Skip to main content

Full text of "Strichartz estimates and the nonlinear Schrödinger equation on manifolds with boundary"

See other formats




Abstract. We establish Strichartz estimates for the Schrodinger equation on 
Riemannian manifolds (r2,g) with boundary, for both the compact case and 
the case that Q is the exterior of a smooth, non-trapping obstacle in Euclidean 
space. The estimates for exterior domains are scale invariant; the range of 
Lebesgue exponents (p, q) for which we obtain these estimates is smaller than 
the range known for Euclidean space, but includes the key LfL"^ estimate, 
which we use to give a simple proof of well-posedness results for the energy 
critical Schrodinger equation in 3 dimensions. Our estimates on compact man- 
ifolds involve a loss of derivatives with respect to the scale invariant index. We 
use these to establish well-posedness for finite energy data of certain semilinear 
Schrodinger equations on general compact manifolds with boundary. 

1. Introduction 

Let (r2,g) be a Riemannian manifold with boundary, of dimension n > 2, and let 
v{t,x) : [0,T] X O ^ C be the solution to the Schrodinger equation 

{tdt + As)v{t,x)^0, v{0,x)^f{x). (1) 

We assume in addition that v satisfies either Dirichlet or Neumann boundary con- 

v{t,x)\g^-^ = or d^v{t,x)\g^-^ = , 

where di, denotes the normal derivative along the boundary. In this work, we 
consider local in time Strichartz estimates for such solutions; these are a family of 
space-time integrability estimates of the form 

IK'IUp([0,T];L<!(O)) < C\\f\\H'(Q) ■ (2) 

Here H''{n) denotes the Sobolev space of order s, defined with respect to the 
spectral resolution of either the Dirichlet or Neumann Laplacian. The Lebesgue 
exponents will always be taken to satisfy p,q > 2, and always the Sobolev index 
satisfies s > 0. 

The consideration of high frequency bump function solutions to ([T]) shows that 
p, q, s must satisfy 

2 n n 

The authors were supported by National Science Foundation grants DMS-0801211, DMS- 
0654415, and DMS-0555162. 



In the case where equahty holds in ([3]) the estimate is said to be scale invariant; 
otherwise, there is said to be a loss of derivatives in the estimate, as it deviates 
from the optimal regularity predicted by scale invariance. 

Strichartz estimates are most well understood over Euclidean space, where f2 = 
M" and gij = Sij. In this case, the scale invariant estimates hold with s — and 
T = oo. See for example Strichartz [21, Ginibre and Velo 8 , Keel and Tao [15j . 
and references therein. Scale invariant estimates for s > then follow by Sobolev 
embedding; such estimates will be called subcritical, as their proof does not use the 
full rate of dispersion for the equation ([T]). 

This paper is primarily concerned with proving scale invariant Strichartz esti- 
mates on the domain exterior to a non-trapping obstacle in R", that is, il = R" \/C 
for some compact set JC with smooth boundary. Non-trapping means that every 
unit speed broken bicharacteristic escapes each compact subset of in finite time. 
While we are only able to prove such estimates for a restricted range of subcritical 
p, q, we do obtain estimates with applications to wellposedness in the energy space 
for semilinear Schrodinger equations when n = 3. 

The key new step in this paper is to establish (for the same range of p, q) scale 
invariant estimates for the semi-classical Schrodinger equation on a general compact 
Riemannian manifold with boundary. The step from these local estimates to the 
case of exterior domains depends on the local smoothing bounds of Burq, Gerard, 
and Tzvetkov When JC is assumed to be non-trapping, they proved that 

ii^-ii.^ao,Ti;.-i(o)) ^ecr(n), [0,1]. (4) 

This inequality is a natural formulation of the local smoothing estimates for Eu- 
clidean space which originated in the work of Constantin and Saut Sjolin jl8j . 
and Vega [23] . The estimate ^ was used in [4] to obtain Strichartz estimates with 
a loss of 1/p derivatives, by combining the gain in regularity in Q with Sobolev 
embedding, in order to prove space-time integrability estimates near the obstacle. 

Improved results were obtained by Anton [1], which show that Strichartz esti- 
mates hold with a loss of ^ derivatives. The approach in [Tj combines the local 
smoothing estimates (|4]) with a semi-classical parametrix construction, rather than 
Sobolev embedding. We remark that further improvement is possible by using the 
parametrix construction of the authors in [2, to yield a loss of ^ derivatives. This is 
currently the best known estimate for critical p, q, that is, | + ^ = f , except for the 
case where K. is strictly convex. For the exterior domain to a strictly convex /C, the 
full range of Strichartz estimates (except for endpoints) was obtained by Ivanovici 
for Dirichlet conditions, using the Melrose- Taylor parametrix construction. 

The use of local smoothing to establish Strichartz bounds has origins in the 
work of Journe, Soffer, and Sogge [14], and of Staffilani and Tataru [19]. Both deal 
with perturbations of the flat Laplacian in R", and establish estimates with no 
loss of derivatives. The paper [14] considered the case of potential terms —A + V, 
whereas [19] considers non-trapping metric perturbations. In both cases, one has 
local smoothing estimates similar to (jl]). 

More recently, Planchon and Vega [17] used a bilinear virial identity to obtain 
the scale invariant estimate ([2]) where p = q = 4, s = ^inn = 3 dimensions (along 
with a range of related inequalities), for the Dirichlet problem on non-trapping ex- 
terior domains. These estimates were applied to semilinear Schrodinger equations. 



showing that for defocusing, energy subcritical nonhnearities, one has global ex- 
istence for initial data in H^{fl). For strictly convex K,, the work [12] establishes 
global existence for the energy critical semilinear equation, focusing or defocusing, 
for small Dirichlet data in H^{^1). 

In the present work, we establish the Strichartz estimates ([2]) for a range of 
subcritical p, q. The key tool is a microlocal parametrix construction previously 
used for the wave equation in [20] and 0] . This approach treats both Dirichlet and 
Neumann boundary conditions, and applies to general non-trapping obstacles. 

Theorem 1.1. Let Q ~ R" \K. be the exterior domain to a compact non-trapping 
obstacle with smooth boundary, and A the standard Laplace operator on f2, subject 
to either Dirichlet or Neumann conditions. Suppose that p > 2 and q < oo satisfy 

+ 2 < 1 „^2, 

Then for the solution v — exp(itA)f to the Schrddinger equation ([!]), the following 
estimates hold 

II"IIlp([0,T];L'!(O)) < C\\f\\H=(n) : (6) 

provided that 

2 n n 

- + - = - - s . 7 
p q 2 

For Dirichlet boundary conditions, the estimates hold with T = oo. 

That one may take T = oo in ^ for Dirichlet boundary conditions is a conse- 
quence of the fact that ((U) holds for T = oo in the Dirichlet case. 

We now consider estimates for compact Rieniannian manifolds fl, with Ag the 
Laplace-Beltrami operator for g. Burq, Gerard, and Tzvetkov showed in 5 that 
for p > 2 estimates hold with a loss of ^ derivatives in case dfl = 0. The same 
result was established for compact manifolds with geodesically concave boundary 
in [12]. For general boundaries, we establish estimates with the same loss of - 
derivatives, valid for {p,q) satisfying ((SJ). For such {p,q), this is an improvement 
over the estimates of 2 , which involve a loss of 4- derivatives. 

Theorem 1.2. Let Q be a compact Riemannian manifold with boundary. Suppose 
that p > 2 and g < oo satisfy ([5]). Then for the solution v = exp(ziAg)/ to the 
Schrddinger equation ([!} , the following estimates hold for fixed finite T 

MLpao,T];Lim < 

forp, q, s satisfying ([T]). 

As with [5] and [lU Corollary 1.5], the loss of i arises as a consequence of using 
a representation for solutions that is valid only in a local coordinate chart; that is, 
on a semi-classical time scale. 

In the last two sections of this paper we present applications of the above the- 
orems to well-posedness of semilinear Schrddinger equations in three space dimen- 
sions with finite energy data. In Section [S] we use Theorem 11.11 and interpolation 
to establish the LfL^ Strichartz estimate. This estimate yields a simple proof 
of well-posedness for small energy data to the energy critical equation on exterior 
domains, a result first established by Ivanovici and Planchon [13]. In Section (6] we 
establish a variant in three dimensions of Theorem 11.21 for the case p = 2, for data 



u localized to dyadic frequency scale A. The estimate involves a loss of (logA)^ 
relative to the estimates of [5]. Following the Yudovitch argument as in O Sec- 
tion 3.3], we use this to establish well-posedness for finite energy data to certain 
semilinear Schrodinger equations, on general three dimensional compact manifolds 
with boundary. The logarithmic loss in the estimates restricts our result to slower 
growth nonlinearities than considered in 5 for manifolds without boundary. For 
particular three dimensional manifolds without boundary, recent results have been 
obtained for the energy critical case by Herr 9 , Herr, Tataru, and Tzvetkov [TU] . 
and lonescu and Pausader [TT] . 

The outline of this paper is as follows. In section 2, we reduce Theorems 1 1 . HfTT2l 
to estimates on the unit scale within a single coordinate chart. Section 3 outlines the 
angular localization approach from |20j . and introduces a wave packet parametrix 
construction. Estimates on the parametrix are then developed in section 4. We 
conclude in sections 5 and 6 with the applications to semilinear Schrodinger equa- 

Throughout this paper we use the following notation. The expression X < Y 
means that X < CY for some C depending only on the manifold, metric, and 
possibly the triple {p,q,s) under consideration. Also, we abbreviate LP{I; L'^{U)) 
by LPL'J{I X U) or by L'^Li{U) when / = [0,r]. li U ^ M" we write L^L?. As 
will be seen below, the last component of an n-vector will take on special meaning, 
hence we will often write x — {x' ,Xn) so that x' denotes the first n~l components. 

We conclude this introduction with a remark on the Sobolev spaces that we use 
in the case of exterior domains. In the above theorems, the Sobolev space H''{il) 
and the operator exp(itA) are defined using the spectral resolution of A subject to 
the chosen Dirichlet or Neumann boundary condition B; in particular, the linear 
evolution preserves H^{n). The space H^{il) is then equal to the subspace of 
H^{fl) satisfying Bu — 0, and for < s < 2, the space H''{il.) can be defined by 
interpolation. For s > 2, these spaces satisfy u e H^{il) if and only if Bu = 
and Au e i/*^^(ri). Thus, for large values of s, a function in H'^{n) satisfies the 
linear compatibility conditions B{A''u) = 0, for k for which this is defined. These 
compatibility conditions are necessary to bootstrap the local smoothing estimates 
to higher orders s, as well as to insure v{t, ■ ) e which is required to 

handle commutator terms with cutoff functions. We will also use that 

Ikliff-aj) ~ + 11(1 - i^Mn^iR^) , 

where V' G is such that 1 — vanishes on a neighborhood of dft, and fl is 

a compact manifold with boundary in which f2 n supp('0) embeds isometrically; for 
example = n [— i?, i?]" with periodic boundary conditions and R sufficiently 

We use H''{il) to denote the space of extendable elements, with no boundary 
conditions. For D, an exterior domain, H'^{Q) consists of restrictions of functions 
in _ff^(M") to r2 with the quotient norm (minimal norm of an extension); for fl 
compact we embed f2 in a compact manifold fi' without boundary, and H'^{fl) 
consists of restrictions of elements H'^{il'). By elliptic regularity, _ff''(ri) C H^{n). 



2. Preliminary reductions 

In this section, we reduce the inequaUties in Theorems 11.11 and 1 1 . 2 1 to estimates on 
solutions to a pseudodifferential equation defined in a coordinate chart near the 
boundary. We start by considering the case of Theorem 11.11 

For ft = M" \ /C, we take -0 S (fi) such that l — ip vanishes on a neighborhood 
of dfl = die. Then vq — {l~ ijj)v satisfies the inhomogeneous Schrodinger equation 
on M": 

{idt + A)vo = [0, A]v , volt^o = (1 - ij)f ■ 

Here, (1 - V')/ G and by Q we have [A, 0]^ e L^H'-^R"-). (Ahhough 

stated only for s G [0,1] in j^, it is easy to see by a bootstrap argument and 
interpolation that (|4]) holds for all s > 0, where H" is the is the intrinsic Sobolev 
space for the Dirichlet/Neumann conditions as above.) The Strichartz estimates 
for Vq then follow from Proposition 2.10 of [4] together with Sobolev embedding. 
While [4 considers the case s G [0, 1], the result of Proposition 2.10 of [3] follows 
for all s > 0, since the free Schrodinger propagator exp(«iA) on R" commutes with 

We are thus reduced to establishing estimates on the term ijjv. We isometrically 
embed a neighborhood of supp(7/;) into a compact manifold (57, g) with boundary, 
where dfl — dfl. Then vi = ipv satisfies the inhomogeneous Schrodinger equation 
on fl: 

{idt + Ag)vi ^ [A, tp]v , vilt^o = -0/- 

By (|4]), we are reduced to establishing the following estimate over a compact man- 
ifold with boundary f2, 

Here we use that [A,-0] vanishes near dil, hence maps 

We next take a Littlewood-Paley decomposition of v in the x variable with 
respect to the spectrum for Ag. Precisely, we write 

oo oo 

V = /3o(-Ag) + J2 l3{'^-''{-A,))v ^ ^ , 

where X^jli /^(■^^^"'•s) = 1 for s > 2, and /3 is supported by s £ [5:2]. The low 
frequency terms are easily dealt with by Sobolev embedding, since the right hand 
side of ([8]) controls the L^H^~2 norm of v for 2 < p < 00. The following square 
function estimate holds, for example by heat kernel methods, 

Mk^^L.^n, « IKE, i-.f )^IL..,(o) < (e, (9) 

where we use p, q > 2 in the last step. By orthogonality, the desired estimate ^ 
would then follow as a consequence of the following estimate. 

Finally, we divide [0,r] into intervals of length 2^-' and note that, since p,q > 2, 
by the Minkowski inequality it suffices to prove the above on each subinterval; that 
is, for T = 2~K To summarize. Theorem 11.11 is thus reduced to establishing the 
following semiclassical result. 



Theorem 2.1. Let be a compaet Riemannian manifold with boundary, and Ag 
the Laplace- Beltrami operator, subject to either Dirichlet or Neumann boundary 
conditions. Suppose that p > 2 and q < oo satisfy ^ and ([7]). 

Suppose also that, for all t, v\{t, ■ ) is spectrally localized for — Ag to the range 
[iA^,4A^]. Then the following estimate holds, uniformly over X, 

hx\\L';__,L->(n) < A'(A5||i;A||i2_^z,2(n) + A-5||(0, + Ag)wA||L2_^i2(n)) . (10) 

We observe that Theorem 11.21 also follows as a consequence of (fTOt . To see this, 
we divide [0,T] into subintervals of length A~^, and note that for v\ — exp{itAg)f\, 
on each subinterval the right hand side of (|10p is bounded by A'* 1 1 /a 1 1 ^2 (j^) • Summing 
the LP norm over a total of « A subintervals leads to 

Applying the square function estimate ([SJ as above yields Theorem II .21 

We will establish ([10]) by the methods developed in [20] and [3] to obtain dis- 
persive estimates for the wave equation on manifolds with boundary. We start 
by taking a finite partition of unity over fi, subordinate to a cover by coordinate 
patches. We restrict attention to a coordinate patch centered on dfl; the interior 
terms can be handled by the methods of :5 , or by the parametrix construction of 
this paper. Thus, let ip € C^{il) be supported in a boundary normal coordinate 
patch along 951. The function ipvx is not sharply spectrally localized, but does 
remain spectrally concentrated in frequencies < A. Precisely, for all fc > 0, 

\\^^x\\Ll_,H''(n) - hx\\Ll_,H>'(n) ^ >^''\\vx\\lI_,lhq) , (11) 

and the same holds with v\ replaced by (idt + A.g)v\. 

Letting a;„ denote geodesic distance to the boundary, and x' coordinates on 557, 
in boundary normal coordinates the Laplace operator takes the form 


where p — \/det gik and g*-' denotes the inverse of the metric gik- Furthermore, 
gm _ gm _ ^.^^ gQ |;jjgj.g a^]-g mixed dx'dx„ terms. 

We now extend the metric g{x',Xn) in an even manner across Xn = 0; the 
new metric g(x' , \xn\), which we also denote by g, is defined on an open subset of 
M", and is of Lipschitz regularity. We extend the solution ijjvx in an odd or even 
fashion, corresponding to Dirichlet or Neumann boundary conditions, to obtain a 
C^'^ function. We will assume is chosen so that ip(x\xn) is independent of Xn 
near a;„ = 0. Since the extended Laplace operator is even, the regularity of g and 
v\ show that the extended solution satisfies the extended equation across x„ = 0, 

(idt + Ag)(V'VA) = [Ag, V]wA + i^iidt + Ag)wA , 

where AgUA is extended oddly/evenly as is v\, and ip is even. 

By choosing suflticiently small coordinate patches, and rescaling if necessary, we 
may assume that g extends to all of M" , such that 

||g'^' - ^'^llco.HK") < CO « 1 , g'' = S'' if \x\>l. 

The odd (respectively even) extension operator maps functions in H^{Wi^) satis- 
fying f{x',0) — (respectively dx„f{x',0) = ) to functions in iJ''(K"), provided 



r e [0, |). The extension also commutes with differentiation in the x' variables. 
We observe that multiplication by functions such as g or p preserves _ff''(R") for 
r G [0, |), and multiplication by dxP preserves H^(M.") for r G [0, ^). This can be 
seen, e.g., from the fact that (^)^~^ is an A2 weight in one dimension, and that 
dxP is a Calderon-Zygmund type multiplier in a;„. 

It follows that the bound pT|) holds to a limited extent for the extension of tpv\ 
to K". To quantify this, we introduce the following family of norms, for r > 0, 

11/11^.,.= J2 (A-i"i|i5,"'/iiL^(E")+A-i"i-nia,",/ii^.(K„)), 


and observe that ||/||^<t,a < ||/||^fr,A if < cr < r. Here N is taken to be a fixed but 
sufficiently large number, which we allow to change in a given inequality. However, 
for the results of this paper N need never exceed n + 2. 
By ([ni) and the above, it holds that for 2 < r < | 

>^^\\^Vx\\Ll_,H'-->-(m") + A"3||(iaj + Ag)(-0UA)||L2_^^.-2,A(R„) 

< A5||z;A||i2_^i2(o) + A-5||(iat + Ag)wA|iLj_,L2(f2) . (12) 

This bound also holds if we replace Ag on the left side by the divergence form 
operator dig^^dj, since the difference p~^{dip)g^Wj maps H"^'^ iJ*""^'^ with 
norm A, provided r G [2, |). Since subsequent estimates will be only in terms of 
the left hand side of ((H]), we may thus set p = 1, and replace Ag by di g^-'dj. 

We next reduce matters to considering solutions that are strictly frequency lo- 
calized on R", and which satisfy an equation with frequency localized coefficients. 
For each /i, we form regularized coefficients g^ by truncating the Fourier transform 
of the g*-' so that 

supp(g;f ) c{iei <c/i}, (13) 

for some small constant c. We observe the following estimates 

hil-g''\\L-(R'^)<f^-\ I|5,%^IIl~(R")<mI"I-S I«I>i. (14) 

With slight abuse of notation we now set 


We will prove in the next section the following estimate for u^{t,x) defined on 
[0,/i~^] X R", which are localized to spatial frequencies w /i. 

Lemma 2.2. Suppose that {p,q,s) are as in Theorem ] 1. 11 andu^it,^) is supported 
in the region ^/x < |^| < Then 

\\u^\\l''_,L'> < 2 +/i^||(iat + Agju^lli,! r2 . (15) 

Furthermore, */u^(t,C) addition localized to |^'| < then (|15p holds for 

p > 2 and q < oo satisfying ([7]) with s > 0; that is, without the restriction 

In the remainder of this section we reduce (jlOp , and hence Theorems 11.11 and 
11.21 to establishing Lemma [121 

We start by considering the frequency components /j, < A of ipvx (by which we 
understand its odd/even extension to R"). Let Pfi^D) denote a Littlewood-Paley 



localization operator on K." to frequencies ~ ii, and consider — (3fj_{D){tpv\). 
For /i < A, ()15p and the Schwarz inequality imply 

<M''(A^||u^||i2_^z,2 +A 2||(iat + Agjw^||i2_^i2) . (16) 

Since s > in our bounds, we may sum over dyadic values of /i < A to establish 
(|10p for the cutoff of ipvx to frequencies < A, provided we bound the i'^ norm over 
/I of the terms in parentheses in (flH)) by the terms in parentheses in (|T0| . By (IT2]) , 
this is a special case of the following estimate, which we establish for all re [2, |), 

I] All li^ II 2, ,+X-'\\{tdt + Ag^^)uXL- 

< A'||V'i;a||l2 +A-3||(iat + Ag)(VwA)||L^ ,h-2,. . (17) 

Since is bounded and commutes with differentiation, this will follow from 
showing the fixed time estimate 

(s2 mA^~ AMiM\i.-2,X < mi^v^Wh-^-^ ■ (18) 

In this estimate we may replace Ag — dig^^dj by g^^didj, and similarly for Ag^. 
This follows since the difference {dig^^)dj maps H^^^'^ — > H^^^'^ with norm A. By 
the Coifman-Meyer commutator estimate (see [221 Prop 4. ID]), for cr e [0,r — 2], 

(j2\\iP^'S\d'AM\\jiy <||V'«A|k.+i < A'^+i||V^,;A||a.-i,. . 

The same holds with Ag] replaced by \dx' ■, [/3^, Ag]], since this has the effect of 
differentiating the coefficients g*^ in x', which remain Lipschitz. Hence 

m^,g\dl{^vx)fn^^...^ < A||V«a||/^.-i,a . 
Next, using (fT4|) and interpolation, we obtain for < (j < 1, 

E ll(g- gM)5'/^/'(V'^^A)||?f.) ' < ||9.(«|k" < A-+1 llVnlli/^+i.- 

Commuting with dx' as above yields 

E ||(g - g^)52/3^(V'i;A)||?,.-2.. < A Uvy\\n.-^,. , 

completing the proof of p8)) . 

To handle frequencies /i > A, we consider separately the tangential and normal 
components of u^. Thus, we decompose 

/3,.(0 = r^(0 + r;(0, 


supp(r^) c u : le'l < flCnl} , supp(r;,) c u : |e'| > \U} . 



First consider a tangential component — T'i^{iljvx). The key idea is that 
and Ag^^u^ are supported where |^'| ~ fj,, whereas dx' weighs as A acting on u^. 
Consequently, for each fixed time, 

and similarly for {idt + Ag^)w^(. Noting that the proof of (IT71) works the same with 
with replaced by T^, then and ^7} together yield 

\Ml''^_^L'^ < (^) A"(A5||t;A||L2_^i2(o) +A-^||(ia, + Ag)wA|lLj_,L2(i2)) ■ 

We sum this over the ^/X disjoint intervals of length contained in [0, A~^], 
and over dyadic values /i > A, to complete the proof of (fTOt for localized to 
frequencies |^'| > |^„|. 

For a normal component = T^{'ipv\), we do not have sufficient decay in powers 
of /i/A to apply the above steps for large s. Instead, we use the fact that P3|) holds 
for all s > in this case, and deduce large s results from the case s — together 
with Sobolev embedding. Given a triple {p,q,s) satisfying ([T]), let go G [2, j) be 
such that 

1 I _ s 
qo q n ' 
hence {p,qo,0) satisfies Then (IT5|) implies 

\Wt^\\L''_,Lm < fl^\\Ui,\\L2 + \\{idt+ Ag^)Uf,\\Li r2 . 

Summing over intervals yields, for r E [2, |), and /i > A, 

II"mIU''_,l<!o < \\u^\\l^_ + \\{idt + Ag^)u^||ii r2 

We apply this inequality to X^^dx'Up,, and observe that 

||[A"^a:r', AgJu^||ii_^^,.-2.A < A||ii^||ii_^^,,A < A^||u^||i2_^^.,A . 

This holds since multiplication by the tangential derivative dx'gfi preserves H^~^ '^, 
provided r < |. Repeated application yields 

\\{X^^dx'TUf,\\LP^_^L'>0 < (^) (A5||M^||i2_^^.,A +||(0t+AgJu^||il_^^.-2,A) . 

We next apply Sobolev embedding to yield 

\\u,,\\li_,L'> ^ fJ-"\\\dx'\ " Wa.||l^_iL™ 

<H"X " sup II (A Ox'Tu^Wlp r,,o 

\a\<n ^ 

We choose re (2 + ^, |), and apply (|17p . We then sum over dyadic values of /i > A 
to establish (fTOj) for localized to frequencies |^'| < ||^n|. 

In proving Lemma 12.21 using the results of |20) , it is convenient to work as in 
that paper with a first order equation. To do so, we start by rescaling the time 
interval of length in ([T5t to an interval of length 1, by considering the function 



v{t, x) = Uf^{^^^t, x). This replaces Ag^ by fi^^ J2 9i{g]^djv), which is a first order 
symbol for |^| « ^. 

We can modify this operator away from the region |^| « ^ without changing 
the estimate p^ . To fit into the framework of [20], we want to work with an 
operator such that solutions to the homogeneous flow remain frequency localized 
to 1^1 « /i if the initial data is supported there. For /3^(^) a Littlewood-Paley cutoff 
to frequencies « fi, we thus set 

P,ix,D)v = fi-'l3^{D)d,{^p,l3^iD)v) + - /3^{Df)Av . 

Then is an elliptic self-adjoint operator on M", with a symbol Pfi{x, ^) such that 

||a|,^^P^(x,o-M"'/|| <co«i. 

We will prove in the next sections the following result. Here we replace the 
parameter by A, and v by u, to follow the notation of [201 Theorem 2.2]. 

Theorem 2.3. Let he obtained by truncating g'^ (x' , |x„|) to frequencies < cX, 
and define P\ as above. Suppose that u\{t,x) is localized to spatial frequencies 
1^1 G [^A, |A], and p, q, s satisfy the conditions of Theorem \l.l\ Then for small e 
the following holds 

II"a||lpl<j(M") < A''+p (^||wa||l-l2(js,„) + \\{idt + -Pa)ma||lil2(R")) ■ (19) 

If in addition u\ is localized to |^„| > then ([15)) holds if s > 0, p > 2, q < oo, 

andl; + ^ = ^-s. 

That this implies Lemma 12.21 follows from the fact that 

II'"a|1l~l2(R") < \\u\\\l^^l^R") + Wiidt + PxhxWLlL-^iR") , 

which follows from self-adjointness of Px. We further note that, by the Duhamel 
principle, it suffices to prove the estimate for the case that {idt + Px)ux = 0. In 
particular, it would suffice to prove (|19p with Ll replaced by on the right hand 
side, as is the case in [20l Theorem 2.2]. 

3. Angular Localization 

We now proceed as in |20[ §3] , and decompose ux in the frequency domain into terms 
localized to angular sectors. This is done by taking a finite dyadic decomposition 
in the f„ variable, where ^ = (Cfn)- Precisely, we write ux — '^^^lUj, where 
Nx — ^ log2 A , and where for 1 < j < Nx, 

supp(u,(i,0) C { le'l e [iA,4A] , € [2-^-^X,2-^+'X] } . (20) 

The "tangential" term is un^, where 

supp(ujv.(t,0) C {l^'l e [iA,4A], 16.1 <Ai}, (21) 

and ui is localized away from tangential co-directions 

supp(5Ii(i,e))c{|e|€[iA,|A], |C„|>|A}. (22) 

The term ui can be handled by the same methods as U2, so we restrict attention 
to 2 < j < Nx. 



The energy of the solution u\ travels along bicharacteristic curves 

For curves which satisfy \£,(t)\ sa A for t e [0,£], then |i| « 1, and i„ w A~^^„. 
In addition, if |f„(0)| w then we wiU have |^„(t)| w 2^^ X for |i| < e2--''. 

Setting 0j = 2~^, the function Uj is thus localized to bicharacteristics which remain 
at angle 9j to the boundary for times up to e6j. As in |20j, we will have good 
estimates on the term Uj on slabs of width e2^^ in t. 

For each j > 2, we let pj{x,£_) be the regularization of the symbol p\{x,£,), 

1 — i 

obtained by truncating the metric coefficients g^ix) to frequencies < cX^Oj ^ . 

Let Sj^k denote the slab {(x, t) :tG [keej, {k + l)£0j]}. The slab S = [0, e] x R" 
is then the union of the Sj^k as k ranges over the integers from to dj^ ■ Define 

^(9) = 

Note that ^ is equivalent to a{q) > i . As in [20], the crucial matter is to now 
show that if Uj satisfies the equation 

then the following estimates are valid for 2 < j < N\ 


+ XUj\\{Xh-Krr)-W\LHS,,,) (23) 

+ X-i9-i\\{Xh-^x„)'G,h^s,,S, 
In case j = N\, that is 9j A^^, then 

h,l|L.L.(s„.) < A^+^<''' + + G.h^ms,.,)) ■ (24) 

Let Cj^k denote the term in parentheses on the right of (|23p (respectively the 
term in parentheses in ([M)) when j — N\). A modification of the arguments in [501 
§6] show that if k{j) denotes any sequence of values of k for which the slabs Sj^k(j) 
are nested, that is, S'j+i C Sj ^i^j^, then 

2^<fcO-) < \\ux\\Ul-(s) + \\i^dt+Px)ux\\i.L-[S) ■ (25) 

The key modification arises from the fact that the symbol p\{XtS^) is not homoge- 
neous of degree 1 in ^. This changes the form of the conjugation of (which is 
the operator P\ after a space-time rescaling by 9j), by the wave packet transform 
T^, which occurs on the bottom of [101 P- 137] and top of [201 P- 145]. The new 
relation is 

T^Q^r; =Dq + ia + K. 

Here, q is the symbol pj rescaled, and Dq the Hamiltonian field of q. The real 
valued function a is defined in (|3ip below. The operator Dq + ia is simply the 
conjugation of Dq by the unimodular function exp(i^/;(t, a;, ^)). This conjugation 
does not affect the arguments of [101 §6], since the only estimates used on K are 
absolute value bounds [101 (6-21)], that follow from the estimates [101 (6.31)]. We 



also note that the estnuates in [2^ use {idt + Px)ux e L?L'^{S), but as noted after 
Theorem [2T3] above this is unimportant. 

The estimates of Theorem 12.31 will then follow from and the branching 

argument on ^^20^ p. 118]. 

In the proof of (|23p -(|24 p . we will from now on work with a fixed j, and will 
abbreviate 9j = 6. We work with an angle-dependent rescaled Uj, setting 

u{t, x) = uj{ejt, e^x) , F{t, x) = 9jFj{9.jt, 6jx) , G{t, x) = ejGj{6jt, Ojx) , 

and q{x,S,) — 6jPj{6jX,9j^S,). Set ^ — X9j, so that q{x,^) w ^ when |^| « fj,. 
Additionally, if |^| ~ fi, then q{x,^) satisfies the following estimates; see [20] (4.1)]. 

We then have 

DtU-q{x,D)u^F + G, 
and the frequency localization condition holds 

'W\ e [l^lA^^],\^n\e[\^l9,2^l9], > a^-^ , 

After translation in time, the estimates psp reduce to showing that, over the slab 
S = [0, e] X M" , 

Mlplhs) < (\\u\\l^lhs) + \\F\\l^lhs) 

+ fii9^\\{^i^x,,)-^u\\L2^s) (27) 

+ fi-i9-^2\\{^J.hnfG\\L2(^S) 

The estimates in ^M^ reduce to showing that, for = , 

IMl^lhs) < A^'+^e^^") (||"IIl=»l^(S) + ||F + G|Uii2(5)) . (28) 

supp(w(t, •)) C 

To establish the inequalities ([27| and (|28l) . we use a wave packet transform to 
construct a suitable representation of u. Define the linear operator on Schwartz 
class functions by 

{T^f)ix,0=f^^ I e~^(^'y-)g{fiHy-x))fiy)dy, 

where we fix (7 a radial Schwartz class function, with 5 supported in a ball of small 
radius c. Taking ||g||L2(B") = (Stt)"-? , it holds that T*Tf, = / and \\T^J\\l2(ri^^) = 
II/IIl=(R5;)- We set 

u(t,a;,e) = iT^uit,-)){x,0- 
By Lemma 4.4 of [20], we may write 

{dt - d^q{x, ■dx+ d^qix, ■ + iq{x, - «^ ' d^q{x, x, (,) 

^F{t,x,C) + G{t,x,0, (29) 

where, over S = [0, e] x M^"^, the quantity 


is bounded by the right hand side of ([27|) when 9 > fj,^^ , and the quantity 

11-^ + G'||Lli2(5) 

is bounded by the right hand side of ((28|) when 9 — fi~i . The proof of this lemma 
rehes only on the bounds ()26|) . and thus applies in our situation. Also, given the 
compact support of g, it can be seen that the ^ support of u, F, G is contained in 
a set where |^'| w /i and ^„ w 9fj, (or |^„| < ^2 when 9 — fi"^). 

Let Qr,t{x,£,) denote the canonical transformation on M^"^ = T*(R") generated 
by the Hamiltonian flow of q{x^^). That is, Qj.,t{x,C) is the time r solution of 

X = d^q{x, C), i = -dxq{x, ^ , (30) 

with initial conditions {x{t),^{t)) — {x,^). Since q{x,^) is independent of time, 
0r,t = 0r-t,o • Also define 

a{x,^) =q{x,^)-^-d^qix,0, ^{t,x,^)= f a{es.t{x,0)ds. (31) 


It follows by time independence of q that a{Qs.t{x, £,)) ds = ip{t — r, x, ^) . 
Equation (P^l) above allows us to write 

{t(i,a;,e) =e-*'^(*'-'«)u(0,eo,t(x,0) 

+ ^ e-»'^(*-'-^-^«)(i?(r,e.,t(x,e)) + G(r,e,,t(x,0)) • 

In the next section we will establish the following estimates for solutions to the 
homogeneous flow equation. 

Theorem 3.1. Suppose f G i^(R^"^) is supported in a set of the form 

Define Wf{t, x) — T* [e"*''''-*''-' (/ o 0o,t)] (x) . Then the following estimate holds for 
s > 0, p > 2, and q < 00 satisfying ([5|) and 

\\Wfh.LHS) < . (33) 

For f £ L'^(M^"^) supported where \^'\ < fi, |^„| w /x, estimate ([55)) holds with 9 — 1, 
for s > 0, p > 2, and q < 00 satisfying 

Since T*T^ = / , it follows by the preceeding steps and variation of parameters 
that this implies the estimates (pS)) . as well as the estimates (P7| in case (5 = 0. 
The reduction of the estimates (|27p to Theorem 13.11 for G 7^ requires the 
spaces introduced by Koch and Tataru [T^, and follows exactly the arguments on 
PPI p. 124-126]. The key fact used in that proof about the Hamiltonian flow of q 
is that in ~ 9 on the support of u(t, x, ^), which holds in our case. 



4. Homogeneous estimates 

In this section we prove Theorem 13. II By duaUty, it suffices to show that 

II < m''^+'^^'^(')||^IIl.'l,'(s) , (34) 

where F{t,£,) is supported as in ([5^ . and we recall that 

I, 2 g ' — 

Let Wt denote the fixed time operator Wtf = Wf{r, x)\r=t- We will show that 
IIM^.M^; hi^L^ < ip-' + \t- r\)-^iti-'e-^ + \t- r|)-^ , (35) 

\\WrW:\\L2^L2<l. (36) 

Interpolation of these estimates yields 

IIW^.WIIl.'^l. < f^^^'-'^Hf^-' + \t-r\r'^'^'--^\f^-'0-^ + \t-r\n'^'-'^- (37) 

For n > 3, we have | < 1 — | < ^^^^ (^ ~ |) ' hence we may ignore the term |i — r| 
in the last factor to obtain 

\\WrW:\\^,-^^,<^,'^^+'^'>e''^^''^\t-r\--. . (38) 

In case n — 2, we use that 6* < 1 to bound 

\\WrW:\\,,^,, < ^^^-i)ei^^-i\^^^ + \t - r\yi^'-'i^ . 

Since | < |(1 — |), in this case we again have ([55]) . In either case, the Hardy- 
Littlewood-Sobolev theorem gives 

In case 6* = 1, as for the normal piece, estimate ([55)) follows for all p, q satisfying 
([7]) with s > 0. Hence for the normal piece the condition ([SJ is not necessary. 

The inequality p6p follows from the fact that is an isometry and 0o,t(a;,^) 
is a symplectomorphism, and hence preserves the measure dx d^. The remainder of 
this section is devoted to proving ([55]). 

The action of WrWf on a function h{y) can be expressed as integration against 
an integral kernel K{r, x;t,y), defined by the formula 

g.(C,.-.>-.^(r-t,.,c)-.(c*„-,y-.o.>^(^^ (y _ zt^,))g{^i (x - z))(3e{0 dz dC . 

Recall that 5 is supported in a ball of small radius and f{x, ^) is assumed to have ^ 
support in a set of the form (j32p , which is essentially preserved by the Hamiltonian 
flow of q for time e. Hence Pe{C) can be taken to be a smooth cutoff to a set of the 
form ([32]) . For convenience, we take /?e(C) to be a product of a cutoff in (' and a 
cutoff in Cn . 

Since Qt,r — ©t-r.Oj it suffices to consider the case r = 0. We abbreviate 
{zt,o,Ct,o) by {zt,Ct), so that 

dtZt{z,Q = d(;qiz,C) , dtCtizX) ^ ~d^q{zX) , (zq, Co) = (z, C) ■ (39) 
The kernel K{{), x; t, y) takes the form 

^^C,x-z)+^i,{uzt,Ct)-^{Ct.y-'Zt) ^(^i _ (a; _ z))pe{C) dz dC . (40) 



Theorem 4.1. Suppose {zt{z ^ Q , C,t{z , C)) are defined by p9p anddz, d^ denote the 
z and Q gradient operators. Then if \C, \ /i, and C,n ~ fJ-&, or |C„| < /i^ in the case 
9 = fJ.^^ • the following bounds hold, 

\d,zt--l\<t, \dczt\<n-H, 

as well as the more precise estimate 




d1q{zsXs) ds 

Furthermore, for second order derivatives we have 

\dlzt\ < {fih} , 

\dzd^zt\ < ^i^'^t{^.^t) , 
Finally, for I > 2 we have 

lid I . 








Proof. The proof is a rescaled version of Theorem 5.1 and Corollary 5.2 of [20], but 
for completeness we sketch the details here. 

Differentiating Hamilton's equations one obtains 






dzd^q d^d^q 
-dzd^q -d^dzq 

To keep all terms of the same order in /i, we take the following rescaled equation, 



dzZt ^idc^zt 
H'^dzC,t dt^Ct 

M^(z,C) = 


dzZt nd(;Zt 
fi~^dzC,t d^Qt 


dzd^q iid^d^q 
-H^^dzd^q - d^d^q 

The key estimate on is that, for j + fc = 2 . 

^-^t, iffc = 2, 


if J = fc = l, 
if J = 2. 


This follows from (f26|) and the property |(9tZt)„| for t e [0,e], when 9 > jj, 2. 
When 9 = the estimates (l26l) are uniform over \/3\ < 2, and (l47t also follows. 
Gronwall's lemma now gives that 

\dzZt\ + fi \dQZt\ + /i" V^Ctl + McCtI < 1 • 

Integrating (j46| and using (|47l) yields (|4T1) . The estimate \d(;(t — I\ S i can then 
be substituted in the integral equation for d(;Zt to give (j42]). 

To show the higher order estimates (j45p . we work with the equation 








Here Ei (t) is a sum of terms of the form 

(/di4+^g)(^t,Ct)(4^4^^t) ■ • ■ (/^'^■4^^t)(/i'^+^-^d;^-^^Ct) ■ • ■ ifi'^^'-'d'^^^'Ct). 



Similarly, E2{t) can be written as a sum of such terms, but with the first factor 
replaced by C,t)- In either case, Im < I for all m and + - ■ ■ Ij+k = 

I. The estimate (j45p now follows by an inductive argument which uses the bounds 


if J =0, 
if J>1. 

Estimates (|43|) and (|44p follow similarly; see the proof of Theorem 5.1 in [2^. □ 

We start the proof of ([551) by noting that absolute bounds on the integrand 
in (|40)) easily yield 


which gives ([55]) for < t < We next consider the cases < t < fi~^9~^ 

and fi^^O^^ < t < e separately. In these two cases, we will respectively integrate 
by parts in PO)) with the two vector fields 

1 - ifit^^ix - z - d^Ct ■ (y ~ zt))' ■ rfc 

1 + fit-^\{x - z - d<;Ct • {y ~ zt)y\ 

1 - iiJit~'^{x - z - (i(;Ct • {y - Zt)) ■ d(^ 

^ 1 12 — 

1 + iit-^\x - z - d^Ct ■ {y - Zt) I 

Both L' and L preserve the phase function in (HUl) . This can be seen by noting 

that ip{t,Zt,Ct) = Jq ct{s, ZsXs) ds, and observing that 

% (^J q{zsXs) - Cr ■ (dc'7)(^s, Cs) ds^ + Ct • %Zt = . 

The expression vanishes at i = since d^zo — 0, and Hamilton's equations show 
that the derivative of the expression with respect to t vanishes identically. 

We begin with the case where fi^^ < t < fjL^^O^^ . Recall that /3e(C) is the 
product of smooth cutoffs to sa 9^ and |C'| ~ ^J■■ Let {S,'^} be a collection of 

{fit)^~ vectors on the lattice of spacing fi2t^2 ^ and (j) a cutoff so that 

where ^™(C') = (C - . 

Define Km{t,x,y) as the integral in (|40p with Pe{C) replaced by l3e{C)4'm{C) so 
that K{0, x; t, y) = J^m. ^mit, x, y) . 

By the estimates PT|) and psj) . we have that for fc > 1, 

|(^^<^^d^)'^a| < 1, for a{t,z,C) = ii^t~^ Zt or a(i, z, C) = i^^^cCti 

which holds not just for t E [jj.^^ , pi^^O^'^] but for all t E [ii^^,e]. Furthermore, 
\{n^t~2d(^,)''(f)j^[(')f3g{C,)\ < 1, since we do not differentiate in Cn- Therefore, inte- 
gration by parts yields the following upper bound on Km{t, x, y), 

{l + ^lh-'^\{x^z~dc(:t■{y-zt)y\)-'' 

X (l + |y - Zfl) ^ [l + ^'^\x~ z\)^^ dzdC, . 
We set = Since t < fi^^9^^ , we have for ( E supp(/36/0m), 

IC-e™|<M^i"^- (48) 



Recall that zt — zt{z,() is the spatial component of 9f,o(z,C)- We let a:™ = 
ztix,£,m) denote the spatial component of 6t,o(a^, ■fm)- We then claim that, for 
C e supp(/3e0„), 

^it-i\x-z-d^Ct■ixT-zt)\<l + ^i\x-z\''. (49) 
Assuming this for the moment, we dominate the integrand for Km by 

(1 + fih-i\{d^Ct ■ [y - xT))'\)-''{i + ^l^y- zt\)-''{i + ^^^x- z\)-'' . (50) 

By dUD and (gS]), we have - zt\ < ^^h^ + \x- z\. Thus, since jd^^Ct - I ^ 1*1, 
we conclude that 

^^'^t-i\{dcCt ■ {y - xT)y\ > M^^^^Ky - xTYl - i^h^y - xT\ 

> tih-^iy - xTYl - ^ih^\y - zt\ + \x- z\) - \t\ . 

The negative terms on the right here are small compared to the last two terms 
in ((50)) . Therefore, we have 

\Km{t,x,y)\ < ^l'^9t-'^ {I + ^lh-'2\iy - xTYiy'' , 

which follows by observing the rapid decay of the integrand in z, and that the 
volume of supp{(j)mPe) is comparable to ^~^9t ~ . 
We next observe that, by (H^ and the estimate 


we have that 

\{xT - x\) - 2^i-hiu - 6)1 « ^^~'t \U 

and since |6„ — 61 = I Cm ^ 6'!, conclude that 

f,h-i\ixT-x[y\^f,-hi\em~ei\- (51) 

Since the lie on a /i^i^^ spaced lattice, we may sum over m to obtain 

\K{o,x;t,y)\ < f^^or"^ ^(i + ^ih-i\{y - xTYiy'' < fi'^Or^ , 


yielding ([35]) for < t < ^j.-^Q-'^. 

To handle the case fi^^O^^ < t < e , we modify the above proof by considering 
an 0{fi^t^) collection of vectors {^m} in a /i2t~2 spaced lattice in R", and an 
associated partition 4'm{C) = 4'i^J'~^t^iC ~ Cm)), satisfying 


We now define Km{t, x, y) as the integral in pOj) with (ie{C) replaced by f3g{()(t)m{0- 
Here, since iJL^t~^ < /i6', we have 

|(A*^t-^d^)Vm(C)/3e(C)l <1- 

Integrating by parts with respect to the vector field L now shows that Km.{t,x,y) 
is bounded by 

{l + fih-i\x - z - d^Ct ■ {y- zt)\y^ 

X (l + ^2 |y _ 2^1) (1 -|_ ^2 _ 2;|) dzdC 



Using (|49|). which holds for any i e [/i ^,e] , we proceed as before and conclude 

\K^{t,x,y)\<^l'^t-'i{l + ^lh--^\y-xT\)~''. 
The n-dimensional analogue of ()5ip is valid here, so we may use the spacing of the 
as above to sum over m and obtain (|35|) for t > that is, 

\K{0,x;t,y)\<fi^t-^ . 

Returning to (|49|) . we first observe that by estimating the Taylor remainder 
using dSll-dlSl) and (gH]), the following holds 

/ih"' IxJ" -zt- {d,zt){x -z)- {d^Zt){U - 01 < 1 + A^la; - ^1^ • 
Furthermore by (|¥T|) and we have 


From the fact that (z, C) — {zf, (t) is a symplectic transformation, we have 

dc.Ct ■ dzjZt - d^^zt ■ dz^Ct = % , 
where • pairs the zt and Q indices. Lastly, by (|411) . 

^j,H~^\d(^zt\ \dz(t\ \x — z\ < fi^t^\x — z\ < fi^ \x — z\ . 
These facts now combine to yield □ 

5. Applications to semilinear Schrodinger equations on exterior 


In this section, we assume that = \ /C is the domain exterior to a smooth 
non-trapping obstacle /C (or any exterior domain where (0]) holds) . We consider the 
initial value problem for the following family of semilinear Schrodinger equations 
in 3 + 1 dimensions, 

idtu + Au±\u\'-^u = 0, u(0,x) = /(x), (52) 

satisfying homogeneous Dirichlet or Neumann boundary conditions 

u{t, x)\q^^ — , or diyu{t, x)\q^ = . (53) 

Precisely, by a solution to ([5^ - ([55)) . we understand that, with F{u) = ±|u|''^^m, 
and u{t) denoting the function u{t, •), 

u{t) ^ e''^ f + i I e'^'-'^"^ F{u{s))ds, (54) 

where exp(ztA) is the unitary Schrodinger propagator defined using the Dirichlet 
or Neumann spectral resolution. Defocusing means that ^"(11) = — Since 
we will work with data, the boundary conditions required of the initial data 
in the Dirichlet case are that / vanish on d^; in the Neumann case the boundary 
conditions are void, i.e. / is the restriction to $7 of a general function in H^{M.'^). 

Planchon and Vega showed in [17] that, for 1 < ?■ < 5 and defocusing nonlinear- 
ities, one has global existence of solutions to the Dirichlet problem for / e H^. A 
crucial ingredient in their proof was the estimate in Theorem 11.11 with p — q — A 
and s = J. They combined this with local smoothing estimates near the boundary 



to obtain well-posedness of solutions for time T > depending on Conser- 
vation of energy and mass 

f 1 1 

/ 7:\dxu(t)\'^ H luUW^^ dx = constant in t 

Jo 2 r + 1 

dx = constant in t 


can then be used to establish global existence of solutions. Given that Theorem 
holds for either Dirichlet or Neumann boundary conditions, we remark that our 
estimates can be used in the same way to obtain global existence of the solution to 
the Neumann problem. 

In the critical case r = 5, local well-posedness for solutions with / G H^, and 
global well-posedness for small data, was proved by Ivanovici in [12] . under the 
assumption that JC is strictly convex, and u satisfies Dirichlet conditions. These 
assumptions are necessary for the Melrose- Taylor parametrix construction used to 
establish Strichartz estimates in jl2) . The Strichartz estimates were in fact shown 
to hold in this setting for the full range of p,q satisfying ([7|), provided p > 2. 
Recently, Ivanovici and Planchon in jT^ extended the well-posedness results to the 
case of general non-trapping /C, for both Dirichlet and Neumann conditions, using 
certain L^L^ estimates from pUj . 

Here, we use our estimates to give a simple proof of the well-posedness results 
for data for the critical case r = 5, and general non-trapping /C. The proof 
proceeds by a contraction argument using a L^L"^ Strichartz estimate. As in jl3j . 
the local results are valid for the Dirichlet and Neumann cases, and the proof yields 
scattering for small data in the Dirichlet case. Precisely, we establish the following. 

Theorem 5.1. Suppose f G H^{fl), where f\dn — ^ if Dirichlet conditions are 
assumed. Then there exists T > 0, and a solution u(t,x) to equation (|54|1 with 
r = 5 on [— T, T] x Q, satisfying 

ueX = C{[-T, T];H^{nj) n L'^ii-T, T];L°°{n)) , 

and the solution is uniquely determined in this function space. Furthermore, if the 
data satisfies < £ for e sufficiently small, one can take T = oo in the case 

of Dirichlet conditions, and T — 1 for Neumann conditions. 

The key ingredient in Theorem 1 5 . 1 1 will be the following Strichartz estimate for u 
given by formula (|54p . and with / and F satisfying the given boundary condition, 

ML-^L^<\\f\\m + \\F\\L^^H^. (55) 

Given (|55p . one applies differentiation and Holder's inequality to obtain 

"I^^ILiffi - ll"llkL-ll"IU??Hi ' (56) 

noting that satisfies Dirichlet conditions if u does. 

We then pose u = uq + v, where uo{t) — exp(ztA)/, and take T small enough 
so that |iwo||Lf,L°' < c, for c to be determined. Estimates ([55]) and ([56]) . together 
with conservation of the norm under exp(itA), show that, for small c, the map 


V I e 

(\uo{s) + v{s)\^{uo{s) + vis))) ds 



maps the ball ||f ||x < c into itself. Similar analysis shows that the map is in fact 
a contraction on this ball, for small c, yielding a fixed point v. If ||/||_ffi l£ £7 then 
one can take T = 00 for the Dirichlet case, or T = 1 for the Neumann case. 

For defocusing Neumann, energy and mass conservation then yield global exis- 
tence. For small norm Dirichlet data, the proof implies \u\'^u E L^{M., H^{n)). This 
yields that such solutions scatter, in the sense that they asymptotically approach 
in the norm a solution to the homogeneous equation. 

In establishing (|55p . it suffices by the Duhamel principle to consider F = Q. The 
proof of ([55]) will be obtained from the following cases of Theorem 11.11 

\\u\\l^'-l'-> < ll/llifi , \\u\\l^l'-' < 11/11^^ • 

The second estimate could be expressed as controlling the L^W^'^ norm of u in 
terms of ||/||_f/i , and we would then apply a fractional Gagliardo-Nirenberg inequal- 
ity to control ||u(t)||L°° by interpolating and W^'^. We can avoid dealing with 
fractional Sobolev spaces on exterior domains, however, by carrying out the 
same steps more directly. The interpolation we will use is the following. 

Lemma 5.2. Suppose that ai, a2 > 0, and u = X^j^o "where 
hjlU- <min(2-J"Vi, 2^"V2). 


Proof. The proof follows by summing the smaller of the bounds, i.e. separating the 


sum depending on whether 2^ > (pi/p2)°i+°2 or not. The bound applies with 

"-^"1,02 2"i — 1 — 1 ' 


We next take a Littlewood-Paley decomposition of the initial data 

where /3(s) is supported in the interval s £ [i, |], and 1 — /?o(s) + J^JLq Pi'^^^'' ^) 
for s > 0. Here, H denotes — A with either Dirichlet or Neumann conditions. Set 

By the spectral localization, 


and we may write u{t) = X^jlo "iC^) i where 

Uj(t) = e e fj, uo{t) =e e /q . 

By the ultracontractivity estimate for H on exterior domains (see Theorem 2.4.2 
and the ensuing comments in [7^ , where /i = 3 in our case) , we can bound 




Together with the case {p, q, s) — (3, 9, ^) of Theorem 1 we have 

l|2-*«,IU3ioe < \\e-^'"f,h^L9 < \\f,\\^^ < 2-*||/,||hi , 
which we combine with Minkowski's inequahty to yield 

/ (f^piuMll-) ^dtV < (f^ \\2iu,\\ls^^) ^ < 11/11?,. . 

In particular, 

sup 2*||u,(t)||i» <PiW, IIpi||l3 < ll/llffi. 

Similar considerations, using the case (p, g, s) — (12, 9, 1) of Theorem 11.11 yield 

sup 2-*||wj(t)|U^ < p2(t) , ||p2||li2 < ■ 

Lemma 15.21 now applies to give the bound 

ll"(i)IU~ <PiWV2(i)^. 
Applying Holder's inequality with the dual indices (|,9) now yields 

ll"lli^L== < / Pi(i)^P2(t)^ dt < \\p,\\l \\p2\\U < WfW'm ■ 

6. Applications to semilinear Schrodinger equations on compact 


In this section we consider a compact 3-dimensional Riemannian manifold il with 
boundary. We assume G : [0, oo) — R is bounded below, with G{0) = 0, and that 

\G'{r)\+r\G"{r)\<{r)i. (57) 

We set F{u) = G'{\u\'^)u, so that 

\F{u)\<{umu\, KF{u)\ < {u}^/' . 

We prove existence, uniqueness, and energy conservation, for initial data uito) S 
H^{fl), to the semilinear Schrodinger equation 

idtu + Au ^ F{u) , u\t=to ^ u{to) , (58) 

satisfying homogeneous Dirichlet or Neumann boundary conditions (I53p . As above, 
by a solution to ([55)) we understand that its integral form holds. 

u{t) = e'(*-*o)'^ (^u{to) ~i e-'^'-'«^'^F{u{s)) ds 


This formulation is seen to be independent of to! that is, if u solves (|59p on an 
interval for some to then it solves the same equation for all to in that interval. 

The key estimates we use involve values of {p, q) which do not satisfy ([5]) . In 
this case, the method of proof yields estimates with a loss of derivatives relative 
to the scale invariant value of s from ([7]). In particular, the following analogue of 
Theorem 12.11 loses - derivatives relative to the case of manifolds without boundary 
considered in [5 . Additionally, there are logarithmic losses due to the endpoint 
p — 2 and q — oo. 



Lemma 6.1. Let n — 3, and suppose that for all t, ux{t, ■ ) is spectrally localized 
for — Ag to the range [;|A^,4A^]. Then the following estimate holds, uniformly for 
6 < q < CO, where F\ = {idt + Ag)u\, 

\\ux\\Ll_,LHn} < CX^-H\ogXf [x^MLl_^L-{n) + ^~^\\Fx\\Ll_^LHn)) ■ (60) 

Proof. We start by noting that the reduction of Theorem 1 2. II to Theorem 12.31 holds 
with uniform constant over q > 6 with p = 2. In particular, in the handling of the 
normal piece, qo = 6 for p = 2, and s < ^ in our estimates, so the use of Sobolev 
embedding works for that piece. Thus, (j60p is a consequence of the estimate 

IIwaIIl^l, < CX^-H\ogXf{\\ux\\L^L^ + \\{idt + Px)ux\\lil^) , 
together with the following estimate, valid if ma is localized to |^„| > 

\\ux\\l^^l-> < CA^-f (logA)2(||uA|UjoL2 + \\{idt + Px)ux\\lil^) , 

where Pa is as in Theorem 12.31 These estimates in turn follows as a consequence 
of the following analogue of 

\\u,\\L2L.^s,.,)<CX'-H\ogX)hf-^[\\u,\\L^LHs,,,) + \\F,h^LH^^^^^ 

+ xhj\\{xh;'^xn)-W\LHs„,) (61) 
+ x-h;'^\\{xh;Kn)'G,\\L^s,.,)) ■ 

To see this, we note that for p ~ 2, the branching argument [20, p. 118] requires 
9^ to converge, and the remaining term 9j ' is bounded by A ' . The additional 

loss of (log A) ^ here comes from the fact that there are ^ log A terms j in the 
decomposition of ux = Uj. We thus have, uniformly in g, 

II"a||l?l. <(iogA)^||(^KP)^||^,^,, 


and it is the norm on the right hand side that is controlled by the branching 

The estimate (|6ip is scale invariant; scaling by 9 reduces it to the following 
analogue of (|27|) . for angularly localized u satisfying {Dt — q{x, D))u = F + G, 

IML^LoiS) ^ //"«(log^)5 6'3-| (^||u||^oei2(5.) + 11^11^1^2(5) 

+ flh^\\{fl-^Xny^u\\L^S) (62) 
+ fJ-^i9-i\\{^J.ixnfG\\L2iS)) , 

where we used that log /i « log A. 

The reduction of (|62|) to homogeneous estimates, that is, bounds on the operator 
W of (pS)) . involves a loss of log/z due to the fact that p = 2. This comes from the 
use of the spaces introduced by Koch and Tataru [16], where the subscript q 
refers to the Hamiltonian flow for ^(a:,^). In case p = 2, one needs to control the 
2-atomic norm Ug of u, whereas Vg C U]^ only for p > 2. To proceed, we note that 
in the atomic decomposition argument of jl6| Lemma 6.4], we may truncate the 
sum u = J2n Vn to n < log/i, since the error is bounded in L°°L'^ by and its 

contribution thus may be estimated in the desired norm using Sobolev embedding. 



Each term Vn is uniformly bounded in , hence the norm of the truncated sum 
is < log//. 

We are thus reduced to establishing the following analogue of (pS)) . 

WWfh^ < c^l'-^log^l)2^2-^\f\\L.^J,.^) . (63) 

To establish we consider WW* as in the proof of Theorem 13.11 Taking 

n = 3 in (j37p . we note the following integral bound for 6 < g < oo, /i and 9 as 

+ i)-(i-f)(/,-i^-2 + i)-^(i-f ) dt < Ctii^'-i^-\logti)0'-l , 
where C is uniformly bounded. The estimate (j63p follows by Schur's lemma. □ 

We use Lemma [6. II to deduce the following analogue of Lemma 3.6 of 5 . This 
version is weaker, both in the logarithmic loss and the loss of A?, but is sufficient 
for our purposes. From now on, we let u\ — pi^X^"^ H)u denote a Littlewood-Paley 
decomposition of w, where A = 2^ and fc > 1. The term fc = contains the low 
frequency terms of m, and the bounds for this term will follow similarly to fc = 1. 

Lemma 6.2. Let u solve ()59|) . Then there are C < oo and e > such that, 
uniformly for 6 < g < oo, the following holds on any time interval [0,T] with 
A-i < T < 1, 

\\ux\\l^[0,T]^Li) < CA"t(logA)^(^||MA||L2([0.T],Hi) + ( ||u|| ([q^tj^^ i ) )'^^^^ . 


Proof. We divide [0,T] into subintervals of length A~^. We apply (pD)) on each such 
subinterval, and square sum over subintervals to obtain 

\\ux\\L-mo,T],Li) < C'A^? (log A)^(^||uA||L2([o,T],ffi) + || -F'a || L2([0,T] .L^)) , 

where Fa — F{u)x. We now take 

2 6 /I 

a = — , r = , e — 6\ 

5' 3 + a' \r 2. 

and observe that 

<X-'\\{u)''{\d^u\ + {u))\\L^L^ 



Sobolev embedding yields 11^2^^, < T^^\\u\\j^2^fji < where 

u^j^-i denotes the sum of ux over A < T^^. Summing (|64l) over A = 2-*=, and 
using Cauchy-Schwarz over fc, we conclude that, with C uniform over g > 6, 

\\u\\lhio,t]m) <CqHl + Wuh^mY^' . (65) 



Suppose now that u satisfies (|59|) on a time interval [0,T], where u{to) e H^{n). 
For suiBciently regular solutions u, we have the conservation laws 


\dMt)\l + G{\u{t)\^)^ / \dMto)\l + G{\u{to)n 


In particular, since — C < G{r) < C{r)i, it follows that ||u||l°°([o,t].hi) ^ 1 + 
||M(to)||j?i, uniformly in T. 

In the following proof, we assume a priori that u £ L°°H^ and prove uniqueness 
of such solutions. The existence of bounded energy solutions, and energy conserva- 
tion, is then proved by a weak- limit argument. 

Theorem 6.3. For each data f G H^{fl), and all T > 0, there exists a unique 
solution u to the equation (j59p . subject to the condition u G -L°°([0, T], i/^(il)). 
Furthermore, the solution satisfies the conservation laws (|66p . 

Proof. We start with the uniqueness of solutions. Since u G L°° ([0,T], H^) it 
follows by Sobolev embedding that F{u) G L°°L^ so w G C{[0,T], L'^), and by 
interpolation u G C{[0,T], H^) for all s < 1. Repeating this argument shows that 
the term in parentheses in (jSQ]) belongs to C^([0, T], L^). 

Let u and v be two solutions to (I59p . with u(0) = v{0). By unitarity of exp(ztA), 


= 2Im {F{u(t)) ~ F{v{t)),u{t) - v{t)) 

<c [ {{uit)r/' + {vitr/')\u{t)-v{t)\^ 

< C{1 + ||u(i)|L2,/5 + \\v{t)\\L.,,.)^\\u{t) - v{t)\\l,^, 
provided q > 5/2. Since \\u{t) — v{t)\\i^6 < C, we may interpolate to bound 

< c\\u{t) ~ vit)\\l',^ . 

Setting g{t) — \\u{t) — v{t)\\'j^2, and noting g{0) = 0, we have upon integrating that 

9{r)^<- f (\W{t)\\%^/, + \\vit)\\%^Adt+ — . 
Q Jo ^ ' 1 

By Holder's inequality and for r G [0,T] 

\<i)\\l2,/^dt < TH|w|li2([o,T],L2<!/5) ^ Cr^q. 


which goes to as g — >■ cx), provided t is small depending on C. Repeating the 
argument yields uniqueness on [0,T]. 

To establish existence and energy conservation for (|59|) with data, we let 
Gj{r) be a family of smooth, compactly supported real valued functions on [0,00), 



uniformly bounded below, such that Gj{r) and G'j{r) converge uniformly on com- 
pact sets to G{r) and G'{r). Additionally, we require that ([57| holds uniformly 
over j for G — Gj. 

We fix a time to and initial data u(to) G H^{i^), and let Uj{t) solve (|59p where F 
is replaced by Fj — Gj{\u\'^)u. We assume for the moment that uj exists globally in 
time, and satisfies the conservation law ()66|) . with G replaced by Gj. In particular 

||"illL°°Hi <G{l + \\u{ta)\\jfi) uniformly over j . 

By jSg), exp(-i(t - to)A)uj{t) is uniformly bounded in C^L^ n L^^ifi C C^iJ^, 
hence by the theorems of Rellich and Arzela-Ascoli, some subsequence of Uj con- 
verges uniformly in the norm on each finite time interval to u{t), in the sense 

lim ||uj(„) - u\\c(\-T,T],L'^) = for aU T < cx) . 

It follows that u E L°°H^, and thus by interpolation that for all s < 1, 

lim ||uj(„) - u\\cn-T.T],H') = fo^' aU T < oo . 

By Sobolev embedding we deduce that Fj{uj) — > F{u) in C([— T, T], L^), hence m is 
the solution to ([5^. unique by above. The conservation of mass in follows by 
uniform convergence in the norm and conservation of mass for uj . To conclude, 
we observe that by energy conservation for Uj and Fatou's lemma, for each ti we 

/ \dMti)\l + G{\u{h)\^)< f \dMto)\l + G{\u{to)\^). (67) 
Jn Jn 
On the other hand, u is the unique solution with data u{ti) at time ti, and the 
inequality is thus symmetric under exchange of to and ti. 

It remains to prove existence of energy conserving solutions to (|59p for 
data, in case G(r) G C^(R). For convenience set to = 0. We introduce w{t) = 
exp(— itA)u(i), and write ([59l) as 

w{t) ^uo-i f e-''^F{e'''^w{s)) ds . (68) 

Since F{z) = G'{\z\'^)z e C^{C), the map u F{u) is globally Lipschitz on 
L^(r2), and one has existence, uniqueness, and Lipschitz dependence on initial data 
for C^L^ solutions of given by the limit of Wn{t), where wo{t) = uq, and 

Wn+iit) ^uo-i f e-'^^F(e'^^u;„(s)) ds . (69) 

Convergence of Wn to w is uniform in the norm on any compact interval. From 
unitarity of exp(itA) on H'' (with norm defined spectrally), and the bound 

\\F{wis))\\Hi <K\\w{s)\\hi, 

one sees from (|69|) and weak limits, and using (|68p to express w'{t), that 

\\w{t)\\Hi < \\uo\\h^ exp{Kt) , \\w'{t)\\Hi < K\\uo\\h^ exp{Kt) . (70) 

It remains to prove the conservation laws (j66p on an interval [0, T] , for a T depending 
only on ||Mo||/fi and F; uniqueness yields global conservation. To do this, we 
will prove for such a T that if uq e i?^, then w G C^([0, T], _ff^), hence u G 
C^([0,r],L2). Together this is sufficient regularity to see that §6^ holds on [0,r] 



for uq E H^. Density and Fatou's lemma yields mass conservation and (|67l) for 
data; uniqueness then yields ([66|). 
We start by noting that 

< \\uis)\\l,i.. + \\u{s)\\h2 < Ms)\\l2 + \\u{s)\\h2 . 

Iterating (IM)) yields ||M||ioo([o j^/j ^^2) < 2||mo||j?2 for some T' > depending on 
IImqII jy2 . It sufhees then to prove, for some C and T depending only on ||uo||_f/i , that 
if T' < T and \\u\\L^(p,T'],m) < oo, then ||m|1l^([o.t'],h2) < C\\uo\\h2. Theorem 
Oand ^ yield 

II"IIl4([o,T'],wi>4) - \Wo\\h^/2 + \\F{u{s))\\h,/2 ds^ 

<\\uq\\h^\\uq\\h2+ I \\F{u{s))\\HA\F{u{s))\\H2ds, 

and we can also use ([59]) to bound ||w||l;^,//2 < ||wo||//2 + \\F{u)\\i^i^^h^. By the 
bounds ([TU]) . we combine these estimates, assuming T' < T < 1, to yield 

II"IIl»h2 + ||F(u)|L^,^2 < C\\uo\\h- + CT^\\u\\l^,h- + CTi\\Fiu)\\L2^^H- , 

where C < ||uo||hi- Taking T small yields the desired result. □ 

We conclude by noting that the above argument shows that u G C{[0,T],H^) 
for all finite T if mq S H^, but possibly with exponential growth of the norm, 
with the growth constant depending on ||Mo||iji- 

Acknowledgements. The authors would like to thank the referee for suggesting 
the application of our methods to semilinear Schrodinger equations on compact 
manifolds in Section [S] 


[1] Anton, R.: Global existence for defocusing cubic NLS and Gross-Pitaevskii equations in 

exterior domains. J. Math. Pures Appl. 89(4), 335-354 (2008) 
[2] Blair, M.D., Smith, H.F., and Sogge, CD.: On Strichartz Estimates for Schrodinger Opera- 
tors in Compact Manifolds with Boundary. Proc. Amer. Math. Soc. 136, 247-256 (2008) 
[3] Blair, M.D., Smith, H.F., and Sogge, CD.: Strichartz estimates for the wave equation on 

manifolds with boundary. Ann. Inst. H. Poincare Anal. Non Lineaire 26(5), 1817-1829 (2009) 
[4] Burq, N., Gerard, P., and Tzvetkov, N.: On nonlinear Schrodinger equations in exterior 

domains. Ann. Inst. H. Poincare Anal. Non Lineaire 21(3), 295-318 (2004) 
[5] Burq, N., Gerard, P., and Tzvetkov, N.: Strichartz inequalities and the nonlinear Schrodinger 

equation on compact manifolds. Amer. J. Math. 126, 569-605 (2004) 
[6] Constantin, P. and Saut, J.C: Local smoothing properties of dispersive equations. J. Amer. 

Math Soc. 1, 413-439 (1988) 
[7] Davies, E.B.: Heat Kernels and Spectral Theory. Cambridge University Press, Cambridge, 


[8] Ginibre, J. and Velo, G.: On the global Cauchy problem for some nonlinear Schrodinger 

equations. Ann. Inst. H. Poincare Anal. Non Lineaire 1(4), 309-323 (1984) 
[9] Herr, S.: The quintic nonlinear Schrodinger equation on three-dimensional ZoU manifolds. 
To appear in Amer. J. Math. 
[10] Herr, S., Tataru, D., and Tzvetkov, N.: Global well-posedness of the energy critical Nonlinear 
Schrodinger equation with small initial data in (T^). Duke Math. J. 159(2), 329-349 (2011) 
[11] lonescu, A.D. and Pausader, B.: The energy-critical defocusing NLS on T^. To appear in 
Duke Math. J. 

[12] Ivanovici, O.: On the Schrodinger equation outside strictly convex obstacles. Anal. PDE 3(3), 
261-293 (2010) 



[13] Ivanovici, O. and Planchon, F.: On the energy critical Schrodinger equation in 3D non- 
trapping domains. Ann. Inst. H. Poincarc Anal. Non Liiicaire 27(5), 1153-1177 (2010) 

[14] Journc, J.L., Soffer, A., and Sogge, CD.; Decay estimates for Schrodinger operators. Comm. 
Pure Appl. Math. 44(5), 573-604 (1991) 

[15] Keel, M. and Tao, T.: Endpoint Strichartz Estimates. Amer. J. Math. 120, 955-980 (1998) 

[16] Koch, H. and Tataru, D.: Dispersive estimates for principally normal operators. Comm. Pure 
Appl. Math. 58, 217-284 (2005) 

[17] Planchon, F. and Vega, L.: Bilinear Virial Identities and Applications. Ann. Sci. Ec. Norm. 
Super. (4) 42(2), 261-290 (2009) 

[18] Sjolin, P.; Regularity of solutions to Schrodinger equations. Duke Alath J. 55, 699-715 (1987) 

[19] Staffilani, G. and Tataru, D.: Strichartz estimates for a Schrodinger operator with nonsmooth 
coefficients. Comm. Partial Differential Equations 27(7-8), 1337-1372 (2002) 

[20] Smith, H.F. and Sogge, CD.; On the norm of spectral clusters for compa<;t manifolds 
with boundary. Acta Math. 198, 107-153 (2007) 

[21] Strichartz, R.: Restriction of Fourier transform to quadratic surfa<;es and decay of solutions 
to the wave equation. Duke Math. J. 44(3), 705-714 (1977) 

[22] Taylor, M.: Pseudodifferential Operators and Nonlinear PDE. Progress in Mathematics, vol. 
100, Birkhauser, Boston, 1991. 

[23] Vega, L.: Schrodinger equations: pointwise convergence to the initial data. Proc. Amer. Math 
Soc 102, 874-878 (1988)