You are on page 1of 29

SPANISH LECTURES ON NLS

LUCA FANELLI

Abstract. We recollect here the notes of a short introductive course


on the nonlinear Schrödinger equation we gave in May 2010 at the Uni-
versidád Autonoma de Madrid.

Contents
1. Introduction 1
2. Dispersive properties of the free Schrödinger group 2
2.1. Exercises 11
3. The local Cauchy problem for NLS 11
3.1. A scaling argument 12
3.2. Local existence: fixed point arguments 12
3.3. Exercises 16
4. The focusing NLS with data in H 1 17
4.1. Energy conservation 17
4.2. Exercises 19
5. Morawetz identities and local smoothing 20
5.1. Exercises 24
6. Interaction Morawetz identities 24
6.1. Applications: a scattering result 26
6.2. Exercises 27
6.3. Some obviously incomplete bibliographical notes 28
References 28

Antongiulio y Giacomo,
bienvenidos a este mundo.

1. Introduction
These are the notes of a short Ph.D. course of five lectures, given in
May 2010 at the Universidád Autonoma de Madrid. Our aim is to give an
introduction about the mathematical techniques and tool which have been
developed in the last years to treat some problem arising in the study of the
nonlinear Schrödinger equation (NLS). In what follows, we will refer to NLS
each time we consider an equation of the form
i∂t u(t, x) + ∆x u(t, x) ± |u|p−1 u(t, x) = 0, (1.1)

Date: May 19, 2010.


1
SPANISH LECTURES ON NLS 2

for the unknown u = u(t, x) : R1+n → C. It is natural to consider equation


(1.1) as a perturbation of the linear Schrödinger equation
i∂t u(t, x) + ∆x u(t, x) = 0, (1.2)
which is the Quantum Mechanics model to describe the evolution of a free
particle in a nonrelativistic regime. Equation (1.2) outstanding with respect
to the standard classification of PDEs into Ellyptic, Parabolic and Hyper-
bolic. On the other hand, as we will explain in the following, it is a relevant
example in the class of dispersive equation, which somehow transversally
cross the above classification. The problems, concerning (1.1), which will be
treated during these lectures are the following ones:
• local (in time) well posedness in suitable Banach spaces;
• global existence vs finite-time blow-up;
• scattering
As a first approach, one should see the NLS (1.1) as a nonlinear perturbation
of (1.2); our first aim is indeed to introduce some basic fact about the linear
equation (1.2), and then use them to solve some of the previous problems in
a perturbative regime. Then, we will pass to consider equation (1.1) in an
intrinsical way; we will introduce some relevant algebraic identities which
turn out to be useful in some critical regimes.
Acknowledgments. I wish to thank Alberto Ruiz for inviting me to give this
course in Madrid. I also sincerely wish to thank all the people who attended
the course, because I learnt a lot while preparing the lectures and these
notes, which are dedicated to them.
I wish also to thank Piero D’Ancona and Luis Vega for some useful advices
about the topic of this course.
Finalmente, agradezco encarecidamente su amistad a Juan Antonio Bar-
celó; gracias a él, cada vez que estoy en Madrid me siento como en casa.

2. Dispersive properties of the free Schrödinger group


Our first lecture is devoted to the Cauchy problem for the linear Schrödinger
equation, i.e. (
(i∂t + ∆) u = 0
(2.1)
u(0, x) = u0 (x)
in the whole space R1+n , n ≥ 1, where the initial datum u0 has to be
chosen in a suitable functional space. Since the operator −∆ is selfadjoint on
L2 (Rn ; C) with form domain H 1 (Rn ; C), it is possible to define, via Spectral
Theorem the functional calculus f (∆), where f : R → C is any Borel-
measurable function. As a consequence, we can define the Schrödinger group
S = S(t) : R → B(H 1 ), S(t) = eit∆ ,
which uniquely defines the solution u of (2.1) as
u(t, ·) = S(t)u0 (·).
Of course u0 has to be chosen in the domain of −∆ and the initial condition
is satisfied in the following sense:
H 1 − lim u(t, ·) = u0 (·).
t→0
SPANISH LECTURES ON NLS 3

On the other hand, using Fourier transform is an equivalent way to proceed.


Let us denote by v(t, ξ) the space Fourier transform of u(t, x), i.e.
Z
1
v(t, ξ) = e−ixξ u(t, x) dx;
(2π)n
if u solves (2.1) and the datum u0 is Fourier transformable, then v is a
solution of the ordinary problem
(
i∂t v = |ξ|2 v
(2.2)
v(0, ξ) = û0 (ξ),
û0 being the Fourier transform of u0 . The unique solution of the previous
problem is hence given by
2
v(t, ξ) = û0 (ξ)e−it|ξ| .
Consequently, if u0 is chosen in a space in which the inverse Fourier formula
holds, we can invert the Fourier transform and define the unique solution u
of (2.1) as
Z
−1 −n 2
u(t, x) = Fξ→x v(t, ξ) = (2π) ei(x·ξ−t|ξ| ) û0 (ξ) dξ. (2.3)

We can recollect all the previous information, to get now the following fun-
damental Lemma.
Lemma 2.1. Let u0 ∈ L2 (Rn ) and denote by
1 i|x|2
Kt (x) = n e 4t , t 6= 0, x ∈ Rn . (2.4)
(4πit) 2
Then, the unique solution u ∈ C(R \ {0}; L2 (Rn )) of (2.1) is given by
Z
1 i|x|2 ix·y i|y|2
u(t, x) = Kt ∗ u0 = n e 4t e− 2t e 4t u0 (y) dy. (2.5)
(4πit) 2
Proof. A direct computation shows that
b t (ξ) = e−it|ξ|2 ;
K (2.6)
as a consequence, denoting by v(t, ξ) the space Fourier transform of the
function u = Kt ∗ u0 , we have
b t (ξ) · û0 (ξ) = û0 (ξ)e−it|ξ|2 ,
v(t, ξ) = K
which proves the claim in view of the previous considerations. ¤
Remark 2.1. Observe again that the initial condition is attained in the fol-
lowing sense:
L2 − lim u(t, ·) = u0 (·).
t→0
These considerations show that the space L2 and, more in general, any
Hilbert space H s , is the natural setting for the Schrödinger equation. In
fact, due to the imaginary unit in the Schrödinger operator, the group eit∆
is unitary on these spaces, in which the solution, as a consequence, well
propagates (in both the directions of time).
SPANISH LECTURES ON NLS 4

Remark 2.2. The convolution formula u = Kt ∗ u0 shows that the free


Schrödinger equation has infinite speed of propagation, in the following
sense: if the initial datum u0 is compactly supported, then the support
of the solution is the whole space, for all times t 6= 0.
We hence now pass to a precise study of the convolution kernel Kt , in order
to obtain a priori informations on the solution of (2.1). A first important
result is the following.
Theorem 2.2 (time-dispersion). Let 2 ≤ q ≤ ∞ and q 0 be its dual exponent.
Then, the Schrödinger group S(t) = eit∆ satisfies the estimate
³ ´
1
−n − 1q
kS(t)u0 kLq . |t| 2
ku0 kLq0 , (2.7)
for any t 6= 0, where n is the dimension of space.
We remark that the symbol . means that the inequality holds modulo a
positive constant C > 0 which is independent of t and u0 .

\0 (ξ) =
Proof. First observe that the case q = 2 is trivial; indeed, since S(t)u
−it|ξ| 2
e û0 (ξ), by Plancherel identity we have
kS(t)u0 kL2 = ku0 kL2 , (2.8)
for all times t ∈ R. Moreover, by (2.4) we see that Kt ∈ L∞ , for all times
n
t 6= 0, and kKt kL∞ . |t|− 2 ; hence, by Young’s Theorem we can estimate
n
kS(t)u0 kL∞ = kKt ∗ u0 kL∞ . |t|− 2 ku0 kL1 . (2.9)
Estimate (2.7) now follows by interpolation (Riesz-Thorin Theorem) be-
tween (2.8) and (2.9). ¤
Remark 2.3. Estimates (2.7) affirm that the spacial Lp -norms of a free
Schrödinger solution, for q ≥ 2, decay in time, with a precise decay-rate
depending on n and q. We remark that this is a property which does not
characterize the class of dispersive equations. For example, also some para-
bolic equations satisfy this kind of a priori estimates (see e.g. the so called
strongly contractive estimates for the Heat Equation). We will spend some
more precise words about the nomenclature dispersive equations in the fol-
lowing lectures.
Although estimates (2.7) represent a precise information about solutions,
they are not ready to be used to treat nonlinear perturbations of (2.1). On
the other hand, as we see in a while, they have some strong consequences
about the boundedness of some global norms (in space and time) of solutions.
We are now ready to introduce the so called Strichartz estimates. Let us
first give a preliminary definition.
Definition 2.3. Let n ≥ 2; we say that a couple of numbers (p, q) is
Schrödinger-admissible if the following relations hold:
2 n n
= − , p ≥ 2, (p, q, n) 6= (2, ∞, 2). (2.10)
p 2 q
SPANISH LECTURES ON NLS 5

Let n = 1; we say that a couple of numbers (p, q) is Schrödinger-admissible


if the following relations hold:
2 1 1
= − , p ≥ 4. (2.11)
p 2 q
2n
Moreover, we will call endpoint pairs the couples (p, q) = (2, n−2 ), in dimen-
sion n ≥ 2, and (p, q) = (4, ∞) in dimension n = 1.
Remark 2.4. Notice that condition (2.10) says that the endpoint pair is not
an admissible pair in dimension n = 2, while it is in all the other dimensions.
Let us now introduce a notation. For a Borel-measurable function f =
f (t, x) : I × Rn → C we define
ÃZ µZ ¶p !1 p
q
kf kLp Lq = kf kLp (I;Lq ) = |f |q dx ,
I

if p, q 6= ∞, and the analogous definition in the case in which one of the


exponents or both of them are infinite. In the case I = R we will omit the
subindex I, just writing kf kLp Lq .
We can now state the main result of this lecture.
Theorem 2.4 (Strichartz Estimates). Let (p, q), (p̃, q̃) be two Schrödinger-
admissible pairs; then, the Schrödinger group satisfies the estimates
keit∆ u0 kLp Lq . ku0 kL2 , (2.12)
°Z °
° °
° e −it∆
F (t, ·) dt°
° ° 2 . kF kLpe0 Lqe0 (2.13)
L
°Z t °
° °
° ei(t−s)∆ F (s, ·) ds° . kF kLpe0 Lqe0 (2.14)
° °
0 Lp Lq

Remark 2.5. At the left-hand side of estimate (2.13) the acting operator is
the dual of the Schrödinger group S(t). Indeed, by estimate (2.12) we have
that
S : L2 3 f 7→ S(t)f ∈ Lp Lq ;
by duality, (2.12) and (2.13) are equivalent to each other, since the adjoint S ∗
with respect to the inner product in L2 is given by S ? (t) = e−it∆ . Observe
also that Z Z
i(t−s)∆
e F (s, ·) ds = S(t)S ∗ (s)F ds.

Hence, a part of the time truncation in the time interval, at the left-hand
side of (2.14) the operator SS ∗ appears. Estimate (2.14) is usually called
T T ∗ -estimate (change the nomenclature S for T in your mind). Finally, we
strongly remark that the pairs (p, q), (p̃, q̃) are not related to each other in
the T T ∗ -estimate, and this turns out to be a crucial fact for the nonlinear
applications. In fact, one should see that the solution of the inhomogeneous
problem (
i∂ut (t, x) + ∆u(t, x) = F (t, x)
(2.15)
u(0, x) = u0 (x)
SPANISH LECTURES ON NLS 6

can be written in terms of the Duhamel formula


Z t
it∆
u(t, ·) = e u0 (·) + i ei(t−s)∆ F (s, ·) ds. (2.16)
0
Remark 2.6. Look to the explicit formula (2.5). It says that, up to a scaling
and a multiplication by a complex number of modulus 1, the Schrödinger
group S(t) is nothing else than the Fourier transform. As a consequence,
if u0 ∈ L2 , then the solution is in L2 for all times. On the other hand,
Strichartz estimates (2.12) are giving a stronger information. In fact, as
a consequence of them, the solution is in Lqx , for 2 ≤ q ≤ 2n/(n − 2),
for almost all times t 6= 0 (we also remark that this is not true for t = 0
and we cannot change the sentence almost all times with all times). In
fact, Strichartz estimates represent a smoothing effect for free Schrödinger
solutions, where the gain here is obtained, via the action of S(t), in terms
of global Lq -regularity in space, during the time evolution.
For the non-endpoint case, we will present in a while a very simple proof
of Theorem 2.4, based on some deep and well known results in Harmonic
Analysis, which we now pass to recall. For the endpoint case, we will prove
a more general result including a larger class of operators groups, which is
due to Keel and Tao.
We first introduce some preliminary results.
Lemma 2.5 (Christ & Kiselev [6]). Let X, Y be two Banach spaces, and let
T be a bounded linear operator from Lβ (R; X) to Lγ (R; Y ) of the form
Z
T f (t) = K(t, s)f (s) ds.

Denote by Te the truncated operator


Z t
e
T f (t) = K(t, s)f (s) ds.
t0

Assume that β < γ; then Te is bounded from Lβ (R; X) to Lγ (R; Y ), with


³ ´
1
− 1 −1
kTek ≤ cβ,γ kT k and cβ,γ = 1 − 2 γ β .

For the proof, see [6]. Now we recall a well known result about fractional
integrals.
Lemma 2.6 (Hardy-Littlewood-Sobolev inequality). Let f : Rn → C be a
measurable function and 0 < γ < n; then
° °
°f ∗ |y|−γ ° p ≤ Cp,γ kf kLr , (2.17)
L
1 1 γ
with p = r + n − 1.
Indeed, the constant Cp,γ in (2.17) is bounded by the weak Lp -norm of
the function |y|−γ . For a simple proof of (2.17), see e.g. (CITARE STEIN).
Finally, we need an abstract functional analysis result.
Lemma 2.7. Let H be a Hilbert space and X be a Banach space, and X ∗ be
its dual space. Let D ⊂ X be a dense vector space in X and T ∈ L(D; H) be
a densely defined linear operator, with its adjoint operator T ∗ ∈ L(H; D∗ ).
Then the following three properties are equivalent:
SPANISH LECTURES ON NLS 7

• there exists a constant C ∈ [0, ∞) such that kT f kH ≤ Ckf kX , for


all f ∈ D;
• the range of T ∗ is dense in X ∗ and there exists a constant C ∈ [0, ∞)
such that kT ∗ vkX ∗ ≤ CkvkH , for all v ∈ H;
• the range of T ∗ T is dense in X ∗ and there exists a constant C ∈
[0, ∞) such that kT ∗ T f kX ∗ ≤ C 2 kf kX , for all f ∈ D.
The constant C is the same in the previous three points. Moreover, if one of
these three conditions holds, then the operators T and T ∗ T extend to bounded
operators from X to H and from X to X ∗ , respectively.
We leave the proof of Lemma 2.7 as an exercise for the reader. We can
now pass to the proof of the non-endpoint Strichartz estimates.
Proof of Theorem 2.4: the non-endpoint case. We divide the proof into two
steps.
Step 1. Let us first prove the T T ∗ estimate (2.14) in the case (p, q) =
(p̃, q̃). Since p > 2, we have p > p0 . As a consequence, we can apply the
0
Christ-Kiselev Lemma 2.5 with the choices β = p, γ = p0 , X = Lq , Y = Lq .
Hence, in order to prove (2.14), it is sufficient to prove the same estimate
for the untruncated integral.
Writing ei(t−s)∆ = eit∆ e−is∆ , by (2.7) and the Schrödinger-admissibility
condition (Definition 2.3) we can estimate
°Z ° °Z °
° ° ° −n +n °
° ei(t−s)∆ F (s, ·) ds° ° kF (s, ·)kLq °
° p q . ° |t − s|
2 q 0
° ° p
L L L
° °
° − p2 °
= °|t| ∗ kF (s, ·)kLq0 ° p .
L
Hence by Hardy-Littlewood-Sobolev we conclude that
°Z °
° °
° ei(t−s)∆ F (s, ·) ds°
° ° p q . kF kLp0 Lq0 . (2.18)
L L
Step 2. The proof is almost done. Indeed, arguing by density, estimates
(2.12) and (2.13) now follow by (2.18) and Lemma 2.7. Finally, use (2.12)
and (2.13) to come back to (2.14). Again, it is sufficient to estimate the
untruncated integral; mixing (2.12) and (2.13) with to different norms, one
can immediately complete the proof of (2.14) in the case (p, q) 6= (p̃, q̃). ¤
Remark 2.7 (Life is hard). Actually, the notations in Lemma 2.7 seem to be
unlucky, because one should because one should call T ∗ (t) = eit∆ ; on the
other hand, the reader will easily understand that there is no way out to
solve this meta-semantical problem.
We now proceed with the Keel and Tao’s Theorem on endpoint Strichartz
estimates.
Theorem 2.8 (Keel & Tao). Let H be an Hilbert space, D ⊂ H a dense
vector subspace and U (t) ∈ B(D, L2 ) be a family of operators such that
kU (t)f kL∞ 2 ≤ C0 kf kH ,
t Lx
∀f ∈ D. (2.19)
moreover, assume that for some σ > 0 and for almost all t, s ∈ R,
kU (t)U ∗ (s)gkL∞
x
≤ C12 |t − s|−σ kgkL1x , (2.20)
SPANISH LECTURES ON NLS 8

for any g ∈ L2 ∩ L1 . Then, for any (p, q), (p̃, q̃) satisfying
1 σ σ
+ = , p≥2 (p, q, σ) 6= (2, ∞, 1), (2.21)
p q 2
the following estimates hold:
kU (t)f kLpt Lqx ≤ C2 kf kH (2.22)
°Z °
° °
° U ∗ (s)F (s) ds° ≤ C2 kF k p0 q0 (2.23)
° ° Lt Lx
H
°Z °
° t °
° U (t)U ∗ (s)F (s) ds° ≤ C22 kF kLp̃0 Lq̃0 . (2.24)
° ° x
0 Lpt Lqx t

Before proving the previous theorem we need to recall the following in-
terpolation lemma.
Lemma 2.9 ([2], 3.13.5 (b)). Let A0 , A1 , B0 , B1 , C0 , C1 be Banach spaces
and let T be a linear operator such that


A0 × B0 → C0
T : A0 × B1 → C1


A1 × B1 → C1 .
Then
T : (A0 , A1 )θ0 ,pr × (B0 , B1 )θ1 ,qr → (C0 , C1 )θ,r ,
for all p, q, r ∈ [1, ∞] such that
1 1
+ ≥ 1, θ0 , θ1 , θ ∈ (0, 1), θ = θ0 + θ1 .
p q
Proof of Theorem 2.8. By the arguments we introduced for the proof of The-
orem 2.4, it is sufficient to prove 2.23. Indeed, as it will be clearRduring the
t
proof, we will prove it also in the case of the³truncated
´ integral 0 dt. Here

we will just prove the endpoint case (p, q) = 2, σ−1 ; we leave to the reader
the proof in the non-endpoint case, which is completely analogous to the one
for Theorem 2.4.
Let us introduce the quadratic form
Z Z
T (F, G) = (U ∗ (s)F (s), U ∗ (t)G(t))H ds dt.

One could easily check that the endpoint version of (2.23) is equivalent to
the following estimate:
|T (F, G)| ≤ C22 kF k 2σ kGk 2σ .
L2t Lxσ+1 L2t Lxσ+1
By symmetry, we just need to consider the double integral in the zone where
s < t. For any j ∈ Z, let us define
Z Z
Tj (F, G) = (U ∗ (s)F (s), U ∗ (t)G(t))H ds dt;
2j ≤t−s<2j+1
Hence we reduced to prove the following claim:
X
|Tj (F, G)| ≤ C22 kF k 2σ kGk
σ+1
2σ . (2.25)
L2t Lx L2t Lxσ+1
j∈Z
SPANISH LECTURES ON NLS 9

Proof of (2.25). By a change of variables, we can write


Z ÃZ 2j+1 !
∗ ∗
Tj (F, G) = (U (s)F (s), U (s + θ)G(s + θ))H dθ ds; (2.26)
2j

hence, by (2.20), Fubini’s Theorem, Cauchy-Schwartz in ds and Hölder L1 −


L∞ in dθ we can easily estimate
|Tj (F, G)| ≤ 2−j(σ−1) kF kL2t L1x kGkL2t L1x , (2.27)
for all j ∈ Z. Analogously, by Cauchy-Schwartz and (2.19) we obtain
¯Z Ã Z 2j+1 ! ¯
¯ ¯
¯ ¯
|Tj (F, G)| = ¯ U ∗ (s)F (s), U ∗ (s + θ)G(s + θ) dθ ds¯
¯ 2j ¯
Z °Z j+1 ° H
° 2 °
° °
≤ kF (s)kL2 · ° U ∗ (s + θ)G(s + θ) dθ° ds.
° 2j °
H
Consequently, by using (2.23) in the non-endpoint case, applied to the func-
e + θ) = G(s + θ)χ[2j ,2j+1 ) (θ), we obtain
tion G(s
Z
|Tj (F, G)| ≤ kF (s)kL2 · kG(s + θ)kLp0 ([2j ,2j+1 ];Lq0 ) ,
θ x

for any non-endpoint couple (p, q), and by Hölder in θ (notice that p0 < 2)
we get
³ ´Z
j σq − σ−1
|Tj (F, G)| ≤ 2 2
kF (s)kL2 · kG(s + θ)kL2 ([2j ,2j+1 ];Lq0 ) .
θ x

We now apply Cauchy-Schwartz in ds and Fubini’s Theorem, to obtain


³ ´ ÃZ ÃZ j+1 ! !1
2 2
j σ
− σ−1
|Tj (F, G)| ≤ 2 q 2
kF kL2t L2x · kG(s + θ)k2Lq0 dθ ds ;
2j

finally, by Hölder L1 − L∞ in dθ it turns out that


³ ´
σ
−j − σq −1
|Tj (F, G)| ≤ 2 2
kF kL2t L2x kGkL2 Lq0 , (2.28)
t x

for any q ∈ [2, 2σ/(σ − 1)). By symmetry (just call now t = s − θ in Tj ), we


also obtained
³ ´
σ
−j − σq −1
|Tj (F, G)| ≤ 2 2
kF kL2 Lq0 kGkL2t L2x . (2.29)
t x

Now we resume the previous estimates. Define lps = Lp (Z; 2js dj); estimates
(2.27), (2.28), (2.29) can be recollected as follows:
0 0
Tj : L2t Lax × L2t Lbx → l∞
β(a,b)
, (2.30)
provided
σ σ
β(a, b) = σ − 1 − − , (2.31)
a b
(
a=b=∞ hest. (2.27),
´ h ´
2σ 2σ
(a, b) ∈ {2} × 2, σ−1 ∪ 2, σ−1 × {2}, est. (2.28), (2.29).
(2.32)
SPANISH LECTURES ON NLS 10

Since (L2t Lpx0 , L2t Lpx1 )θ,s = L2t Lp,s


x , being L
p,s the standard Lorentz space, it

immediately follows by (2.30) and Lemma 2.9 (with the choices r = 1, p =


q = 2) that
0 0 β(a,b)
Tj : L2t La ,2 × L2t Lb ,2 → l1 , (2.33)
for any couple (a, b) such that (1/a, 1/b) is in the convex hull of the points
(0, 0), (1/2, 1/2), (1/2, (σ − 1)/2σ), ((σ − 1)/2σ), 1/2), in which the endpoint
(a, b) = (2σ/(σ − 1), 2σ/(σ − 1)) is included.
Finally, by the embedding Lp,r ⊂ Lp,s , if r ≤ s, (2.33) implies
0 0 β(a,b)
Tj : L2t La × L2t Lb → l1 , (2.34)
³ ´
2σ 2σ
and the claim (2.25) is proved, once observed that β σ−1 , σ−1 = 0. ¤

We conclude this lecture with s-admissible. Let us introduce the notation


kf kḢ s := k |D|s f kL2 ,
where |D|s = F −1 (|ξ|s F), and F, F −1 are the Fourier transform and its
inverse. With these notations, we have that −∆ = |D|2 . Since eit∆ com-
mutes with |D|s for any positive s ≥ 0 (which is immediate to see passing
in Fourier), by estimate (2.12) it follows that
k |D|s eit∆ u0 kLp Lr . ku0 kḢ s , (2.35)
for any s ≥ 0, where (p, r) is any Schrödinger admissible couple. Now, by
Sobolev embedding, we have that
keit∆ u0 kLp Lq . k |D|s eit∆ u0 kLp Lr , (2.36)
provided
n n n
s− =− , s< .
r q r
It is hence natural to give the following definition.
Definition 2.10 (s-admissible pairs). Let n ≥ 2 and s ≥ 0; we say that a
couple of numbers (p, q) is (Schrödinger) s-admissible if the following rela-
tions hold:
2 n n
= − − s, p ≥ 2, (p, q, n) 6= (2, ∞, 2). (2.37)
p 2 q
Let n = 1; we say that a couple of numbers (p, q) is (Schrödinger) s-
admissible if the following relations hold:
2 1 1
= − − s, p ≥ 4. (2.38)
p 2 q
Observe that the lower bound on p, in both cases n = 1 and n ≥ 2 gives
on upper bound on q and s. It is now immediate to prove the following
result, as a corollary of Theorem 2.4.
Theorem 2.11 (s-admissible Strichartz Estimates). Let (p, q), (p̃, q̃) be two
Schrödinger s-admissible pairs; then, the Schrödinger group satisfies the es-
timates
keit∆ u0 kLp Lq . ku0 kḢ s , (2.39)
°Z °
° °
° e−it∆ F (t, ·) dt° . k |D|s F k pe0 qe0 (2.40)
° ° L L
L2
SPANISH LECTURES ON NLS 11
°Z t °
° °
° ei(t−s)∆ F (s, ·) ds° . k |D|2s F kLpe0 Lqe0 (2.41)
° °
0 Lp Lq

2.1. Exercises.
Exercise 2.12. Prove identity (2.6).
Exercise 2.13. Formally prove that formula (2.16) defines a solution of the
inhomogeneous problem (2.15).
Exercise 2.14. Prove Lemma 2.7.
Exercise 2.15. Define Lorentz spaces as real interpolation of Lp -spaces,
namely
Lp,q = (Lp0 , Lp1 )θ,q , p−1 = (1 − θ)p−1 −1
0 + θp1 ,
provided
p0 < p 1 , p0 < q ≤ ∞, 0 < θ < 1.
Let (p, q), (p̃, q̃) be two non-endpoint Schrödinger admissible couples and r, s
satisfying
p̃0 ≤ s0 < r ≤ p, r ≥ 2.
Prove that
keit∆ f kLp,r Lq,r . kf kL2 ,
°Z °
° °
° e−it∆ F (t, ·) dt° . kF k p̃0 ,s0 q̃0 ,s0 ,
° ° 2 L L
L
°Z t °
° °
° e i(t−s)∆
F (s, ·) ds° . kF kLp̃0 ,s0 Lq̃0 ,s0 .
° °
0 Lp,r Lq,r
Be careful, because
(Lp0 Lq0 , Lp1 , Lq1 )θ,r 6= Lp,r Lq,r ,
except for a few cases (see [9]). Hence the previous estimates do not follow
by interpolation of standard Strichartz estimates.
Hint: Interpolate between time-dispersive estimates and then estimate
convolutions in Lorentz spaces (see [18]). For further details about Lorentz
spaces, see [2].

3. The local Cauchy problem for NLS


We now pass to study the local (in time) Cauchy problem for the nonlinear
Schrödinger equation
(
i∂t u + ∆u = ±|u|γ−1 u
(3.1)
u(0, x) = u0 (x),

where u = u(t, x) : R1+n → C and u0 is assumed to be in a suitable Hilbert


space which will be specified each time. Let us first introduce some heuristics
to conjecture for which values of γ one should expect to have local well
posedness for (3.1).
SPANISH LECTURES ON NLS 12

3.1. A scaling argument. Let us observe that the equation in (3.1) is


invariant under the scaling
µ ¶
2 t x
uλ (t, x) := λ 1−γ u , ,
λ2 λ
for any λ > 0, i.e. if u is a solution of (3.1) then uλ solves the same equation
with a scaled initial datum. Notice the lack of scaling invariance in the case
γ = 1. By an explicit computation we have that, for any s ≥ 0,
kuλ (t, ·)kḢ s = λc(n,s,γ) ku(t, ·kḢ s , (3.2)
where
2 n − 2s
c(n, s, γ) = + . (3.3)
1−γ 2
4
Hence we have that c(n, s, γ) ≤ 0 if γ ≤ 1 + n−2s =: γc,s . The value γc,s
is consequently expected to be s-critical. Indeed, assume the initial datum
is in Ḣ s , assume γ > γc,s , and suppose that a local (in time) solution u ∈
C([0, T ); Ḣ s )of (3.1) exists; it is hence possible to construct a scaled solution
with arbitrarily large Ḣ s -norm, at arbitrary times, by sending λ → ∞ in
the previous computation. Then we give the following definition.
Definition 3.1. Let n ≥ 1, s ≥ 0 and γ > 1; we say that
4
• γ is s-subcritical if 1 < γ < 1 + n−2s
4
• γ is s-critical if γ = 1 + n−2s
4
• γ is s-supercritical if γ > 1 + n−2s .
We give three important examples:
(1) the quintic NLS in dimension 1
i∂ut + ∂x2 u = ±|u|4 u
is L2 -critical
(2) the quintic NLS in dimension 3
i∂ut + ∆u = ±|u|4 u
is Ḣ 1 -critical (also said energy-critical)
(3) the cubic NLS in dimension 3
i∂ut + ∆u = ±|u|2 u
1
is Ḣ 2 -critical.

3.2. Local existence: fixed point arguments. We now pass to state


some Theorems in which, thanks to the linear techniques which have been
introduced in the previous section, we prove local well-posedness for some
nonlinear Schrödinger equation. We first specify which kind of solutions we
are searching for. We give the following definition.
Definition 3.2. Let X be a Banach space, u0 ∈X and let I ⊂ R be a closed
time interval containing 0. Define the map
Z t
it∆
T v = e u0 + i ei(t−s)∆ |v|γ−1 v(s, ·) ds, (3.4)
0
SPANISH LECTURES ON NLS 13

for t ∈ I. We say that u ∈ C(I; X) is a local solution of (3.1) if u is a fixed


point of the map T , i.e. T u = u and
X − lim u(t, ·) = u0 .
t→0

Before stating the first theorems about local well-posedness, let us state
the following Lemma.
Lemma 3.3. Let I ⊂ R be a closed time interval containing 0, and u ∈
C(I; L2 (Rn )) be a local solution of (3.1). Then
ku(t, ·)kL2x = ku0 kL2 , (3.5)
for all times t ∈ I.
Proof. Assuming first that the solution u ∈ C(I; H 1 ), then (3.5) follows by
integration by parts. Then, the general statement can be proved by density
arguments. ¤
Let us start with the L2 -theory for subcritical nonlinearities. In this case,
we can state a global existence result.
Theorem 3.4 (L2 -subcritical NLS). Let u0 ∈ L2 (Rn ), n ≥ 1 and 1 < γ <
1 + n4 . Then there exists a unique global solution u ∈ C(R; L2 (Rn )).
Proof. We search for a fixed point of the Duhamel operator T defined in
(3.4). Let I ⊂ R be an open interval containing 0, and (p, q), (p̃, q̃) be two
Schrödinger 0-admissible pairs to be chosen later; moreover, define
p q
X = L∞ 2
I L ∩ LI L , k · kX = k · kL∞
I L
2 + k · kLp Lq .
I

By Strichartz estimates (2.12), (2.14) we hence estimate


kT ukX . ku0 kL2 + k|u|γ kLp̃0 Lq̃0 = ku0 kL2 + kukγ γ p̃0 . (3.6)
I LI Lγ q̃0

Hence, by Hölder inequality in time we have


kT ukX . ku0 kL2 + |I|θ · kukγLr Lγ q̃0 , (3.7)
I

where |I| is the measure of the interval I and


1 1 1 γ
0
= + , θ= . (3.8)
γ p̃ r s s
Now we impose that the couple (r, γ q̃ 0 ) is Schrödinger 0-admissible, i.e.
2 n n
= − 0. (3.9)
r 2 γ q̃
An explicit computation shows that conditions (3.8) and (3.9) are consistent
with the Schrödinger admissibility of (p̃, q̃) if and only if 1 < γ < 1 + n4 .
This proves that, for a suitable choice of the exponents p, q, p̃, q̃, the following
estimate holds:
kT ukX . ku0 kL2 + |I|θ kukγX . (3.10)
As a consequence, since θ > 0, we have proved that, for a sufficiently small
interval I, the operator T maps a ball in X into itself. With an analogous
computation, we see that T is a contraction on X, and this completes the
proof of the local existence.
SPANISH LECTURES ON NLS 14

Now, since the mass (i.e. the L2 -norm) is conserved, as stated in Lemma
3.3, as an immediate consequence 3.4 we have that the maximal existence
interval is in fact I = R, in other words the solution is global in time. In
fact, denote for example by T+ > 0 the maximal positive existence time
and assume by contradiction that T+ < ∞. By the mass conservation, we
have limt→T+ ku(t, ·)kL2 = ku0 kL2 < ∞; hence we can start again with the
fixed point argument and find a bigger positive lifespan T = T+ + T1 , which
contradicts the fact that T+ is maximal. ¤
Remark 3.1. In fact, the proof of Theorem 3.4 shows that the local solution
which has been constructed has got an additional regularity information;
indeed, we proved that u ∈ L∞ (R; L2 ) ∩ Lp (R; Lq ), with the best possible
choice of p, q.
Remark 3.2. Notice that the continuity in time of the solution is immediate
a posteriori, once it has been obtained as a fixed point of the map (3.4).
We now pass to the study of the L2 -critical case. We first state a global
result with small data.
Theorem 3.5 (L2 -critical NLS with small data). Let u0 ∈ L2 (Rn ), n ≥ 1
and γ = 1 + n4 . There exists a small ²0 > 0 depending on n and u0 , for
which, if
ku0 kL2 < ²0 , (3.11)
then the Cauchy problem (3.1) has a unique global solution u ∈ C(R; L (Rn )).
2

Proof. Let us divide the proof into two steps.


Step 1. Let I ⊂ R be an open interval containing 0, and let (p, q) be a
Schrödinger admissible pair to be chosen later. We first construct a local
solution in S = LpI Lq . We proceed in an analogous way to the subcritical
case. Let us now start by estimate (3.6), where we substitute X by S. We
impose the 0-admissibility of the couple (γ p̃0 , γ q̃ 0 ), i.e.
2 n n
0
= − 0, (3.12)
γ p̃ 2 γ q̃
which together with the admissibility of (p̃, q̃) is satisfied if and only if γ =
1 + n4 . As a consequence we obtain the estimate
4
1+ n
kT ukS . ²0 + kukS . (3.13)
Now denote by δ := kukS , and observe that the inequality
4
δ . ²0 + δ 1+ n
has always a range of solutions in δ, provided ²0 is sufficiently small. This,
together with (3.13), implies that a ball with a suitable radius δ > 0 in S is
mapped into itself by T .
In a completely analogous way, we prove that T is a contraction on S,
hence there exists a unique fixed point u ∈ S, which locally solves the Cauchy
problem.
Notice that the maximal interval for which this construction can be done
is Imax = R, since the smallness ²0 is obtained just in terms of the initial
datum. Hence, for small initial data, we constructed a global solution u ∈ S.
SPANISH LECTURES ON NLS 15

Step 2. In order to conclude the proof, we just need to prove that the
solution we constructed in fact also in L∞ 2
R L . To do this, let I = (−T0 , T0 ),
for an arbitrary T0 > 0. By the previous step, we have that kukS < ∞,
where S = LpI Lq for a suitable choice of (p, q). Now we come back to (3.4)
and estimate by Strichartz
4
1+ n
kT ukL∞
I L
2 . ku0 kL2 + kuk
S .
Hence kukL∞
I L
2 < ∞, and the proof is complete, since T0 is arbitrary. ¤
Remark 3.3. Notice that the previous proof does not ensure kukS < ∞ in
the case I = R. In other words, we cannot prove that a Strichartz norm is
bounded in the whole lifespan of the solution with simple techniques. This
is a very difficult problem and, as we see in the following, it turns out to be
crucial to prove global existence with large data.
In view of the previous remark, we now prove the following Theorem.
Theorem 3.6 (L2 -critical NLS with large data). Let ku0 kL2 = A < ∞ and
γ = 1 + n4 . There exist a maximal interval I = (−T∗ , T ∗ ), with T∗ , T ∗ > 0,
such that problem (3.1) has a unique solution u ∈ C((−T∗ , T ∗ ); L2 (Rn )).
Proof. The proof is analogous to the previous case. With the same notations
of Theorem 3.5, by Strichartz estimate we obtain
4
1+ n
kT ukS . keit∆ u0 kS + kukS . (3.14)
Now we need to obtain the smallness ²0 by another argument. Since u0 ∈ L2 ,
we have by Strichartz that keit∆ u0 kS < ∞. Hence, by the absolute conti-
nuity of the Lebesgue integral, if the interval I is chosen to be sufficiently
small, i.e. |I| ≤ ², for a suitable ² > 0, we have that keit∆ u0 kS < ²0 , for
some small ²0 depending on u0 and the constant in the Strichartz estimate.
Consequently, we obtain
4
1+ n
kT ukS . ²0 + kukS . (3.15)
Now the proof follows exactly as in the previous case. ¤
Remark 3.4. Notice that now we cannot ensure that the maximal existence
interval is the whole line R. On the other hand, the previous proof suggests
the following sentence:
• let u be the local solution of the L2 -critical problem (3.1) with max-
imal time interval, and let S = Lp ((−T∗ , T ∗ ); Lq ) be the Strichartz
norm we used to close the fixed point. Assume that
kukS < ∞; (3.16)
then T∗ , T ∗ = +∞, i.e. the solution is global.
It should be now clear which is the simple proof of the previous sentence.
In other words, if one can ensure that a suitable Strichartz norm stays finite
also when t approaches the maximal existence time, the solution can be
extended to all times.
In the following, we will refer to this kind of statements as global exis-
tence criteria.
SPANISH LECTURES ON NLS 16

Proving a priori estimates of kind (3.16) for solutions of nonlinear dis-


persive equations was one of the most investigated topics in the last years.
We will introduce later some of of the tools which have been introduced to
solve these kind of problems. See also the bibliographic notes to have an
idea about the most important contributions.
The local theory (and global, when it is immediate) in the spaces Ḣ s
can be performed by the same methods. We refer to the exercises for some
examples.

3.3. Exercises.
Exercise 3.7. Let t0 =
6 0, δ > 0 and consider the following nonlinear
equation:
i∂t u + |t − t0 |δ ∆u = ±|u|γ−1 u (3.17)
Conjecture the critical exponent γc,s,δ for the local well-posedness of the
Cauchy problem associated to (3.17) in Ḣ s , with s ≥ 0. Moreover, prove
the local theory in L2 for any subcritical γ and for the critical value γc,s,δ
in the case of small data.
Hint. For the L2 -theory, first transform (3.17) into a nonlinear Schrödinger
equation with time dependent nonlinearity, by a change o variables. Then
use the Lorentz version of Strichartz estimates (see Exercise 2.15).
Exercise 3.8. Check all the algebraic conditions on the exponents in The-
orems 3.4 and 3.5.
Exercise 3.9. Let u ∈ C((−T, T ); H 2 (Rn )) be a solution of
i∂t u + ∆u = ±|u|γ−1 u,
Define the energy E(t) as follows:
Z Z
1 2 1
E(t) = |∇u(t, x)| dx ± |u(t, x)|γ+1 dx. (3.18)
2 γ+1
Prove that E(t1 ) = E(t2 ), for any t1 , t2 ∈ (−T, T ).
Exercise 3.10. Consider the Cauchy problem (3.1) with datum u0 ∈ Ḣ 1 (Rn ).
4
Prove that, if 1 < γ < γc,1 = 1 + n−2 , then there exists a unique local solu-
1 n
tion u ∈ C((−T, T ); Ḣ (R )).
4
Moreover, prove the same when γ = 1+ n−2 and ku0 kḢ 1 < ²0 , with ²0 > 0
sufficiently small.
Exercise 3.11. In space dimension n = 3, consider the quintic Schrödinger
equation with datum u0 ∈ Ḣ 1 . Prove that the boundedness of the norm in
L10 (I; L10 ), where I is the maximal existence interval, is a criterion for the
global well-posedness in Ḣ 1 .
Exercise 3.12. In space dimension n = 3, consider the cubic Schrödinger
1
equation with datum u0 ∈ Ḣ 2 . Prove that the boundedness of the norm in
L5 (I; L5 ), where I being the maximal existence interval, is a criterion for
1
the global well-posedness in Ḣ 2 .
SPANISH LECTURES ON NLS 17

4. The focusing NLS with data in H 1


As it is clear from what we introduced in the previous section, the local
theory for the Cauchy problem (3.1) does not see the difference between the
cases of nonlinearities with positive or negative sign. On the other hand, in
the L2 -theory, we concluded the global existence in the subcritical case and
in the critical case with small data, since in both the cases ±|u|γ−1 u the
mass is conserved.
When we try to perform the same arguments for the H 1 -theory, the prob-
lem of global existence is quite more complicated, already at the level of
subcritical nonlinearities. The aim of this section is to introduce some clas-
sical arguments which permit to solve this problem at least in some range
of nonlinearities.

4.1. Energy conservation. Let γ > 1 and consider the Cauchy problem
(
i∂t u + ∆u + ²|u|γ−1 u = 0
(4.1)
u(0, x) = u0 (x),
¡ ¢
where ² = ±1, and u0 ∈ H 1 (Rn ). Let u ∈ C (−T, T ); H 1 (Rn ) be a local
solution of (4.1) and define, for any t ∈ [−T, T ], the energy E(t) as follows:
Z Z
1 2 ²
E(t) = |∇u(t, x)| dx − |u(t, x)|γ+1 dx.
2 γ+1
Notice that E(t) < ∞, for any t ∈ (−T, T ), since u ∈ H 1 . Moreover, the
computation made in exercise 3.9, which is justified for H 2 -solutions, can be
extended to H 1 -solutions by standard density arguments. Consequently, the
energy E(t) is conserved during the evolution also in this case, i.e. E(t) =
E(0), for all times t ∈ (−T, T ). Now observe that the sign of E(t) depends
on ². We give the following definition.
Definition 4.1. We say that equation (4.1) is defocusing if ² = −1; on
the other hand, if ² = 1, we say that (4.1) is focusing.
In the defocusing case, since E is positive we have the immediate bound
k∇u(t, ·)kL2 (Rn ) ≤ E(0) < ∞.
Moreover, the mass (i.e. the L2 -norm) is conserved, hence we can extend
the local solution in H 1 to be global in time (in the H 1 -subcritical case
4
γ < 1 + n−2 ).
On the other hand, in the focusing case the situation is quite different
and the global theory needs some additional arguments which we are going
to introduce in the following results.
In the following, we will concentrate on the fosusing NLS with data in
H 1 , for L2 -subcritical nonlinearities.
Theorem 4.2. Let u0 ∈ H 1 , and 1 < γ < 1 + n4 ; let u ∈ C((−T∗ , T ∗ ); H 1 )
be a local solution of (4.1), with ² = 1, and maximal time existence. Then
T∗ = T ∗ = ∞.
Moreover, assume γ = 1 + n4 ; then there exists ²0 > 0 such that, if
ku0 kL2 < ²0 , then T∗ = T ∗ = ∞.
SPANISH LECTURES ON NLS 18

Proof. First we remark that the local solution is constructed by a fixed point
argument (see the previous section and Exercise (3.10)). Now we want to
conclude that the solution is in fact global. In order to do this, since the L2 -
norm is conserved, it is sufficient to prove that the Ḣ 1 -norm does not blow
up when t approaches the maximal time. Consider the energy functional
Z Z
1 2 1
E(t) = |∇u(t, x)| dx − |u(t, x)|γ+1 dx,
2 γ+1
and recall the Gagliardo-Nirenberg inequality
2+(γ−1)· 2−n (γ−1) n
kf kγ+1
Lγ+1
≤ Cγ+1 kf kL2 2
k∇f kL2 2
, (4.2)
4
valid for 1 ≤ γ ≤ 1 + n−2 (notice that, for γ = 1 this is an identity, while for
4
γ = 1 + n this is the Sobolev embedding of Ḣ 1 in L(n+2)/(n−2) ). By (4.2)
we obtain
µ ¶
2 1 Cγ+1 2+(γ−1)· 2−n (γ−1) n −2
E(0) = E(t) ≥ k∇u(t)kL2 − ku(t)kL2 2
k∇u(t)kL2 2
.
2 γ+1
(4.3)
Let us first consider the case γ < 1 + n4 . Since ku(t)kL2 is constant in t, and
(γ − 1) n2 − 2 < 0, inequality (4.3) implies that
k∇u(t)kL2 ≤ CE(0), (4.4)
for some C > 0, which proves the claim.
In the case γ = 1 + n4 , by (4.3) we have
µ ¶
2 1 Cγ+1 4
E(0) ≥ k∇ukL2 − ku(0)kL2 .
n
2 γ+1
As a consequence, the choice
γ+1
²0 < (4.5)
2Cγ+1
gives (4.4), and the proof is complete. ¤
Remark 4.1. As it is shown in the previous proof, the best choice of ²0
is related to the best constant Cγ+1 in the Gagliardo-Nirenberg inequality
(4.2). As Weinstein proved in (CITARE WEINSTEIN BENE), the best
Cγ+1 is reached by the unique positive and radial solution with minimal
energy (ground state) of the elliptic equation
4
−∆Q + Q = |Q| n Q;
Weinstein gave an approximation of the L2 -norm of the ground state, which
determines a threshold for the initial data u0 ∈ H 1 in the focusing mass-
critical NLS, as we saw in the previous theorem.
We now state another famous result, due to Glassey, about a blow-up
criterion for the focusing NLS with data in H 1 . It is based on a convexity
method called virial identity. First we prove an important lemma on which
we will come back in the next sections.
SPANISH LECTURES ON NLS 19

Theorem 4.3 (Virial identity). Let u0 ∈ H 1 , and assume that |·|u0 (·) ∈ L2 .
Let u ∈ C((−T∗ , T ∗ ); H 1 (Rn )) be a local solution of (4.1) in the maximal
(−T∗ , T ∗ ), and define
Z
V (t) = |x|2 |u(t, x)|2 dx, (4.6)

for any t ∈ (−T∗ , T ∗ ). Then, the following identities hold:


Z
dV
= 4= u(∇u · x) dx (4.7)
dt
Z Z
d2 V 2 4n(γ − 1)
= 8 |∇u(t, x)| dx − ² |u|γ+1 dx = 16E(t). (4.8)
dt2 γ+1
Notice that, when γ = 1 + n4 , the previous identity reads
Z Z
d2 V 2 16²
= 8 |∇u(t, x)| dx − |u|γ+1 dx = 16E(t).
dt2 γ+1
We leave the proof of the previous Lemma to the reader (see the exercises).
Now we pass to show a consequence of the virial identitiy (4.8).
Theorem 4.4 (Zakharov-Glassey). Let u0 ∈ H 1 , with | · |u0 (·) ∈ L2 , and
assume γ = 1 + n4 . There exists k0 > 0 such that, for any k > k0 , the unique
solution u ∈ C((−T∗ , T ∗ ); H 1 ) of the focusing (4.1) with initial datum ku0
blows up in finite time.
Proof. By identity (4.8), we have
Z
V (t) = 8t2 E(t) + 4t= u0 (∇u0 · x) dx + V (0),

for all t ∈ (−T∗ , T ∗ ). Since the energy is conserved and V (0) is assumed to
be finite, this shows that V (t) is a parabola. Now, notice that, if k is large
enough, then
Z Z
k2 k γ+1
E(t) := E(t; ku0 ) = |∇u0 (x)| dx − |u0 (x)|γ+1 dx < 0.
2 γ+1
Consequently, V (t) becomes negative after a finite time T + , which proves
that the solution stops to exist in a finite time. ¤
4.2. Exercises.
Exercise 4.5. Consider equation (3.17) and define the energy as follows:
Z Z
a(t) 1
E(t) = |∇u(t, x)|2 dx ± |u(t, x)|γ+1 dx,
2 γ+1
where a(t) = |t − t0 |δ . Prove that, if u is a sufficiently regular solution of
(3.17), then the following identity holds:
d 1
E(t) = a0 (t)k∇u(t, ·)k2L2 .
dt 2
Exercise 4.6. Use the result of the previous exercise to state when a local
H 1 -solution of (3.17) is in fact global, distinguishing between defocusing
and focusing cases.
Exercise 4.7. Prove the Gagliardo-Nirenberg inequality (4.2).
SPANISH LECTURES ON NLS 20

Exercise 4.8. Prove Lemma 4.3. (Hint. For regular solution, just compute
the derivatives and integrate by parts. Then argue by density.)
Exercise 4.9. In dimension n ≥ 3, prove that, for any f ∈ Ḣ 1 (Rn ) with
| · |f (·) ∈ L2 (Rn ), the following inequality holds:
kf kL2 ≤ 4k |x|f kL2 k∇f kL2
Exercise 4.10. Let u be an H 1 -solution of the focusing NLS with negative
initial energy, in space dimension ≥ 3; we saw that there exists a time T such
that V (t) = k|·|u(t)kL2 → 0, when t → T . Prove that limt→T k∇u(t, ·)kL2 =
∞. (Hint. Use the previous exercise.)

5. Morawetz identities and local smoothing


We now pass to investigate the family of so called Moraewtz identities,
which the virial Theorem 4.3 is an example of. They are algebraic identities
holding on solution of Schrödinger equations with relevant consequences on
the evolution of their solutions. Let us go back to the linear Schrödinger
equation (
i∂t u + ∆u = 0 in R1+n
(5.1)
u(0, x) = u0 (x);
now consider a solution of (5.1), a linear operator K on L2 and define the
quantity Θ(t) = (u, Ku), denoting with the brackets the inner product in
L2 . If we derive formally with respect to time we obtain
d
Θ(t) = i(u, [∆, K]u) = i(u, (∆K − K∆)u), (5.2)
dt
which is in fact the Eisenberg equation. In an analogous way we also obtain
d2
Θ(t) = −(u, [∆, [∆, K]]u). (5.3)
dt2
Now we specialize the previous identities in the case in which K is a multi-
plication operator; in this case the commutators are explicit; we can easily
prove the following Lemma.

RLemma 5.1. Let φ : Rn → R be a positive function and define Θ(t) =


2
φ|u| dx, being u a solution of (5.1). The following formal identities hold:
Z
d
Θ(t) = 2= u∇φ · ∇u dx (5.4)
dt
Z Z
d2
Θ(t) = 4 ∇uD φ∇u dx − |u|2 ∆2 φ dx,
2
(5.5)
dt2
where D2 φ is the Hessian matrix of φ and ∆2 φ = ∆(∆φ) is the bi-laplace.
Proof. The proof is based on an explicit (formal) computation, in the fol-
lowing assume we are working with smooth solutions. First notice that
[∆, φ] = (∆φ) + 2∇φ · ∇, (5.6)
hence (5.2) gives
Z Z
d 2
−i Θ(t) = |u| ∆φ dx + 2 u∇φ · ∇u dx (5.7)
dt
SPANISH LECTURES ON NLS 21

Integrating by parts gives


µZ Z ¶
2
< |u| ∆φ dx + 2 u∇φ · ∇u dx = 0,

hence (5.4) follows by (5.7). In order to prove (5.5) consider identity (5.3);
since ∆ is a symmetric operator and [∆, φ] is antisymmetric, with respect
to the L2 -product, we notice that
(u, [∆, [∆, φ]]u) = (u, ∆[∆, φ]u) − (u, [∆, φ]∆u) (5.8)
= (∆u, [∆, φ]u) + ([∆, φ]u, ∆u)
= 2<(∆u, [∆, φ]u).
Again, now (5.5) follows by (5.3) and integration by parts in (5.8). ¤
Remark 5.1. The formal computation in the previous proof is justified for
Ḣ 2
R -solutions. Indeed, the term which requires more regularity is of the form
∇φ·∇u∆u, and φ is an abstract function, at this level. On the other hand,
the result of the previous Theorem holds also on H 1 -solutions, and the proof
can be performed by standard density arguments.
For the NLS, we have an analogous result.
Lemma 5.2 (Virial-type identities
R for NLS). Let φ : Rn → R be a positive
2
function and define Θ(t) = φ|u| dx, being u a solution of
iut + ∆u = ±|u|γ−1 u. (5.9)
The following formal identities hold:
Z
d
Θ(t) = 2= u∇φ · ∇u dx (5.10)
dt
Z Z Z
d2 2 2 2 2(γ − 1)
Θ(t) = 4 ∇uD φ∇u dx − |u| ∆ φ dx ± |u|γ+1 ∆φ dx,
dt2 γ+1
(5.11)
where D2 φ is the Hessian matrix of φ and ∆2 φ = ∆(∆φ) is the bi-laplace.
We leave the proof of the previous Lemma as an exercise for the reader.
Notice that the result generalize the virial identity, Theorem 4.3.
Remark 5.2. Notice that, for stationary solutions u = u(x) of (5.9), identity
(5.11) reads as
Z Z
2 n(γ − 1)
|∇u| dx ± |u|γ+1 dx = 0,
2(γ + 1)
in the case φ = |x|2 : this is the so called Pohozaev identity.
We now introduce two important radial multipliers:
• the virial multiplier φ(x) = |x|2
• the Morawetz multiplier φ(x) = |x|
For the virial we have
D2 |x|2 = 2Id, ∆2 |x|2 = 0,
SPANISH LECTURES ON NLS 22

then we conclude by (5.5) that


Z Z
d2
|x| |u| dx = 8 |∇u|2 dx = 16E,
2 2
(5.12)
dt2
being E the energy, which is a constant of the motion.
Remark 5.3. In order to prove Theorem 4.3, in the nonlinear case, just
denote by f (u) = |u|γ−1 u, write equation (4.1) as
ut = i∆u ∓ if (u),
and compute
dV
= i(u, [∆, |x|2 ]u) ∓ i(f (u), φu) ± i(u, φf (u)) = i(u, [∆, |x|2 ]u).
dt
This proves (4.7), as we saw above. The proof of (4.8) is analog.
By using the Morawetz multiplier in Lemma 5.1, we can prove the follow-
ing a priori estimate.
Theorem 5.3 (Morawetz estimate). Let n ≥ 3, and u be a solution of (5.1),
1
with u0 ∈ Ḣ 2 . The following a priori estimate hold:
Z Z
|∇τ u(t, x)|2
dx dt . ku0 k2 1 . (5.13)
|x| Ḣ 2

Proof. By (5.4), (5.5), we have


Z Z Z
d
= u∇φ · ∇u = 4 ∇uD2 φ∇u dx − |u|2 ∆2 φ dx.
dt
We now choose φ = |x|; notice that
(
−8πδ0 , n=3
∆2 |x| =
−(n − 1)(n − 3)|x|−3 , n ≥ 4,

hence ∆2 |x| < 0. Moreover, we use the following formula, which is true for
radial functions,
φ0
∇uD2 φ∇u = φ00 |∂r u|2 + |∂τ u|2 , (5.14)
r
where ∂r , ∂τ denote, respectively, the radial and tangential components of
the gradient. Then we obtained
Z Z
|∇τ u(t, x)|2 d
0≤ dx . = u∂r u dx,
|x| dt
for all times t ∈ R. Integrating in time yields
Z T Z ¯Z ¯ ¯Z ¯
|∇τ u(t, x)|2 ¯ ¯ ¯ ¯
dx dt . ¯ u(−T )∂r u(−T ) dx¯ + ¯ u(T )∂r u(T ) dx¯¯
¯ ¯ ¯
−T |x|
. ku(−T )k 12 + ku(T )k 12 ,
Ḣ Ḣ
1
for any T ∈ R, and the proof follows now by the conservation of the Ḣ 2 -
norm. ¤
SPANISH LECTURES ON NLS 23

Notice that estimate (5.13) is a global (in space and time) tangential
smoothing for solutions of the free Schrödinger equation. This kind of result
is in general false for the complete gradient. A famous local result in this
direction is the following.
Theorem 5.4 (Local smoothing). Let n ≥ 1, and u be a solution of (5.1),
1
with u0 ∈ Ḣ 2 . The following a priori estimate hold:
Z Z
1
sup |∇(t, x)|2 dx dt . ku0 k2 1 . (5.15)
R>0 R |x|≤R Ḣ 2

Proof. We give a proof of this result which just works in dimensions n ≥ 3.


For a general and standard proof, based on Fourier techniques, in all the
dimensions, see the references at the end of these notes.
We work with a new multiplier which Rmixes the virial and Morawetz ones.
r
For any R > 0, let us consider φR (r) = 0 φ0R (s) ds, with
(
(n−1)r
0 , r≤1
φR (r) = 1 2nR R
2 − 2nr n−1 , r > 1.
An explicit computation shows that
(
1 n−1
00 · , r≤1
φR (r) = R1 R2n n (n−1)
R · 2nrn , r > 1,
(
2 −4πδx=0 − R12 δ|x|=R , n=3
∆ φR = n−1 (n−1)(n−3)
− 2 δ|x|=R − 2r3
χ[1,+∞) , n ≥ 4.
Observe that ∆2 φ < 0 and φ0 , φ00 ≥ 0; the proof now follows exactly as in
the previous theorem, once we observe that
¯ Z ¯
¯ ¯
¯= u∇φR · ∇u dx¯ . kuk 1 . (5.16)
¯ ¯ Ḣ 2

¤
Remark 5.4. Notice that the local smoothing multiplier used in the proof
of Theorem 5.4 permits to obtain further informations on the solutions.
Indeed, one can easily check that the following trace estimate was proved:
Z Z
1
sup 2 |u|2 dσ(x) dt . ku0 k2 1 ,
R>0 R |x|=R Ḣ 2

where dσ is the surface measure on the sphere SR = {|x| = R}.


Remark 5.5. The multiplier technique we used to prove Theorem 5.4 is
adaptable to a more general class of Schrödinger equations; for example, we
can treat the defocusing NLS, and the Schrödinger equation with an external
potential (see the exercises at the end of the Section).
Remark 5.6. We also point out that the natural initial space which appears
1
in the integration by parts is Ḣ 2 . In the case of the linear Schrödinger
equation, estimate (5.15) immediately implies that
Z Z
1 1
sup ||D|s+ 2 u|2 dx dt . ku0 k2Ḣ s ,
R>0 R |x|≤R
SPANISH LECTURES ON NLS 24

for any s ≥ 0, since |D|s commutes with eit∆ .


Finally notice that the previous estimate implies the following weaker one:
1
kω(x)|D|s+ 2 eit∆ u0 kL2t L2x . ku0 kḢ s , (5.17)
for any s ≥ 0, where
1
ω(x) = 1 1 ,
−²
|x| 2 + |x| 2 +²
for any small ² > 0.
5.1. Exercises.
Exercise 5.5. Prove formula (5.14) for radial functions φ.
Exercise 5.6. Prove that there does not exists a positive function φ : R →
R, φ ≥ 0, with φ0 ∈ L∞ , such that φ0 > 0, φ00 < 0 on the whole real line.
Exercise 5.7. Prove Lemma 5.2.
Exercise 5.8. Prove (5.16).
Exercise 5.9. Let n ≥ 3, V ∈ C ∞ (Rn ; R) and u be a regular solution of
(
iut + ∆u + V (x)u = 0
1 (5.18)
u(0, x) = u0 ∈ Ḣ 2 .
Assume that ∂r V ≥ 0, for any x ∈ Rn ; prove that the Morawetz and local
smoothing estimates (5.13), (5.15) hold.
Exercise 5.10. Assume that V ∈ S, is a rapidly decaying function, with
∂r V ≥ 0. Use the result of the previous exercise to prove that, for any
non endpoint 0-admissible couple (p, q), any solution of (5.18) satisfies the
Strichartz estimate
1
k|D| 2 ukLpt Lqx . ku0 k 12 .

Hint. Write the solution of (5.18) via Duhamel, treating V u as a pertur-
1
bation term. Then apply a T T ∗ -argument, at the level of Ḣ 2 , by mixing
free Strichartz estimates with dual of the local smoothing which has been
proved in the previous exercise.

6. Interaction Morawetz identities


The final section of these notes is devoted the so called Interaction Morawetz
identities (or bilinear Morawetz identities) which have been introduced by
Colliander-Keel-Staffilani-Takaoka and Tao in [7] to solve the quintic de-
focusing Schrödinger equation in 3D. They are the bilinear version of the
Morawetz identities we introduced in the previous section.
Let first u be a solution of the linear Schrödinger equation i∂t u + ∆u = 0,
and define Z Z
H(t) = |u(x)|2 |u(y)|2 φ(|x − y|) dx dy, (6.1)
for some positive radial function φ = φ(|x|) : Rn → R. Now define w(x, y) =
u(x)u(y), and it is immediate to see that w is a solution of
i∂t w + ∆xy w = 0 (x, y) ∈ Rn × Rn .
SPANISH LECTURES ON NLS 25

Hence it is clear that we can reduce to the computations made in the previous
section, to prove the following result.
Theorem 6.1. Let u be a solution of the NLS (5.9), φ = φ(|x|) : Rn →
[0, ∞) be a positive radial multiplier and consider the quantity H(t) in (6.1).
Then, the following formal identities hold:
Z Z
Ḣ(t) = 4= j(x) · ∇φ(x − y)m(y) dx dy (6.2)
Z Z
Ḧ(t) = 4 G(x, y)D2 φ(x − y)G(y, x) dx dy (6.3)
Z Z
− m(x)m(y)∆2 φ(x − y) dx dy
Z Z
4(γ − 1)
± |u(x)|γ+1 |u(y)|2 ∆φ(x − y) dx dy
γ+1
Z Z
=4 G(x, y)D2 φ(x − y)G(y, x) dx dy (6.4)
Z Z
+ ∇(m(x)) · ∇(m(y))∆φ(x − y) dx dy
Z Z
2(γ − 1)
± |u(x)|γ+1 |u(y)|2 ∆φ(x − y) dx dy,
γ+1
where m(x) = |u(x)|2 , j(x) = =u(x)∇u(x), and
G(x, y) = u(y)∇u(x) − u(x)∇u(y).
We leave the proof as a relevant exercise for the reader.
Remark 6.1. It is also possible to define an interactive quantity H(t) con-
taining two different solutions of (5.9), see the bibliographical notes in the
following.
Now we concentrate on identity (6.3). Notice that, when we use the virial
multiplier φ = |x|2 we obtain by (6.3) that
Z Z Z Z
2 2 4n(γ − 1)
Ḧ(t) = 16 |u(x)| |∇u(x)| dx ± |u(x)| dx |u(x)|γ+1 dx.
2
γ+1
Now we state the analogous result which can obtained by using the Morawetz
multiplier φ = |x|.
Theorem 6.2 (Colliander-Keel-Staffilani-Takaoka -Tao [7]). In space di-
mension n = 3, consider the defocusing Schrödinger equation (5.9) with
1 < γ ≤ 1 + 4/(n − 2) and initial datum u0 ∈ H 1 (R3 ); let u be a local H 1 -
solution in a interval I. Prove that the following a priori estimate holds:
Z Z
3 1
|u(t, x)|4 dx dt . M (0) 2 E(0) 2 , (6.5)
I

where M (0) = ku0 k2L2 , E(0) = 21 k∇u0 k22 + 16 ku0 k6L6 .


We leave this proof to the reader (see the Hint in the exercises below).
Another interesting result can be obtained in 1D using Theorem 6.1.
SPANISH LECTURES ON NLS 26

Theorem 6.3 (Planchon-Vega [20]). In space dimension n = 1, consider


the defocusing Schrödinger equation (5.9) with 1 < γ ≤ 9 and initial datum
1
u0 ∈ H 1 (R); let u be a global Ḣ 4 -solution in a interval I. Prove that the
following a priori estimate holds:
Z Z Z Z
2 2
|∂x (|u(x)| )| dx dt + |u(x)|γ+3 dx dt . ku0 k2 1 . (6.6)
Ḣ 4
4 1
Proof. First notice that 9 = 1 + 1−2· 14
is the 4 -critical nonlinearity for (5.9).
Notice that the right-hand side of (6.4), with the choice φ = |x|, can be now
controlled from below with the left-hand side of (6.6). In order to conclude
the proof of (6.6), it is now sufficient to prove that
¯ ¯
¯ ¯
¯Ḣ(t)¯ . ku0 k2 1 . (6.7)
Ḣ 4
Indeed, we can prove, using the convexity of H(t), that
Z Z
1
lim Ḣ(t) = ± u0 (ξ)|2 |b
|b u0 (η)|2 |ξ − η| dξ dη, (6.8)
t→±∞ 2
which implies (6.7). We remand to [20] for the proof of (6.8). ¤
Remark 6.2. Notice that the part of (6.6) regarding the term ∂x |u|2 is a
global smoothing for the NLS; this is the only result we know in this direc-
tion.
Remark 6.3. The techniques used to prove Theorem 6.3 can be extended to
the 2D case by means of Radon Transforms. See [20] for the details.
6.1. Applications: a scattering result. We devote this final section to
an application of Theorem 6.3 to the following defocusing NLS in 1D:
(
i∂t u + ∂x2 u = |u|8 u
(6.9)
u(0) = u0 ∈ H 1 (R).
As we remarked before, (6.9) is 14 -critical in 1D. Here we assume u0 ∈ H 1 ,
by the arguments of the previous sections we know that (6.9) is globally well
posed in H 1 . We prove the following result.
Theorem 6.4. Let u ∈ C(R; H 1 ) the unique global solution in H 1 of (6.9).
There exist two functions u± ∈ H 1 (R) such that
2
lim ku(t) − eit∂x u± kH 1 = 0. (6.10)
t→±∞

Remark 6.4. The property (6.10) can be rephrased, using the terminology
of Scattering Theory, in terms of asymptotic completeness of the wave
operators associated to free hamiltonian H0 = −∂x2 and the perturbed one
via the nonlinear term in (6.9).
Proof of Theorem 6.4. Let us write the global solution we already constructed
as Z t
2 2
u(t) = eit∂x u0 + i ei(t−s)∂x |u|8 u(s) ds (6.11)
0
Notice that a weaker version of estimate (6.6) is
Z Z Z Z
2 2
|∂x (|u(x)| )| dx dt + |u(x)|12 dx dt . M (0)3 E(0), (6.12)
SPANISH LECTURES ON NLS 27

where M and E are as usual the mass and the energy, which are two invari-
ants of the motion. Moreover, by the Gagliardo-Nirenberg inequality
3 1
kf kL∞ . kf kL4 6 k∂x f kL4 2 , (6.13)

and (6.12), we obtain (for f = |u|2 )


3 ° °1
kuk2L8 L∞ = k|u|2 kL4t L∞ . kuk 2
12
°∂x |u|2 ° 4 2 < ∞.
t x x L L tx tx

This implies that


° Z ∞ ° Z
° 2 ° 1

°∂x e−is∂ x |u| u(s) ds°
8
. E(0) 2 ku(s)k8L∞ ds → 0, (6.14)
° ° x
t L2 t

as t tends to infinity. Analogously, by Hölder we have


8 1
kukL9t L18
x
≤ kukL9 8 L∞ kukL9 ∞ L2 < ∞,
t x t x

which implies that


°Z ∞ ° Z
° −is∂x2
° ∞
° e |u| u(s) ds°
8
. ku(s)k9L18 ds → 0, (6.15)
° ° x
t L2 t

as t tends to infinity. When time goes to −∞, the conclusions are analogous,
by the same computations. Now we define
Z ±∞
2
u± = u0 + i e−is∂x |u|8 u(s) ds, (6.16)
0

and we obtain by (6.11), (6.14), and (6.15) that


° °
° 2 °
°u − eit∂x u± ° 1 → 0,
H

as t → ±∞, which completes the proof. ¤

6.2. Exercises.
Exercise 6.5. Prove Theorem 6.1. Hint. For (6.2), (6.3) just use the
Morawetz identities of the previous section; for (6.4) integrate by parts one
term in (6.3).
Exercise 6.6. Prove Theorem 6.2. Hint. Use the multiplier φ = |x| in
Theorem 6.1.
Exercise 6.7. Prove the 1D Gagliardo-Nirenberg inequality (6.13).
Exercise 6.8. Consider the operator
Z Z
Tu = u(x)∇u(x) · ∇φ(x − y)|u(y)|2 dx dy,
R3 R3

with φ(|x|) = |x|α , x, y ∈ R3 . Find the unique α ∈ R such that the following
estimate holds:
|T u| . k∇uk2L2 .
SPANISH LECTURES ON NLS 28

6.3. Some obviously incomplete bibliographical notes. Section 2.


For Strichartz estimates, see the standard references [13], [15], and the ref-
erences therein. The Christ-Kiselev Lemma was proved in [6]. For a detailed
discussion of the topic see [4] and the references therein. For exercise 2.15
see [10].
Section 3. Most of the subcritical theory was performed by Ginibre and
Velo in [13]. For the quintic 3D-NLS with data in H 1 , see [5]; the same
problem with data in Ḣ 1 was first treated in [3], then completely solved in
the defocusing case in [7]. The same problem in the focusing case was solved
for radial solutions by Kenig and Merle in [16]. See the previous papers and
the references therein.
Sections 4,5,6. For a detailed references list see [4]. Morawetz identities
appeared in [17]. The standard references for the local smoothing are [8],
[21], and [22]. The local smoothing multiplier appears intrinsecally in [1]
and explicitly in [12]. Finally, for interaction Morawetz identities see [7],
[20] and the references therein.

References
[1] J.A. Barceló, A. Ruiz, and L. Vega, Some dispersive estimates for
Schrödinger equations with repulsive potentials J. Funct. Anal. 236 (2006), 1–24.
[2] Bergh, J., and Löfström, J., Interpolation spaces. Springer Verlag, Berlin,
1976.
[3] Bourgain, J., Global wellposedness of defocusing critical nonlinear Schrödinger
equation in the radial case, J. Amer. Math. Soc. 12, no. 1, (1999), 145–171.
[4] Cazenave T., Semilinear Schrödinger equations, Courant Lecture Notes in
Mathematics, vol. 10, American Mathematical Society, Providence, 2003.
[5] Cazenave, T.m and Weissler, F. B., The Cauchy problem for the nonlinear
Schrödinger equation in H 1 , Manuscripta Math. 61 (1988), no. 4, 477–494.
[6] M. Chirst, and A. Kiselev, Maximal functions associated to filtrations, J.
Funct. Anal. 179(2) (2001), 409–425.
[7] J. Colliander, M. Keel, G. Staffilani, H. Takaoka, and T. Tao, Global
well-posedness and scattering for the energy-critical nonlinear Schrd̈inger equa-
tion in R3 , Annals of Math. 167 No. 3 (2008), 767–865.
[8] P. Constantin, and J.-C. Saut, Local smoothing properties of dispersive equa-
tions, Journ. AMS (1988), 413–439.
[9] Cwikel, M., On (LP0 (A0 ), LP1 (A1 ))θ,q , Proc. Amer. Math. Soc. 44 (1974), 286-
292.
[10] Fanelli, L., Semilinear Schrödinger equation with time dependent coefficients,
Math. Nach. 282 (2009), 976–994.
[11] Fanelli L., Montefusco E., On the blow-up threshold for weakly coupled
nonlinear Schrödinger equations, J. Phys. A 40 (2007), 14139–14150.
[12] Fanelli, L., and Vega, L., Magnetic virial identities, weak dispersion and
Strichartz inequalities, Math. Ann. 344 (2009), 249–278.
[13] Ginibre, J., and Velo, G., The global Cauchy problem for the nonlinear
Schrödinger equation revisited, Ann. I.H.P., Analyse Non Lynéare 2 (1985),
309–327.
[14] Glassey, R.T., On the blowing up of solutions to the Cauchy problem for
nonlinear Schrödinger equations, J. Math. Physics 18 (1977), 1794–1797.
[15] Keel, M., and Tao, T., Endpoint Strichartz estimates, Amer. J. Math. 120
(1998) no. 5, 955–980.
[16] Kenig, C., and Merle, F., Global well-posedness, scattering and blow-up for
the energy-critical, focusing, non-linear Schrödinger equation in the radial case,
Invent. Math. 166 (2006), 645-675.
SPANISH LECTURES ON NLS 29

[17] Morawetz, C., Time decay for the nonlinear Klein-Gordon equations, Proc.
Roy. Soc. Ser. A 306 (1968), 291–296.
[18] O’Neil, R., Convolution operators and L(p, q) spaces, Duke Math. J. 30 (1963),
129–142.
[19] Ozawa T., Remarks on proofs of conservation laws for nonlinear Schrodinger
equations. Calc. Var. PDE 25 (2006) 403408.
[20] Planchon, F., and Vega, L. Bilinear virial identities and Applications, Ann.
Scient. Ec. Norm. Sup. 42 (2009), 263–292.
[21] P. Sjölin, Regularity of solutions to the Schrödinger equations, Duke Math. J.
55 (1987), 699–715.
[22] L. Vega, The Schrödinger equation: pointwise convergence to the initial date,
Proc. AMS 102 (1988), 874–878.

Luca Fanelli: Universidad del Pais Vasco, Departamento de Matemáticas,


Apartado 644, 48080, Bilbao, Spain
E-mail address: luca.fanelli@ehu.es

You might also like