in terms of the
 ON LOCAL AND GLOBAL THEORY FOR NLS 21

mass density ρ and the momentum p~ if we write

p − 1 p+1 p+1 p − 1 p+1
λ |u| =2 2 λ ρ 2
p+1 p+1
and
1 1
σjk = 2<(∇j u∇k u) = (2pj pk + ∇j ρ∇k ρ),
ρ 2
but we will not use this formulation in these notes.

4.1. One particle Morawetz inequalities. We are ready to state the main the-
orem of this section:
Theorem 4.1. [6, 9, 40, 47] Consider u ∈ Ct (R; C0∞ (Rn )) a smooth and compactly
supported solution to (4.6) with u(x, 0) = u(x) ∈ C0∞ (Rn ). Then for n ≥ 2 we have
that
p−1 |u(y, t)|2 |u(x, t)|p+1
Z Z
− n−3 2 2
CkD 2 (|u| )kL2 L2 + (n − 1)λ dxdydt
t x p + 1 Rt Rnx ×Rny |x − y|

≤ ku0 k2L2 sup |My (t)|,

t∈R
where
x−y
Z


My (t) = · = u(x)∇u(x) dx,
Rn |x − y|
α α f (ξ) = |ξ|α u
D is defined on the Fourier side as D d b(ξ) for any α ∈ R and C is a
positive constant that depends only on n, [45]. For n = 1 the estimate is
p−1 1
k∂x (|u|2 )k2L2 L2 + λ kukp+3
Lp+3 Lp+3
≤ ku0 k3L2 sup k∂x ukL2 .
t x p+1 t x 2 t∈R

Remarks on Theorem 4.1.

1. By the Cauchy-Schwarz inequality it follows that for any n ≥ 2,

sup |My (t)| . ku0 kL2 sup k∇u(t)kL2 .
0,t t∈R

A variant of Hardy’s inequality gives

sup |My (t)| . sup ku(t)k2 1 ,
0,t t∈R Ḣ 2

For details, see [19].

2. Concerning our main theorem, we note that both the integrated functions in
the second term on the left hand side of the inequalities are positive. Thus when
λ > 0, which corresponds to the defocusing case, and for H 1 data say, we obtain
for n ≥ 2:
n−3 3 1 3 1
kD− 2 (|u|2 )kL2t L2x . ku0 kL2 2 sup k∇u(t)kL2 2 . M (u0 ) 2 E(u0 ) 2 ,
t∈R

and for n = 1
3 1 3 1
k∂x (|u|2 )k2L2 L2 . ku0 kL2 2 sup k∂x u(t)kL2 2 . M (u0 ) 2 E(u0 ) 2 .
t x
t∈R
22 N. PAVLOVIĆ AND N. TZIRAKIS

These are easy consequences of the conservation laws of mass (2.3) and energy (2.2).
They provide the global a priori estimates that are used in quantum scattering in
the energy space, [19].

3. Analogous estimates hold for the case of the Hartree equation iut + ∆u =
λ(|x|−γ ? |u|2 )u when 0 < γ < n, n ≥ 2. For the details, see [24]. We should point
out that for 0 < γ ≤ 1 scattering fails for the Hartree equation, [23], and thus the
estimates given in [24] for n ≥ 2 cover all the interesting cases.

4. Take λ > 0. The expression

n−3
kD− 2 (|u|2 )kL2t L2x ,
for n = 3, provides an estimate for the L4t L4x norm of the solution. For n = 2 by
Sobolev embedding one has that
1
kuk2L4 L8 = k|u|2 kL2t L4x . kD 2 |u|2 kL2t L2x . CM (u0 ),E(u0 ) .


t x

For n ≥ 4 the power of the D operator is negative but some harmonic analysis and
interpolation with the trivial inequality
1
kD 2 ukL∞ 2 . kuk
t Lx
1
L∞ 2
t Ḣx

provides an estimate in a Strichartz norm. For the details see [47].

5. In the defocusing case all the estimates above give a priori information for the
1
Ḣ 4 -critical Strichartz norm. We remind the reader that the Ḣ s -critical Strichartz
norm is kukLqt Lrx where the pair (q, r) satisfies 2q + nr = n2 − s. In principle the corre-
1
lation of k particles will provide a priori information for the Ḣ 2k critical Strichartz
1
norm. In 1d an estimate that provides a bound on the Ḣ 8 critical Strichartz norm
has been given in [8].

6. To make our presentation easier we considered smooth solutions of the NLS

equation. To obtain the estimates in Theorem 4.1 for arbitrary H 1 functions we
have to regularize the solutions and then take a limit. The process is described in
[19].

7. A more general bilinear estimate can be proved if one correlates two different
solutions (thus considering different density functions ρ1 and ρ2 ). Unfortunately,
one can obtain useful estimates only for n ≥ 3. The proof is based on the fact that
−∆2 |x| is a positive distribution only for n ≥ 3. For details the reader can check
[9]. Our proof shows that the diagonal case when ρ1 = ρ2 = |u|2 provides useful
monotonicity formulas in all dimensions.

Proof. We define the Morawetz action centered at zero by

Z
M0 (t) = ∇a(x) · p~(x) dx, (4.9)
R
n
where the weight function a(x) : R → R is for the moment arbitrary. The minimal
requirements on a(x) call for the matrix of the second partial derivatives ∂j ∂k a(x)
 ON LOCAL AND GLOBAL THEORY FOR NLS 23

to be positive definite. Throughout our paper we will take a(x) = |x|, but many
estimates can be given with different weight functions, see for example [7] and [29].
If we differentiate the Morawetz action with respect to time we obtain:
Z Z
∂t M0 (t) = ∇a(x) · ∂t p~(x) dx = ∇j a(x)∂t pj (x) dx
Rn Rn

p − 1 p+1

Z Z
1
∇j ∇k a(x) δkj − ∆ρ+λ ∇j ∇k a(x) < ∇j u∇k u dx,




= |u| dx+2
Rn 2 p+1 Rn

where we use equation (4.8). We rewrite and name the equation as follows
p − 1 p+1

Z Z
1
∇j ∇k a(x) < ∇j u∇k u dx.



∂t M0 (t) = ∆a(x) − ∆ρ+λ |u| dx+2
Rn 2 p+1 Rn
(4.10)
Notice that for a(x) = |x| the matrix ∇j ∇k a(x) is positive definite and the same
is true if we translate the weight function by any point y ∈ Rn and consider
∇x,j ∇x,k a(x − y) for example. That is for any vector function on Rn , {vj (x)}nj=1 ,
with values on R or C we have that
Z
∇j ∇k a(x) v j (x)vk (x)dx ≥ 0.


Rn
x 1 xj xk

To see this, observe that for n ≥ 2 we have ∇j a = |x|j and ∇j ∇k a = |x| δkj − |x|2 .
Summing over j = k we obtain ∆a(x) = n−1 |x| . Then

1 k xj xk
j 1  x · ~v (x)
2 
∇j ∇k a(x)v j (x)vk (x) = δj − 2
v (x)vk (x) = |~v (x)|2 − ≥0
|x| |x| |x| |x|
by the Cauchy-Schwarz inequality. Notice that it does not matter if the vector
function is real or complex valued for this inequality to be true. In dimension one
(4.10) simplifies to
p − 1 p+1
Z
1
+ 2|ux |2 dx.


∂t M0 (t) = axx (x) − ∆ρ + λ |u| (4.11)
R 2 p+1
In this case for a(x) = |x|, we have that axx (x) = 2δ(x). Since the identity (4.10)
does not change if we translate the weight function by y ∈ Rn we can define the
Morawetz action with center at y ∈ Rn by
Z
My (t) = ∇a(x − y) · p~(x) dx.
Rn

We can then obtain like before

p − 1 p+1

Z
1
∂t My (t) = ∆x a(x − y) − ∆ρ + λ |u| dx (4.12)
R n 2 p +1
Z
∇x,j ∇x,k a(x − y) < ∇x,j u∇x,k u dx.



+2 (4.13)
Rn

Recall that
 
λ(p − 1) p+1 1
Z Z
∂t M0 = ∆a(x) |u| − ∆ρ dx + (∂j ∂ k a(x))σkj dx
Rn p+1 2 R n
24 N. PAVLOVIĆ AND N. TZIRAKIS

for a general weight function a(x).

If we pick a(x) = |x|2 , then ∆a(x) = 2n and ∂j ∂k a(x) = 2δkj . Therefore
2nλ(p − 1)
Z Z
∂t M = |u|p+1 dx + 2 |∇u|2 dx
p+1 R n n
 Z Z R  Z
1 2 λ 2λ
=8 |∇u| dx + |u|p+1 dx − (4 − n(p − 1)) |u|p+1 dx
2 Rn p + 1 Rn p+1 Rn
Z
2λ p+1
= 8E(u(t)) − (4 − n(p − 1)) |u| dx. (4.14)
p+1 Rn

Thus if we define the quantity

Z
V (t) = a(x)ρ(x)dx,
Rn

with a(x) = |x|2 , we have that

Z Z
∂t V (t) = a(x)∂t ρ(x)dx = −2 a(x) ∇ · p~ dx = 2M (t) (4.15)
Rn Rn

using integration by parts. Thus

Z
4λ
∂t2 V (t) = 16E(u(t)) − (4 − n(p − 1)) |u|p+1 dx. (4.16)
p+1 Rn

Another useful calculation is the following. Set

8t2 λ
Z
K(t) = k(x + 2it∇)uk2L2 + |u|p+1 dx.
p+1 Rn

Then we have:
8t2 λ
Z Z
K(t) = kxuk2L2 + 4t2 k∇uk2L2 − 4t x · p dx + |u|p+1 dx
Rn p + 1 R n
Z Z
2
= a(x)ρ(x)dx + 8t E(u(t)) − 2t ∇a · p dx
n n
ZR Z R
= a(x)ρ(x)dx + 8t2 E(u0 ) − 2t ∇a · p dx, (4.17)
Rn Rn
2
with a(x) = |x| . However
Z Z
∂t a(x)ρ(x)dx = ∇a · p dx
Rn Rn

and thus
Z
∂t K(t) = −2t ∂j a(x)∂t pj dx + 16tE(u0 ) = −2t∂t M (t) + 16tE(u0 ).
Rn

If we use (4.14) we have that

Z
4λt
∂t K(t) = (4 − n(p − 1)) |u|p+1 dx.
p+1 Rn

Notice that for p = 1 + n4 , the quantity K(t) is conserved.

 ON LOCAL AND GLOBAL THEORY FOR NLS 25

4.2. Two particle Morawetz inequalities. We now define the two-particle Morawetz
action Z
M (t) = |u(y)|2 My (t) dy
Rn
y

and differentiate with respect to time. Using the identity above and the local
conservation of mass law we obtain four terms
Z Z
2
∂t M (t) = |u(y)| ∂t My (t) dy + ∂t ρ(y)My (t) dy
Rn
y Rn
y

p − 1 p+1

Z
1
= |u(y)|2 ∆x a(x − y) − ∆ρ + λ |u| dxdy
Rn n
y ×Rx
2 p+1
Z
|u(y)|2 ∇x,j ∇x,k a(x − y) < ∇x,j u∇x,k u dxdy



+2
Rn n
y ×Rx
Z
−2 ∇y,j pj (y)∇x,k a(x − y)pk (x)dxdy
Rn n
y ×Rx
Z
= I + II + III + 2 pj (y)∇y,j ∇x,k a(x − y)pk (x)dxdy
Rn n
y ×Rx

by integration by parts with respect to the y−variable. Since

∇y,j ∇x,k a(x − y) = −∇x,j ∇x,k a(x − y)
we obtain that
Z
∂t M (t) = I + II + III − 2 ∇x,j ∇x,k a(x − y)pj (y)pk (x)dxdy (4.18)
Rn n
y ×Rx

= I + II + III + IV
where Z
1

I= |u(y)|2 ∆x a(x − y) − ∆ρ dxdy,
Rn n
y ×Rx
2
p − 1 p+1

Z
II = |u(y)|2 ∆x a(x − y) λ |u| dxdy,
Rn n
y ×Rx
p+1
Z
|u(y)|2 ∇x,j ∇x,k a(x − y) < ∇x,j u∇x,k u dxdy,



III = 2
Rn n
y ×Rx
Z
IV = −2 ∇x,j ∇x,k a(x − y)pj (y)pk (x)dxdy.
Rn n
y ×Rx

n−1
Claim: III + IV ≥ 0. Assume the claim. Since ∆x a(x − y) = |x−y| we have that
n−1 |u(y)|2 p−1 |u(y)|2
Z Z
|u(x)|p+1 dxdy.


∂t M (t) ≥ −∆ρ dxdy+(n−1)λ
2 Rn ×R n |x − y| p + 1 Rn ×Rn |x − y|
y x y x

But recall that on one hand we have that −∆ = D2 and on the other that the
1 c
distributional Fourier transform of |x| for any n ≥ 2 is |ξ|n−1 where c is a positive
constant depending only on n. Thus we can define
Z
−(n−1) f (y)
D f (x) = c dy
Rn |x − y|
26 N. PAVLOVIĆ AND N. TZIRAKIS

and express the first term as

n−1 |u(y)|2 n−1
Z
n−3
< D−(n−1) |u|2 , D2 |u|2 >= CkD− 2 |u|2 k2L2x


−∆ρ dxdy = c
2 Rn n
y ×Rx
|x − y| 2
by the usual properties of the Fourier transform for positive and real functions.
Integrating from 0 to t we obtain the theorem in the case that n ≥ 2.

Proof of the claim: Notice that

Z  
∇x,j ∇x,k a(x−y) |u(y)|2 < ∇x,j u(x)∇x,k u(x) −pj (y)pk (x) dxdy


III+IV = 2
Rn n
y ×Rx
Z  ρ(y) 
∇x,j ∇x,k a(x−y) < u(x)(∇x,j u(x))u(x)(∇x,k u(x)) −pj (y)pk (x) dxdy.


=2
Rn n
y ×Rx
ρ(x)
Since
∇x,j ∇x,k a(x − y) = ∇y,j ∇y,k a(y − x)
by exchanging the roles of x and y we obtain the same inequality and thus
Z  ρ(y)
∇x,j ∇x,k a(x−y) < u(x)(∇x,j u(x))u(x)(∇x,k u(x)) −pj (y)pk (x)


III+IV =
Rn n
y ×Rx
ρ(x)

ρ(x) 
< u(y)(∇y,j u(y))u(y)(∇y,k u(y)) − pj (x)pk (y) dxdy.


+
ρ(y)
Now set z1 = u(x)∇x,k u(x) and z2 = u(x)∇x,j u(x) and apply the identity
<(z1 z̄2 ) = <(z1 )<(z2 ) + =(z1 )=(z2 )
to obtain
< u(x)(∇x,j u(x))u(x)(∇x,k u(x)) = < u(x)∇x,k u(x) < u(x)∇x,j u(x)

1
+= u(x)∇x,k u(x) = u(x)∇x,j u(x) = ∇x,k ρ(x)∇x,j ρ(x) + pk (x)pj (x)


4
and similarly

1
< u(y)(∇y,j u(y))u(y)(∇y,k u(y)) = ∇y,k ρ(y)∇y,j ρ(y) + pk (y)pj (y).
4
Thus
Z
1 ρ(y)
III + IV = ∇x,j ∇x,k a(x − y) ∇x,k ρ(x)∇x,j ρ(x)dxdy
4 Rny ×Rnx ρ(x)
Z
1 ρ(x)
+ ∇y,j ∇y,k a(x − y) ∇y,k ρ(y)∇y,j ρ(y)dxdy
4 Rny ×Rnx ρ(y)
Z  ρ(y) ρ(x) 
+ ∇y,j ∇y,k a(x−y) pk (x)pj (x)+ pk (y)pj (y)−pk (x)pj (y)−pk (y)pj (x) dxdy.
Rn n
y ×Rx
ρ(x) ρ(y)
Since the matrix ∇x,j ∇x,k a(x − y) = ∇y,j ∇y,k a(x − y) is positive definite, the first
two integrals are positive. Thus,
III + IV ≥
Z  ρ(y) ρ(x) 
∇x,j ∇x,k a(x−y) pk (x)pj (x)+ pk (y)pj (y)−pk (x)pj (y)−pk (y)pj (x) dxdy.
Rn n
y ×Rx
ρ(x) ρ(y)
 ON LOCAL AND GLOBAL THEORY FOR NLS 27

Now if we define the two point vector

s s
ρ(y) ρ(x)
Jk (x, y) = pk (x) − pk (y)
ρ(x) ρ(y)
we obtain that
Z
III + IV ≥ ∇x,j ∇x,k a(x − y)J j (x, y)Jk (x, y)dxdy ≥ 0
Rn n
y ×Rx

and we are done.

The proof when n = 1 is easier. First, an easy computation shows that if
a(x, y) = |x − y| then ∂xx a(x, y) = 2δ(x − y). In this case from (4.18) we obtain
p−1
Z Z
1
|u(y)|2 2δ(x − y) − ρxx dxdy + 2 |u(x)|2 λ |u(x)|p+1 dx



∂t M (t) =
Ry ×Rx 2 R p+1
Z Z
+4 |u(x)|2 |ux |2 dx − 4 p2 (x)dx.
R R
But Z Z 
1 2
2
∂x |u(x)|2


|u(y)| 2δ(x − y) − ρxx dxdy = dx.
Ry ×Rx 2 R

In addition a simple calculation shows that

 2  2 1 2
|u(x)|2 |ux |2 = <(uux ) + =(uux ) = ∂x |u|2 + p2 (x).
4
Thus
 2
4|u(x)|2 |ux |2 − 4p2 (x) = ∂x |u|2
and the identity becomes
p−1
Z  2 Z
∂x |u|2 dx + 2 |u(x)|2 λ |u(x)|p+1 dx


∂t M (t) = 2 (4.19)
R R p+1
which finishes the proof of the theorem. 

5. Applications.

In this section we present a few applications of the decay estimates that were
established in Section 4.

5.1. Blow-up for the energy sub-critical and mass (super)-critical prob-
lem . We show a criterion for blow-up for the energy subcritical and mass critical
or super-critical
4 4
1+ <p<1+
n n−2
focusing (λ = 1) problem which is due to Zakharov and Glassey. In our presentation
we follow [21]. In addition we assume that our data have some decay (which will
be specified below).
28 N. PAVLOVIĆ AND N. TZIRAKIS

From the lwp theory we have a well-defined solution in (0, T ? ) of the following
initial value problem:
iut + ∆u + |u|p−1 u = 0, x ∈ Rn , t ∈ R,

1 n 2 n 2 (5.1)
u(x, 0) = u0 (x) ∈ H (R ) ∩ L (R , |x| dx),
4 4 4
for any 1 + n ≤p<1+ n−2 when n ≥ 3, and for 1 + n ≤ p < ∞ when n = 1, 2.

Recall from Section 4 that for the variance, which was introduced as follows:
Z
V (t) = |x|2 |u(x, t)|2 dx,
Rn
we calculated that (see (4.15) and (4.16) and expressions leading to them):
∂t V (t) = 2M (t), (5.2)
where Z Z
M (t) = ~x · p~ dx = ~x · =(ū∇u) dx,
Rn Rn
and Z
4
∂t2 V (t) = 16E(u(t)) + (4 − n(p − 1)) |u|p+1 dx. (5.3)
p+1 Rn
Hence (5.3) together with conservation of energy and the fact that p ≥ 1 + n4 ,
implies:
∂t2 V (t) ≤ 16E(u0 ),
which we can integrate twice to obtain:
V (t) ≤ 8t2 E(u0 ) + tV 0 (0) + V (0)
= 8t2 E(u0 ) + 2tM (0) + V (0)
Z
= 8t2 E(u0 ) + 4t ~x · =(u0 ∇u0 ) dx + kxu0 k2L2 . (5.4)
Rn
Since
u0 ∈ Σ = H 1 (Rn ) ∩ L2 (Rn , |x|2 dx),
the coefficients of the second degree polynomial in t on the right hand side of (5.4)
are finite. Now if the initial data have negative energy, that is if
E(u0 ) < 0,
2
the coefficient of t is negative. On the other hand, for all times
Z
V (t) = |x|2 |u(x, t)|2 dx ≥ 0.
Rn
Therefore V (t) starts with a positive value V (0) and at some finite time the second
order polynomial V (t) will cross the horizontal axis. Thus T ? is finite. By the
blow-up alternative of the lwp theory this gives that
lim ku(t)kH 1 = ∞,
t→T ?

if in addition to u0 ∈ H 1 , we have that kxu0 kL2 < ∞ and E(u0 ) < 0.

Remark 5.1. We make a few comments:

(1) Note that the assumption E(u0 ) < 0 is a sufficient condition for finite-time
blow-up, but it is not necessary. One can actually prove that for any E0 > 0
there exists u0 with E(u0 ) = E0 and T ? < ∞. For details consult [4].
 ON LOCAL AND GLOBAL THEORY FOR NLS 29

(2) One can reasonably ask whether she can prove the same result for H 1 data?
The authors in [38] prove such a result with the additional assumption of
radial symmetry for any n ≥ 2. For the L2 −critical case (p = 1 + n4 ) the
radial assumption is not needed. See the papers [39, 20, 37] for details.
(3) Many results have been devoted to the rate of the blow-up for the focusing
problem. A variant of the local well-posedness theory provides the following
result:
If u0 ∈ H 1 and T ? < ∞, then there exists a δ > 0 such that for all
0 ≤ t < T ? we have that
δ
k∇u(t)kL2 ≥ 1 n−2 .
(T ? − t) p−1 − 4
Note that the above gives a lower estimate but not an upper estimate. The
authors in [33] have provided an upper estimate for the L2 -critical case that
is very close to the one above.

5.2. Global Well-Posedness for the L2 -critical problem. We have seen that
in the mass-critical case when p = 1 + n4 the local existence time depends not only
on the norm of the initial data but also on the profile. This prevents the use of
the conservation of mass law in order to extend the solutions globally, even in the
defocusing case (λ = −1).

5.2.1. Defocusing problem under the finite variance assumption. In the case when
λ < 0, the conjecture was (for a long time) that T ? = ∞. Although the conjecture
is proven to be true in [11, 12, 13], in these notes we present a positive answer to
an easier problem where we consider the corresponding problem for H 1 data (that
can be large), but in addition we assume finiteness of the variance. This scenario
can be analyzed using methods of Section 4 and as such it fits well into the flow of
our presentation.
Recall that
8t2
Z
K(t) = k(x + 2it∇)uk2L2 + |u|p+1 dx
p+1 Rn
is a conserved quantity for p = 1 + n4 . Thus
8t2
Z
K(t) = k(x + 2it∇)uk2L2 + |u|p+1 dx = kxu0 k2L2 .
p+1 Rn
We approximate the data with an H 1 sequence such that u0,n → u0 in L2 and
have finite variance. The corresponding solutions satisfy un ∈ C(R, H 1 (Rn )) and
xun ∈ C(R, L2 (Rn )). The conservation law for K(t) implies that
8t2
Z
|un |p+1 dx ≤ C
p + 1 Rn
and thus Z
4 C
|un |2+ n dx ≤ 2
R n t
for all t ∈ (0, T ? ). By continuous dependence this implies that
Z
4 C
|u(x, t)|2+ n dx ≤ 2
Rn t
30 N. PAVLOVIĆ AND N. TZIRAKIS

for a.a. t ∈ (0, T ? ). Thus if T ? < ∞ one can integrate the above quantity from any
t < T ? to T ? and obtain that
Z T? Z
4
|u(x, t)|2+ n dxdt < C.
t Rn
Since on the other hand we have that
4 4
2+ n 2+ n
u ∈ Lt ((0, t); Lx )
we conclude
2+ 4 2+ 4
Lt n ((0, T ? ); Lx n ) < ∞.
But this contradicts the blow-up alternative for this problem and thus T ? = ∞.
2+ 4 2+ 4
Actually since the L2 Strichartz norm Lt n Lt n is bounded we also have scattering
(more on that later).

5.2.2. Focusing problem. Now let us derive a global well-posedness condition for
the focusing equation
4
iut + ∆u + |u| n u = 0. (5.5)
2
We have already seen that for small enough L data the problem, focusing or
defocusing, has global solutions. We have also mentioned the result in [14] that
gives a sharp criterion for global existence for the focusing problem. Here we
reproduce the result in [51] which states that if one assumes small L2 data (but not
arbitrarily small), which are, in addition, in H 1 , global well-posedness follows by
discovering the sharp constant of the Gagliardo-Nirenberg inequality.
More precisely since
4 4 4
2+ n
ku(t)k 4 ≤ Ck∇u(t)k2L2 ku(t)kLn2 = Ck∇u(t)k2L2 ku0 kLn2 ,
L2+ n
one can easily see that the energy functional
Z Z
1 2 1 4
E(u)(t) = |∇u(t)| dx − 4 |u(t)|2+ n dx
2 2+ n
is bounded from below as follows
 
1 4
E(u(t)) = E(u0 ) ≥ k∇u(t)k2L2 − Cku0 kLn2 . (5.6)
2
Thus for ku0 kL2 < η, η a fixed number, we have that

k∇u(t)kL2 + ku(t)kL2 ≤ CM (u0 ),E(u0 ) < ∞.

By the blow-up alternative of the H 1 theory we see that Tmax = ∞.

The question remains what is the optimal η. It was conjectured that, even with
L2 data, the optimal η is the mass of the ground state Q, which is the solution to
the elliptic equation:
4
−Q + ∆Q = |Q| n Q,
that can be obtained by using the the ansatz u(x, t) = eit Q(x) in (5.5). It is shown
that Q is unique, positive, spherically symmetric and very smooth (see [4] for exact
references). Also Q satisfies certain identities (Pohozaev’s identities) that can be
 ON LOCAL AND GLOBAL THEORY FOR NLS 31

obtained by multiplying the elliptic equation by ū and x · ∇u and take the real part
respectively. In particular the identities imply that E(Q) = 0. In [51] Weinstein
discovered that the mass of the ground state is related to the best constant of the
Gagliardo-Nirenberg inequality. More precisely by minimizing the functional
4
k∇u(t)k2L2 kukLn2
J(u) = 4
2+ n
,
kuk 2+ 4
Lx n
Weinstein showed that the best constant of the Galgiardo-Nirenberg inequality
1 4
2+ n C 4

4 ku(t)kL2+ n4 ≤ k∇u(t)k2L2 ku(t)kLn2 ,

2+ n 2
is
−4
C = kQkL2n .
Hence we can revisit (5.6) to obtain
4 !
1 ku0 kLn2
E(u0 ) ≥ k∇u(t)k2L2 1− 4 .
2 kQkLn2
Therefore, if ku0 kL2 < kQkL2 , we have a global solution.
Moreover the condition is sharp in the sense that for any η > kQkL2 , there ex-
ists u0 ∈ H 1 such that ku0 kL2 = η, and u(t) blows-up in finite time. To see that, set
η
γ= > 1,
kQkL2
and consider u0 = γQ. Then ku0 kL2 = η and
4 4
4
2+ n γ 2+ n − γ 2 γ 2+ n − γ 2
E(u0 ) = γ E(Q) − k∇Qk2L2 = − k∇Qk2L2 < 0.
2 2
Since u0 = γQ ∈ Σ and E(u0 ) < 0, by the Zakharov-Glassey argument we have
blow-up in finite time.
Remark 5.2. As consequence of the pseudo-conformal transformation
n i|x|2 t x
u(x, t) → (1 − t)− 2 e− 4(1−t) u( , ),
1−t 1−t
we actually have blow-up even for η = kQkL2 . We cite [4] for the details. It is
interesting that the blow-up rate is 1t and thus at least in the L2 -critical case the
lower estimate we gave is not optimal for all blow-up solutions.

5.3. Blow-up for the L2 -critical problem. We now prove that for the focusing
L2 -critical problem, the mass at the origin concentrates the mass of the ground
state. We assume radial H 1 data with n ≥ 2. Both assumptions (radiality and
dimension) have been removed but the proof is more elaborate. For the detials of
the H 1 theory see [4] and the references therein.
Theorem 5.3. Consider (5.5) with u0 ∈ H 1 (Rn ) ∩ {radial} in dimensions n ≥
2. Let ρ be any function (0, ∞) → (0, ∞) such that lims↓0 ρ(s) = ∞ and that
1
lims↓0 s 2 ρ(s) = 0. If u is the maximal solution of (5.5) and T ? < ∞ then
lim inf
?
ku(t)kL2 (Ωt ) ≥ kQkL2 ,
t↑T
32 N. PAVLOVIĆ AND N. TZIRAKIS

where  
1
Ωt = x ∈ Rn : |x| < |T ? − t| 2 ρ(T ? − t) .

To prove the theorem we note that a result of W. Strauss states that a radial
bounded sequence of functions in H 1 contains a subsequence that converges strongly
2n
in Lp for 2 < p < n−2 . Now set
1
λ(t) =
k∇u(t)kL2
so that
lim λ(t) = 0.
t↑T ?

We claim that
lim inf
?
ku(t)kL2 (|x|<λ(t)ρ(T ? −t)) ≥ kQkL2 .
t↑T
δ
The result then follows since ρ is arbitrary and k∇u(t)kL2 ≥ 1 .
(T ? −t) 2

We prove the claim by contradiction. Assume there exists tn ↑ T ? such that

lim ku(t)kL2 (|x|<λ(tn )ρ(T ? −tn )) < kQkL2 .
n→∞

Set
n
vn (t) = λ(t) 2 u(tn , λ(tn )x).
Clearly
kvtn kL2 = 1,

k∇vn kL2 = 1, (5.7)

E(vn ) = λ(tn )2 E(u(tn )) = λ(tn )2 E(u0 ).

In particular
1 1 4
2+ n
E(vn ) = − 4 kvn k 4 ,
2 2+ L2+ n
n
and
lim E(vn ) = 0.
n→∞
Thus
4
2+ n 2 + n4
lim kvn k 4 →6= 0. (5.8)
n→∞ L2+ n 2
Since vn is a bounded H 1 sequence there exists a subsequence which we still denote
4
by vn that converges weakly to w in H 1 and strongly by Strauss’ result in L2+ n .
By the properties of weak and strong limits and (5.8) we have that
E(w) ≤ 0, w 6= 0.
By the sharp Gagliardo-Nirenberg inequality this means
kwkL2 ≥ kQkL2 .
 ON LOCAL AND GLOBAL THEORY FOR NLS 33

Now given M > 0 we have that

kwkL2 (|x|<M ) = lim kvn kL2 (|x|<M )
n→∞
= lim ku(tn )kL2 (|x|<M λ(tn ))
n→∞
≤ lim inf ku(tn )kL2 (|x|<λ(tn )ρ(T ? −t)) ,
n→∞
since ρ(s) → ∞ as s ↓ 0. But since M was arbitrary, we obtain
lim inf ku(tn )kL2 (|x|<λ(tn )ρ(T ? −t)) ≥ kwkL2 > kQkL2 ,
n→∞
reaching the contradiction.

5.4. Quantum scattering in the energy space. Consider the defocusing L2 -

super-critical problem
iut + ∆u − |u|p−1 u = 0, x ∈ Rn , t ∈ R,

1 n (5.9)
u(x, 0) = u0 (x) ∈ H (R ),
for any 1 + n4 < p < 1 + n−24
.
We define the set of initial values u0 which have a scattering state at +∞ (by time
reversibility all the statements are equivalent at −∞):
R+ = (u0 ∈ H 1 : T ? = ∞, u+ = lim e−it∆ u(t) exists ). (5.10)
t→∞
Now define the operator
U : R+ → H 1 .
This operator sends u0 to the scattering state u+ . If this operator is injective then
we can define the wave operator

Ω+ = U −1 : U (R+ ) → R+
which sends the scattering state u+ to u0 . Thus the first problem of scattering is
the existence of wave operator:
• Existence of wave operators. For each u+ there exists unique u0 ∈ H 1 such that
u+ = limt→∞ e−it∆ u(t).
If the wave operator is also surjective we say that we have asymptotic completeness
(thus in this case the wave operator is invertible):
• Asymptotic completeness. For every u0 ∈ H 1 there exists u+ such that u+ =
limt→∞ e−it∆ u(t).
Both statements make rigorous the idea that we have scattering if, as time goes to
infinity, the nonlinear solution of the NLS behaves like the solution of the linear
equation.
Using the decay estimates of section 4 we can solve the scattering problem for
every p > 1 + n4 . Well-defined wave operators for this range of p is easy and it is
almost a byproduct of the local theory. But asymptotic completeness is hard. In
dimensions n ≥ 3 this was proved in [18] and for n = 1, 2 in [35, 36]. The proofs are
complicated since they were achieved before the interaction Morawetz estimates.
Using the interaction Morawetz estimates we can prove the scattering properties
in two simple steps. To make the presentation clear we will only show the n = 3
case with the cubic nonlinearity. But keep in mind that the interaction Morawetz
estimates give global a priori control on quantities of the form
kukLqt Lrx ≤ CM (u0 ),E(u0 ) ,
34 N. PAVLOVIĆ AND N. TZIRAKIS

for certain q and r in all dimensions. It turns out that in the L2 -supercritical case
this is enough to give scattering for any p > 1 + n4 and n. Finally for completeness
we also outline the wave operator question.
Theorem 5.4. For every u+ ∈ H 1 (R3 ) there exists unique u0 ∈ H 1 (R3 ) such that
the maximal solution u ∈ C(R; H 1 (R3 )) of iut + ∆u = |u|2 u, satisfies
lim ke−it∆ u(t) − u+ kH 1 (R3 ) = 0.
t→∞

Proof: For u+ ∈ H 1 define the map

Z ∞
A(u)(t) = eit∆ u+ + i ei(t−s)∆ (|u|2 u)(s)ds.
t
What is the motivation behind this map? Recall that
Z t
u(t) = eit∆ u0 − i ei(t−s)∆ (|u|2 u)(s)ds,
0
Z t
e−it∆ u(t) = u0 − i e−is∆ (|u|2 u)(s)ds. (5.11)
0
If the problem scatters we have that limt→∞ ke−it∆ u(t) − u+ kH 1 = 0 and thus
Z ∞
u+ = u0 − i e−is∆ (|u|2 u)(s)ds (5.12)
0
1
in H sense. Now subtracting (5.12) from (5.11) we have that
Z ∞
u(t) = eit∆ u+ + i ei(t−s)∆ (|u|2 u)(s)ds.
t
By Strichartz estimates we have that
keit∆ u+ kLq Wx1,r . ku+ kH 1 < ∞.
t

By the monotone convergence theorem there exists T = T (u+ ) large enough such
that for q < ∞ we have
keit∆ u+ kLq Wx1,r . .
t

The trick here is to use the smallness assumption to iterate the map in the interval
(T, ∞). But our local theory was performed in the norms
kukS 1 (I×Rn ) = kukS 0 (I×Rn ) + k∇ukS 0 (I×Rn )
where
kukS 0 (I×Rn ) = sup kukLqt∈I Lrx .
(q,r)−admissible
But this norms contain L∞
t . So momentarily we will go to the smaller space
10
1, 10
X = L5t L5x ∩ Lt3 Wx 3
.
For this norm we also have that for large T
keit∆ u+ kX[T ,∞) . .
Furthermore Strichartz estimates show that
kA(u)kX[T ,∞) .  + kuk3X[T ,∞) .
 ON LOCAL AND GLOBAL THEORY FOR NLS 35

The main step here is Sobolev embedding

kf kL5t L5x . kf k 1, 30
L5t Wx 11

30
where the pair (5, 11 ) is Strichartz admissible. The details are as follows: Notice
that the dual pair of ( 10 10 10 10
3 , 3 ) is ( 7 , 7 ).
Z ∞  
it∆
kukL5t L5x . ke u+ kL5t L5x + k ei(t−s)∆ |u|2 u(s) dskL5t L5x
T
Z ∞  
. keit∆ u+ k 1 30 +k ei(t−s)∆ |u|2 u(s) dsk 1, 30
L5t Wx 13 T L5t Wx 13

.  + ku3 k 10 10 + k(∇u)u2 k 10 10
Lt7 Lx7 Lt7 Lx7

.  + kuk2L5 L5 kuk 10 10 + kuk2L5 L5 k∇uk 10 10

t x Lt3 Lx3 t x Lt3 Lx3

.  + kuk2L5 L5 kuk 10 1, 10 .  + kuk3X[T ,∞) .

t x Lt3 Wx 3

Similarly we derive
k∇uk 10 10 .  + kuk2L5 L5 k∇uk 10 10 .  + kuk3X[T ,∞) .
Lt3 Lx3 t x Lt3 Lx3

and
kuk 10 10 .  + kuk2L5 L5 kuk 10 10 .  + kuk3X[T ,∞) .
Lt3 Lx3 t x Lt3 Lx3

Thus for T large enough we have that

kukX[T ,∞) . .
More precisely to obtain the last claim one has to estimate kAkX , kA(u) − A(v)kX
and prove that the map A is a contraction. Thanks to the  we derive simultaneously
this property along with the estimate
kukX[T ,∞) . .
It remains to show that the solution is in C([T, ∞); H 1 (R3 )). But by Strichartz
again and using any admissible pair we have
kukLq Wx1,r . ku+ kH 1 + kuk3X[T ,∞) . ku+ kH 1 .
t∈[T ,∞)

In particular ψ = u(T ) ∈ H 1 and we have a strong H 1 solution of the equation

with initial data u(T ) = ψ. But we know that the solutions of this equation are
global and thus u(0) is well-defined. Finally
Z ∞
e−it∆ u(t) − u+ = i e−is∆ (|u|2 u)(s)ds,
t
  Z ∞  
−it∆
∇ e u(t) − u+ = i e−is∆ ∇(|u|2 u) (s)ds,
t
ke−it∆ u(t) − u+ kH 1 . k∇uk 10
3
10 kuk2L5 L5x . kuk3X[T ,∞)
L[t,∞) Lx3 [t,∞)

But for T large enough we have that kukX[T ,∞) .  and thus
lim ke−it∆ u(t) − u+ kH 1 = 0.
t→∞

Therefore u(0) = u0 ∈ H 1 satisfies the assumptions of the theorem. We end with

asymptotic completeness.
36 N. PAVLOVIĆ AND N. TZIRAKIS

Theorem 5.5. If u0 ∈ H 1 (R3 ) and if u ∈ C(R; H 1 (R3 )) where u is the solution of

iut + ∆u = |u|2 u, then there exists u+ such that
lim ke−it∆ u(t) − u+ k = 0.
t→∞

The proof is based on a simple proposition assuming the interaction Morawetz

estimates. This was the hardest part in the earlier proofs of quantum scattering.
Proposition 5.6. Let u be a global H 1 solution of the cubic defocusing equation
on R3 . Then
kukS 1 (R×R3 ) ≤ C.

Proof: We know that kukL4t L4x ≤ C for energy solutions. Thus we can pick 
small to be determined later and a finite number of intervals {Ik }k=1,2,...,M , with
M < ∞ such that
kukL4t∈I L4x ≤ 
k

for all k. If we apply the Strichartz estimates on each Ik we obtain for some α < 1
kukS 1 (Ik ) . ku(0)kH 1 + kuk2α
L4 4 kuk3−2α
S 1 (Ik ) , (5.13)
t∈Ik Lx

kukS 1 (Ik ) . ku(0)kH 1 + 2α kuk3−2α

S 1 (Ik ) .
We can pick  so small such that
kukS 1 (Ik ) ≤ K.
Since the number of intervals are finite and the conclusion can be made for all Ik0 s
the proposition follows.

Remarks. 1. Where do we use the condition p > 1 + n4 ? This is a delicate matter.

It is not hard to see that the interaction Morawetz estimates are global estimates
of Strichartz type but are not L2 scale invariant. If one inspects the right hand side
1
of the interaction inequalities, a simple scaling argument shows that these are H 4
invariant estimates. Thus only in the case that p > 1 + n4 we can take advantage
of an non L2 estimate such as L4t L4x . This is the heart of the matter in proving
(5.13). In the case that p = 1 + n4 we need to have a global L2 Strichartz estimate
10 10
like Lt3 Lx3 in dimensions 3. Estimates of this sort can never come from Morawetz
estimates due to scaling.
2. Notice that the proposition gives a global decay estimate for the nonlinear solu-
tion.

Let’s finish the proof of asymptotic completeness. Note that

Z t
e−it∆ u(t) = u0 − i e−is∆ (|u|2 u)(s)ds,
0
Z τ
e−iτ ∆ u(τ ) = u0 − i e−is∆ (|u|2 u)(s)ds.
0
Thus
ke−it∆ u(t) − e−iτ ∆ u(τ )kH 1 = ku(t) − e−i(t−τ )∆ u(τ )kH 1 . kuk3S 1 ≤C
(t,τ )
 ON LOCAL AND GLOBAL THEORY FOR NLS 37

again by Strichartz estimates. Thus as t, τ → ∞ we have that

ke−it∆ u(t) − e−iτ ∆ u(τ )kH 1 → 0.
By completeness of H 1 there exists u+ ∈ H 1 such that e−it∆ u(t) → u+ in H 1 as
t → ∞. In particular in H 1 we have
Z ∞
u+ = u0 − i e−is∆ (|u|2 u)(s)ds
0
and thus
ke−it∆ u(t) − u+ kH 1 . kuk3S 1 .
(t,∞)

As t → ∞ the conclusion follows.

More remarks. What about energy scattering for p ≤ 1 + n4 . The critical case
has been solved in [11, 12, 13]. For p < 1 + n4 the problem is completely open.
We have already mentioned that scattering makes rigorous the intuition that as
time increases, for a defocusing problem, the nonlinearity |u|p−1 u becomes negligi-
ble. From this observation one expects that the bigger the power of p the better
chance the solution has to scatter. Thus the question: Is there any threshold p0
with 1 < p0 ≤ 1 + n4 such that energy scattering does fail? The answer is yes and
p0 = 1 + n2 . This is in [43] for higher dimensions and in [1] for dimension one. More
precisely using the pseudo-conformal conservation law and decay estimates that we
discuss later in the notes, they showed that for any 1 < p ≤ p0 , U (−t)u(t) doesn’t
converge even in L2 . Thus the wave operators cannot exist in any reasonable set.
The problem remains open for
2 4
1+ <p<1+
n n
and for general energy data. For partial results see [24] and the references therein.

5.5. Quantum scattering in the Σ space. If we are willing to abandon the

energy space can we improve scattering in the range 1 + n2 < p < 1 + n4 ? Recall
that
Σ = H 1 (Rn ) ∩ L2 (Rn , |x|2 dx).
We will not go into the details but a few comments can clarify the situation. Exactly
like the energy case it is enough to prove that
kukS 1 (R×R3 ) ≤ C.
How one can obtain this estimate for different values of p? First recall that for
8t2
Z
2
K(t) = k(x + 2it∇)ukL2 + |u|p+1 dx
p + 1 Rn
we have that Z t
K(t) − K(0) = θ(s)ds,
0
where Z
4t
θ(t) = (4 − n(p − 1)) |u|p+1 dx.
p+1 Rn
Using this quantity and a simple analysis one can obtain the following proposition:
38 N. PAVLOVIĆ AND N. TZIRAKIS

Proposition 5.7. Consider the defocusing NLS

iut + ∆u = |u|p−1 u

(5.14)
u(x, 0) = u0 (x) ∈ H 1 (Rn ).
4
for any 1 < p < 1 + n−2 , n ≥ 3 (1 < p < ∞ for n = 1, 2). If in addition
kxu0 kL2 < ∞ and
u ∈ Ct0 (R; H 1 (Rn ))
solves (5.14), then we have:
i) If p > 1 + n4 then for any 2 ≤ r ≤ 2n
n−2 (2 ≤ r ≤ ∞ if n = 1, 2 ≤ r < ∞ if n = 2)
1 1
ku(t)kLr ≤ C|t|−n( 2 − r )
for all t ∈ Rn .
4 2n
ii) If p < 1 + n then for any 2 ≤ r ≤ n−2 (2 ≤ r ≤ ∞ if n = 1, 2 ≤ r < ∞ if
n = 2)
1 1
ku(t)kLr ≤ C|t|−n( 2 − r )(1−θ(r))
where (
0 if 2 ≤ r ≤ p + 1
θ(r) = [r−(p+1)][4−n(p−1)]
(r−2)[(n+2)−p(n−1)] if r > p + 1.

Remarks. 1. Notice that for p ≥ 1 + n4 the decay is as strong as the linear

equation. Recall here the basic L1 − L∞ estimate of the linear problem and its
interpolation with Plancherel’s theorem.
2. Using these estimates and the standard theory we have developed one can prove
that global solutions defined in the Σ space obey
kukS 1 (R×R3 ) ≤ C,
for any √
2−n+ n2 + 12n + 4 4
1+ <p<1+ .
2n n−2
The existence of wave operators and asymptotic completeness follows easily. Of
course √
2 2 − n + n2 + 12n + 4 4
1+ <1+ <1+ .
n 2n n
3. The existence of the wave operators can go below the above threshold in all
dimensions. Indeed one can cover the full range p > 1 + n2 . The subject is rather
technical and we refer to [4] for more details.

References
[1] J. E. Barab, Nonexistence of asymptotically free solutions for nonlinear Schrödinger equa-
tion, J. Math Phys. 25 (1984), pp. 3270–3273.
[2] J. Bourgain, Global solutions of nonlinear Schrödinger equations, American Mathematical
Society, Providence, RI, 1999.
[3] J. Bourgain, Global well-posedness of defocusing 3D critical NLS in the radial case, J. Amer.
Math. Soc., 12 (1999), 145–171.
[4] T. Cazenave, Semilinear Schrödinger equations, CLN 10, eds: AMS, 2003.
[5] T. Cazenave, and F. B. Weissler, Some remarks on the nonlinear Schrödinger equation in
critical case, Nonlinear semigroups, partial differential equations and attractors (Washington
DC 1987), 18–29. Lecture Notes in Mathematics, 1394. Springer, Berlin, 1989.
 ON LOCAL AND GLOBAL THEORY FOR NLS 39

[6] J. Colliander, M. Grillakis, and N. Tzirakis, Commutators and correlations estimates with
applications to NLS, Comm. Pure Appl. Math. 62 (2009), no. 7, 920–968.
[7] J. Colliander, M. Grillakis, and N. Tzirakis, Remarks on global a priori estimates for the
nonlinear Schrödinger equation, Proc. Amer. Math. Soc. 138 (2010), no. 12, 4359–4371.
[8] J. Colliander, J. Holmer, M. Visan and X. Zhang, Global existence and scattering for rough
solutions to generalized nonlinear Schrödinger equations on R, Commun. Pure Appl. Anal.
7 (2008), no. 3, 467–489.
[9] J. Colliander, M. Keel, G. Staffilani, H. Takaoka and T. Tao, Global existence and scattering
for rough solutions to a nonlinear Schrödinger equation on R3 , Comm. Pure Appl. Math.,
57 (2004), no. 8, 987–1014.
[10] J. Colliander, M. Keel, G. Staffilani, H. Takaoka, and T. Tao, Global well-posedness and
scattering in the energy space for the critical nonlinear Schrödinger equation in R3 , Ann.
of Math. (2), 167 (2008), no. 3, 767–865.
[11] B. Dodson, Global well-posedness and scattering for the defocusing, L2 −critical, nonlinear
Schrödinger equation when d ≥ 3, preprint, http://arxiv.org/abs/0912.2467.
[12] B. Dodson, Global well-posedness and scattering for the defocusing, L2 −critical, nonlinear
Schrödinger equation when d = 2, preprint, http://arxiv.org/abs/1006.1375.
[13] B. Dodson, Global well-posedness and scattering for the defocusing, L2 −critical, nonlinear
Schrödinger equation when d = 1, preprint, http://arxiv.org/abs/1010.0040.
[14] B. Dodson, Global well-posedness and scattering for the mass critical nonlinear
Schrödinger equation with mass below the mass of the ground state, preprint,
http://arxiv.org/abs/1104.1114.
[15] L. C. Evans Partial Differential Equations: Second Edition (Graduate Studies in Mathe-
matics).
[16] J. Ginibre and G. Velo, The global Cauchy problem for the nonlinear Schrödinger equation
revisited, Ann. Inst. H. Poincaré Anal. Non Linéaire, 2 (1985), no. 4, 309–327.
[17] J. Ginibre and G. Velo, Scattering theory in the energy space for a class of Hartree equations,
in Nonlinear Wave Equations, Y. Guo Ed. Contemporary Mathematics 263, AMS 2000.
[18] J. Ginibre and G. Velo, Scattering theory in the energy space for a class of nonlinear
Schrödinger equations, J. Math. Pure Appl., 64 (1985), 363-401.
[19] J. Ginibre, and G. Velo, Quadratic Morawetz inequalities and asymptotic completeness in
the energy space for nonlinear Schrödinger and Hartree equations, Quart. Appl. Math. 68
(2010), 113–134.
[20] L. Glangetas and F. Merle, A geometrical approach of existence of blow up solutions in H 1
for nonlinear Schrödinger equation, Rep. No. R95031, Laboratoire d’Analyse Numérique,
Univ. Pierre Marie Curie (1995).
[21] R. T. Glassey, On the blowing up of solutions to the Cauchy problem for nonlinear
Schrödinger equations, J. Math. Phys. 18 (1977), 1794–1797.
[22] M. Grillakis, On nonlinear Schrödinger equations, Commun. Partial Differential Equations
25, no. 9-10, (2005), 1827-1844.
[23] N. Hayashi and Y. Tsutsumi, Scattering theory for the Hartree type equations, Ann. Inst.
H. Poincaré Phys. Théor. 46 (1987), 187–213.
[24] J. Holmer, and N. Tzirakis Asymptotically linear solutions in H 1 of the 2D defocusing
nonlinear Schrödinger and Hartree equations, J. Hyperbolic Differ. Equ. 7 (2010), no. 1,
117–138.
[25] T.Kato, On nonlinear Schrödinger equations II. H s solutions and unconditionally well-
posedness, J. Anal. Math. 67 (1995), 281-306.
[26] T.Kato, On nonlinear Schrödinger equations, Ann. Inst. H. Poincare Phys. Theor. 46 (1987),
113–129.
[27] M. Keel, and T. Tao, Endpoint Strichartz estimates, Amer. J. Math. 120 (1998), no. 5,
955–980.
[28] C. E. Kenig and F. Merle, Global well-posedness, scattering and blow up for the energy-
critical, focusing, nonlinear Schrödinger equation in the radial case, Invent. Math. 166
(2006), 645–675.
[29] R. Killip, and M. Visan The focusing energy-critical nonlinear Schrödinger equation in
dimensions five and higher, Amer. J. Math. 132 (2010), 361–424.
[30] R. Killip and M. Visan, Energy supercritical NLS: Critical Ḣ s − bounds imply scattering,
http://arxiv.org/abs/0812.2084.
40 N. PAVLOVIĆ AND N. TZIRAKIS

[31] J. E. Lin, and W. A. Strauss, Decay and scattering of solutions of a nonlinear Schrödinger
equation, J. Funct. Anal. 30 (1978), no.2, 245–263.
[32] F. Linares and G. Ponce, Introduction to Nonlinear Dispersive equations, UTX, Springer-
Verlag, 2009.
[33] F. Merle, and P. Raphael Sharp upper bound on the blow-up rate for the critical nonlinear
Schrödinger equation, Geom. Funct. Anal. 13 (2003), no. 3, 591-642.
[34] C. Morawetz, Time decay for the nonlinear Klein-Gordon equation, Proc. Roy. Soc. A, 306
(1968), 291–296.
[35] K. Nakanishi, Energy scattering for nonlinear Klein-Gordon and Schrödinger equations in
spatial dimensions 1 and 2, J. Funct. Anal. 169 (1999), 201–225.
[36] K. Nakanishi, Energy scattering for Hartree equations, Math. Res. Lett. 6, no 1, (1999),107–
118.
[37] H. Nawa, Asymptotic and limiting profiles of blowup solutions of the nonlinear Schrdinger
equation with critical power, Comm. Pure Appl. Math. 52 (1999), no. 2, 193–270.
[38] T Ogawa, and Y. Tsutsumi, Blow-up of H 1 solutions for the nonlinear Schrödinger equation,
J. Differential Equations 92 (1991), 317–330.
[39] T Ogawa, and Y. Tsutsumi, Blow-up of H 1 solutions for the one dimensional nonlinear
Schrödinger equation with critical power nonlinearity, Proc. Amer. Math. Soc. 111 (1991),
487–496.
[40] F. Planchon, and L. Vega, Bilinear Virial Identities and Applications, Ann. Sci. c. Norm.
Supr. (4) 42 (2009), no. 2, 261–290.
[41] M. Strichartz, Restrictions of Fourier transforms to quadratic surfaces and decay of solu-
tions of wave equations, Duke Math. J. 44 (1977), 705–714.
[42] E. Ryckman and M. Visan, Global well-posedness and scattering for the defocusing energy-
critical nonlinear Schrödinger equation in R1+4 , Amer. J. Math. 129 (2007), 1–60.
[43] W. Strauss, Nonlinear scattering theory, Scattering Theory in Mathematical Physics, ed.
La Vita and Marchard, Reidel, (1974), 53–78.
[44] C.Sulem, and P-L. Sulem, The nonlinear Schrödinger equation. Self-focusing and wave
collapse, Springer-Verlag, New York, 1999.
[45] E. M. Stein, Singular integrals and differentiability properties of functions, Princeton Univ.
Press, 1967.
[46] T. Tao, Nonlinear dispersive equations. Local and global analysis CBMS 106, eds: AMS,
2006.
[47] T. Tao, M. Visan, and X. Zhang The nonlinear Schrödinger equation with combined power-
type nonlinearities, Comm. Partial Differential Equations 32 (2007), no. 7-9, 1281–1343.
[48] T. Tao, M. Visan, and X. Zhang, Global well-posedness and scattering for the mass-critical
nonlinear Schrödinger equation for radial data in high dimensions, Duke Math. J. 140
(2007), 165–202.
[49] Y. Tsutsumi, L2 solutions for nonlinear Schrödinger equations and nonlinear groups, Funk-
cial. Ekvac. 30 (1987) 115–125.
[50] M. Visan, The defocusing energy-critical nonlinear Schrodinger equation in dimensions five
and higher, Ph. D. Thesis UCLA, 2006.
[51] M. I. Weinstein, Nonlinear Schrödinger equations and sharp interpolation estimates, Comm.
Math. Phys. 87 (1983), 567–576.

N. Pavlović, Department of Mathematics, University of Texas at Austin.

E-mail address: natasa@math.utexas.edu

N. Tzirakis, University of Illinois at Urbana-Champaign.

E-mail address: tzirakis@math.uiuc.edu