Global Regularity for General Non-Linear WaveEquations I (6 + 1) and Higher Dimensions
Jacob Sterbenz
Department of Mathematics Princeton University Princeton
New Jersey USA
ABSTRACT
Following work of Tataru [Tataru D (1998) Local and global results for wavemaps I Comm Partial Differential Equations 23(9ndash10)1781ndash1793 Tataru D(1999) On the equation ampu frac14 jHuj2 in 5thorn 1 dimensions Math Res Lett 6
(5ndash6)469ndash485] we solve the division problem for wave equations with genericquadratic non-linearities in high dimensions Specifically we show that non-linearwave equations which can be written as systems involving equations of the form
ampf frac14 fHf andampf frac14 jHfj2 are well-posed with scattering in eth6thorn 1THORN and higherdimensions if the Cauchy data are small in the scale invariant lsquo1 Besov space _BBsc1This paper is the first in a series of works where we discuss the global regularity
properties of general non-linear wave equations for all dimensions 4 n
Key Words Semi-linear wave equation Strichartz estimates
1991 Mathematics Subject Classification 35L70
Correspondence Jacob Sterbenz Department of Mathematics Princeton UniversityPrinceton NJ 08544 USA E-mail sterbenzmathprincetonedu
COMMUNICATIONS IN PARTIAL DIFFERENTIAL EQUATIONS
Vol 29 Nos 9 amp 10 pp 1505ndash1531 2004
1505
DOI 101081PDE-200037764 0360-5302 (Print) 1532-4133 (Online)
Copyright 2004 by Marcel Dekker Inc wwwdekkercom
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
1 INTRODUCTION
In this paper our aim is to give a more or less complete description of the globalregularity properties of generic homogeneous quadratic semi-linear wave equationson eth6thorn 1THORN and higher dimensional Minkowski space The equations we will considerare all of the form
ampf frac14 NethfDfTHORN eth1THORN
Here amp frac14 2t thorn Dx denotes the standard wave operator on Rnthorn1 and N is a
smooth function of f and its first partial derivatives which we denote by Df Forall of the nonlinearities we study here N will be assumed to be at least quadraticin nature that is
NethX Y THORN frac14 OethjethX Y THORNj2THORN ethX Y THORN 0
The homogeneity condition we require N to satisfy is that there exist a (vector) ssuch that
Nethlsf lsthorn1DfTHORN frac14 lsthorn2NethfDfTHORN eth2THORN
where we use multiindex notation for vector N The condition (2) implies thatsolutions to the system (1) are invariant (again solutions) if one performs the scaletransformations
fethTHORNff lsfethlTHORN eth3THORN
The general class of equations which falls under this description contains virtually allmassless non-linear field theories on Minkowski space including the Yang Millsequations (YM) the wavendashmaps equations (WM) and the MaxwellndashDirac equations(MD) We list the schematics for these systems respectively as
ampA frac14 ADAthorn A3 ethYMTHORNampf frac14 jDfj2 ethWMTHORNampu frac14 ADu ethMDTHORNampA frac14 jDuj2
The various values of s for these equations are (respectively) s frac14 1 s frac14 0 ands frac14 eth12 1THORN For a more complete introduction to these equations see for instancethe works Foschi and Klainerman (2000) and Bournaveas (1996) For the purposesof this paper will will only be concerned with the structure of these equations at thelevel of the generic schematics (YM)ndash(MD)
The central problem we will be concerned with is that of giving a precisedescription of the regularity assumptions needed in order to guarantee that the
1506 Sterbenz
ORDER REPRINTS
Cauchy problem for the system (1) is globally well posed with scattering (GWPS)That is given initial data
feth0THORN frac14 f tfeth0THORN frac14 g eth4THORN
we wish to describe how much smoothness and decay ethf gTHORN needs to possess in orderfor there to exist a unique global solution to the system (1) with this given initialdata We also wish to show that the solutions we construct depend continuouslyon the initial data and are asymptotic to solutions of the linear part of (1) We willdescribe shortly in what sense we will require these notions to hold
Our main motivation here is to be able to prove global well-posedness for non-linear wave equations of the form (1) in a context where the initial data may not bevery smooth and furthermore does not possess enough decay at space-like infinityto be in L2 Also we would like to understand how this can be done in situationswhere the equations being considered contain no special structure in the non-linearity For instance this is of interest in discussing the problem of small dataglobal well-posedness for the MaxwellndashKleinndashGordon and YangndashMills equationswith the Lorentz gauge enforced instead of the more regular Coulomb gauge Thisprovides a significant point of departure from earlier works on the global existencetheory of non-linear wave equations which for the general case requires precisecontrol on the initial data in certain weighted Sobolev spaces (see Klainerman 1985)or else requires the non-linearity to have some specific algebraic structure (perhapscoming from a gauge transformation) which allows one to exploit lsquolsquonull formrsquorsquoidentities or apply standard Strichartz estimates directly to the equation beingconsidered (see Tao 2001 Tataru 1998)
From the point of view of homogeneity we are lead directly to considerations ofthe low regularity properties of Eq (1) as follows By a simple scaling argumenta onecan see that the most efficient L2 based regularity assumption possible on the initialdata involves sc frac14 n
2 s derivatives Again by scale invariance and looking at unitfrequency initial datab one can see that if we are to impose only an L2 smallness con-dition on the initial data which contains no physical space weights then sc frac14 n
2 s isin fact the largest amount of derivatives we may work with This leads us to considerthe question of GWPS for initial data in the homogeneous Besov spaces _BBscp forvarious values of p In this work we will concentrate solely on the case p frac14 1 Thisis the strongest scale and translation invariant control on the initial data possibleand will be crucial for the kind of non-linearities we work with here In fact it doesnot seem possible to push any type of global regularity for equations of the type (1)which contain derivatives in the non-linearity down to the scale invariant Sobolevspace _HHsc frac14 _BBsc2 unless the equations under consideration possess a great deal ofspecial structure in the non-linearity This has been done for the wave-maps
aIn conjunction with finite time blowup for large data This phenomena is known to happenfor higher dimensional equations with derivative non-linearities even in the presence ofpositive conserved quantities (see eg Cazenave et al 1998)bThat is initial data sets where the Fourier transform is supported in the unit frequency
annulus fx 12 lt jxj lt 2g
Global Regularity for NLWE 1507
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
equations (see Tao 2001) and more recently for the MaxwellndashKleinndashGordonequations in eth6thorn 1THORN and higher dimensions with the help of the Coulomb gauge(see Rodnianski preprint) Both of these results depend crucially on the fact thatthe underlying gauge group of the equations is compact
In recent years there has been much progress in our understanding of the lowregularity local theory for general non-linear wave equations of the form (1) Inthe lower dimensional setting ie when n frac14 2 3 4 it is known from counterexam-ples of Lindblad (see Lindblad 1996) that there is ill posedness for initial data inthe Sobolev space Hs0 where s0 sc thorn 5n
4 Intimately connected with this phenom-ena is the failure of certain space-time estimates for the linear wave equation knownas Strichartz estimates Specifically one does not have anything close to an L2ethL4THORNestimate in these dimensions Such an estimate obviously plays a crucial role (viaDuhamelrsquos principle) in the quadratic theory However using the Strichartz esti-mates available in these dimensions along with Picard iteration in certain functionspaces one can show that the Lindblad counterexamples are sharp in that thereis local well-posedness for initial data in the spaces Hs when sc thorn 5n
4 lt s (see forexample Klainerman and Selberg 2002)
In the higher dimensional setting ie when the number of spatial dimensions isn frac14 5 or greater one does have access to Strichartz estimates at the level of L2ethL4THORN(see Keel and Tao 1998) and it is possible to push the local theory down to HscthornEwhere 0 lt E is arbitrary (see Tataru 1999)
In all dimensions the single most important factor which determines the localtheory as well as the range of validity for Strichartz estimates is the existence of freewaves which are highly concentrated along null directions in Minkowski spaceThese waves known as Knapp counterexamples resemble a single beam of lightwhich remains coherent for a long period of time before dispersing For a specialclass of non-linearities known as lsquolsquonull structuresrsquorsquo interactions between thesecoherent beams are effectively canceled and one gains an improvement in the localtheory of equations whose nonlinearities have this form (see for example Klainermanand Machedon 1993 Klainerman and Selberg 2002)
In both high and low dimensional settings the analysis of certain null structuresspecifically non-linearities containing the Q0 null fromc has led to the proof that thewavendashmaps model equationsd are well posed in the scale invariant lsquo1 Besov space _BB
n21
(see Tataru 1998 2001) While the proof of this result is quite simple for high dimen-sions it relies in an essential way on the structure of the Q0 null form In fact there isno direct way to extend the proof of this result to include the less regular nonlinea-rities of the form fHf or for that matter the Qij null formse which show up in theequations of gauge field theory However the high dimensional non-linear interac-tion of coherent waves is quite weak (eg giving the desired range of validity forStrichartz estimates) and one would expect that it is possible to prove local wellposedness for quadratic equations with initial data in the scale invariant lsquo1 Besov
cThis is defined by the equation Q0ethfcTHORN frac14 afacdNot the rough schematic we have listed here but rather equations of the formampf frac14 GethfTHORNQ0ethffTHORNeThese are defined by QijethfcTHORN frac14 ifjc jfic
1508 Sterbenz
ORDER REPRINTS
space without resorting to any additional structure in the nonlinearity For n frac14 5dimensions it may be that this is not quite possible although we provide no convin-cing evidence except for the fact that there is no obvious way to add over our loca-lized estimates in that dimension in order to obtain a full set of estimates that worksin all of spacendashtime Fourier space In fact every estimate we prove here leads to alogarithmic divergence in the distance to the cone in Fourier space for the case ofeth5thorn 1THORN dimensions so in this sense our argument breaks down completely in thatregime However for n frac14 6 and higher dimensions we will prove that in fact no nullstructure is needed for there to be well posedness in _BBsc1 This leads to the statementof our main result which is as follows
Theorem 11 (Global Well Posedness) Let 6 n be the number of spatial dimen-sions For any of the generic equations listed above YM WM or MD let ethf gTHORN be a(possibly vector valued) initial data set Let sc frac14 n
2 s be the corresponding L2
scaling exponent Then there exists constants 0 lt E0C such that if
kethf gTHORNk _BBsc 1 _BBsc11 E0 eth5THORN
there exits a global solution c which satisfies the continuity condition
kckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN Ckethf gTHORNk _BBsc 1 _BBsc11 eth6THORN
The solution c is unique in the following sense There exists a sequence of smoothfunctions ethfN gN THORN such that
limN1
kethf gTHORN ethfN gN THORNk _BBsc 1 _BBsc11 frac14 0
For this sequence of functions there exists a sequence of unique smooth global solu-tions cN of (1) with this initial data Furthermore the cN converge to c as follows
limN1
kc cNkCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0
Also c is the only solution which may be obtained as a limit (in the above sense) ofsolutions to (1) with regularizations of ethf gTHORN as initial data Finally c retains anyextra smoothness inherent in the initial data That is if ethf gTHORN also has finite_HHs _HHs1 norm for sc lt s then so does c at fixed time and one has the followingestimate
kckCeth _HHsTHORNCeth1THORNeth _HHs1THORN Ckethf gTHORNk _HHs _HHs1 eth7THORN
In a straightforward way the function spaces we iterate in allow us to show thefollowing scattering result
Theorem 12 Using the same notation as above we have that there exists data setsethf gTHORN such that if c is the solution to the homogeneous wave equation with the
Global Regularity for NLWE 1509
ORDER REPRINTS
corresponding initial data the following asymptotics hold
limt1kcthorn ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth8THORN
limt1kc ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth9THORN
Furthermore the scattering operator retains any additional regularity inherent inthe initial data That is if ethf gTHORN has finite _HHs norm then so does ethf gTHORN andthe following asymptotics hold
limt1kcthorn ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth10THORN
limt1kc ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth11THORN
2 PRELIMINARY NOTATION
For quantities A and B we denote by AB to mean that A C B for somelarge constant C The constant C may change from line to line but will alwaysremain fixed for any given instance where this notation appears Likewise we usethe notation A B to mean that 1
C B A C B We also use the notation
A B to mean that A 1C B for some large constant C This is the notation we will
use throughout the paper to break down quantities into the standard cases A Bor A B or B A and AB or B A without ever discussing which constantswe are using
For a given function of two variables etht xTHORN 2 R R3 we write the spatial andspace-time Fourier transform as
ffetht xTHORN frac14Z
e2pixxfetht xTHORNdx
~ffetht xTHORN frac14Z
e2piethttthornxxTHORN fetht xTHORNdt dx
respectively At times we will also write Ffrac12f frac14 ~ff For a given set of functions of the spatial variable only we denote byWethf gTHORN the
solution of the homogeneous wave equation with Cauchy data ethf gTHORN If F is afunction on spacendashtime we will denote by WethFTHORN the function W Feth0THORN tFeth0THORNeth THORN
Let E denote any fundamental solution to the homogeneous wave equation ieone has the formula ampE frac14 d We define the standard Cauchy parametrix for thewave equation by the formula
amp1F frac14 E F WethE FTHORN
1510 Sterbenz
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
1 INTRODUCTION
In this paper our aim is to give a more or less complete description of the globalregularity properties of generic homogeneous quadratic semi-linear wave equationson eth6thorn 1THORN and higher dimensional Minkowski space The equations we will considerare all of the form
ampf frac14 NethfDfTHORN eth1THORN
Here amp frac14 2t thorn Dx denotes the standard wave operator on Rnthorn1 and N is a
smooth function of f and its first partial derivatives which we denote by Df Forall of the nonlinearities we study here N will be assumed to be at least quadraticin nature that is
NethX Y THORN frac14 OethjethX Y THORNj2THORN ethX Y THORN 0
The homogeneity condition we require N to satisfy is that there exist a (vector) ssuch that
Nethlsf lsthorn1DfTHORN frac14 lsthorn2NethfDfTHORN eth2THORN
where we use multiindex notation for vector N The condition (2) implies thatsolutions to the system (1) are invariant (again solutions) if one performs the scaletransformations
fethTHORNff lsfethlTHORN eth3THORN
The general class of equations which falls under this description contains virtually allmassless non-linear field theories on Minkowski space including the Yang Millsequations (YM) the wavendashmaps equations (WM) and the MaxwellndashDirac equations(MD) We list the schematics for these systems respectively as
ampA frac14 ADAthorn A3 ethYMTHORNampf frac14 jDfj2 ethWMTHORNampu frac14 ADu ethMDTHORNampA frac14 jDuj2
The various values of s for these equations are (respectively) s frac14 1 s frac14 0 ands frac14 eth12 1THORN For a more complete introduction to these equations see for instancethe works Foschi and Klainerman (2000) and Bournaveas (1996) For the purposesof this paper will will only be concerned with the structure of these equations at thelevel of the generic schematics (YM)ndash(MD)
The central problem we will be concerned with is that of giving a precisedescription of the regularity assumptions needed in order to guarantee that the
1506 Sterbenz
ORDER REPRINTS
Cauchy problem for the system (1) is globally well posed with scattering (GWPS)That is given initial data
feth0THORN frac14 f tfeth0THORN frac14 g eth4THORN
we wish to describe how much smoothness and decay ethf gTHORN needs to possess in orderfor there to exist a unique global solution to the system (1) with this given initialdata We also wish to show that the solutions we construct depend continuouslyon the initial data and are asymptotic to solutions of the linear part of (1) We willdescribe shortly in what sense we will require these notions to hold
Our main motivation here is to be able to prove global well-posedness for non-linear wave equations of the form (1) in a context where the initial data may not bevery smooth and furthermore does not possess enough decay at space-like infinityto be in L2 Also we would like to understand how this can be done in situationswhere the equations being considered contain no special structure in the non-linearity For instance this is of interest in discussing the problem of small dataglobal well-posedness for the MaxwellndashKleinndashGordon and YangndashMills equationswith the Lorentz gauge enforced instead of the more regular Coulomb gauge Thisprovides a significant point of departure from earlier works on the global existencetheory of non-linear wave equations which for the general case requires precisecontrol on the initial data in certain weighted Sobolev spaces (see Klainerman 1985)or else requires the non-linearity to have some specific algebraic structure (perhapscoming from a gauge transformation) which allows one to exploit lsquolsquonull formrsquorsquoidentities or apply standard Strichartz estimates directly to the equation beingconsidered (see Tao 2001 Tataru 1998)
From the point of view of homogeneity we are lead directly to considerations ofthe low regularity properties of Eq (1) as follows By a simple scaling argumenta onecan see that the most efficient L2 based regularity assumption possible on the initialdata involves sc frac14 n
2 s derivatives Again by scale invariance and looking at unitfrequency initial datab one can see that if we are to impose only an L2 smallness con-dition on the initial data which contains no physical space weights then sc frac14 n
2 s isin fact the largest amount of derivatives we may work with This leads us to considerthe question of GWPS for initial data in the homogeneous Besov spaces _BBscp forvarious values of p In this work we will concentrate solely on the case p frac14 1 Thisis the strongest scale and translation invariant control on the initial data possibleand will be crucial for the kind of non-linearities we work with here In fact it doesnot seem possible to push any type of global regularity for equations of the type (1)which contain derivatives in the non-linearity down to the scale invariant Sobolevspace _HHsc frac14 _BBsc2 unless the equations under consideration possess a great deal ofspecial structure in the non-linearity This has been done for the wave-maps
aIn conjunction with finite time blowup for large data This phenomena is known to happenfor higher dimensional equations with derivative non-linearities even in the presence ofpositive conserved quantities (see eg Cazenave et al 1998)bThat is initial data sets where the Fourier transform is supported in the unit frequency
annulus fx 12 lt jxj lt 2g
Global Regularity for NLWE 1507
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
equations (see Tao 2001) and more recently for the MaxwellndashKleinndashGordonequations in eth6thorn 1THORN and higher dimensions with the help of the Coulomb gauge(see Rodnianski preprint) Both of these results depend crucially on the fact thatthe underlying gauge group of the equations is compact
In recent years there has been much progress in our understanding of the lowregularity local theory for general non-linear wave equations of the form (1) Inthe lower dimensional setting ie when n frac14 2 3 4 it is known from counterexam-ples of Lindblad (see Lindblad 1996) that there is ill posedness for initial data inthe Sobolev space Hs0 where s0 sc thorn 5n
4 Intimately connected with this phenom-ena is the failure of certain space-time estimates for the linear wave equation knownas Strichartz estimates Specifically one does not have anything close to an L2ethL4THORNestimate in these dimensions Such an estimate obviously plays a crucial role (viaDuhamelrsquos principle) in the quadratic theory However using the Strichartz esti-mates available in these dimensions along with Picard iteration in certain functionspaces one can show that the Lindblad counterexamples are sharp in that thereis local well-posedness for initial data in the spaces Hs when sc thorn 5n
4 lt s (see forexample Klainerman and Selberg 2002)
In the higher dimensional setting ie when the number of spatial dimensions isn frac14 5 or greater one does have access to Strichartz estimates at the level of L2ethL4THORN(see Keel and Tao 1998) and it is possible to push the local theory down to HscthornEwhere 0 lt E is arbitrary (see Tataru 1999)
In all dimensions the single most important factor which determines the localtheory as well as the range of validity for Strichartz estimates is the existence of freewaves which are highly concentrated along null directions in Minkowski spaceThese waves known as Knapp counterexamples resemble a single beam of lightwhich remains coherent for a long period of time before dispersing For a specialclass of non-linearities known as lsquolsquonull structuresrsquorsquo interactions between thesecoherent beams are effectively canceled and one gains an improvement in the localtheory of equations whose nonlinearities have this form (see for example Klainermanand Machedon 1993 Klainerman and Selberg 2002)
In both high and low dimensional settings the analysis of certain null structuresspecifically non-linearities containing the Q0 null fromc has led to the proof that thewavendashmaps model equationsd are well posed in the scale invariant lsquo1 Besov space _BB
n21
(see Tataru 1998 2001) While the proof of this result is quite simple for high dimen-sions it relies in an essential way on the structure of the Q0 null form In fact there isno direct way to extend the proof of this result to include the less regular nonlinea-rities of the form fHf or for that matter the Qij null formse which show up in theequations of gauge field theory However the high dimensional non-linear interac-tion of coherent waves is quite weak (eg giving the desired range of validity forStrichartz estimates) and one would expect that it is possible to prove local wellposedness for quadratic equations with initial data in the scale invariant lsquo1 Besov
cThis is defined by the equation Q0ethfcTHORN frac14 afacdNot the rough schematic we have listed here but rather equations of the formampf frac14 GethfTHORNQ0ethffTHORNeThese are defined by QijethfcTHORN frac14 ifjc jfic
1508 Sterbenz
ORDER REPRINTS
space without resorting to any additional structure in the nonlinearity For n frac14 5dimensions it may be that this is not quite possible although we provide no convin-cing evidence except for the fact that there is no obvious way to add over our loca-lized estimates in that dimension in order to obtain a full set of estimates that worksin all of spacendashtime Fourier space In fact every estimate we prove here leads to alogarithmic divergence in the distance to the cone in Fourier space for the case ofeth5thorn 1THORN dimensions so in this sense our argument breaks down completely in thatregime However for n frac14 6 and higher dimensions we will prove that in fact no nullstructure is needed for there to be well posedness in _BBsc1 This leads to the statementof our main result which is as follows
Theorem 11 (Global Well Posedness) Let 6 n be the number of spatial dimen-sions For any of the generic equations listed above YM WM or MD let ethf gTHORN be a(possibly vector valued) initial data set Let sc frac14 n
2 s be the corresponding L2
scaling exponent Then there exists constants 0 lt E0C such that if
kethf gTHORNk _BBsc 1 _BBsc11 E0 eth5THORN
there exits a global solution c which satisfies the continuity condition
kckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN Ckethf gTHORNk _BBsc 1 _BBsc11 eth6THORN
The solution c is unique in the following sense There exists a sequence of smoothfunctions ethfN gN THORN such that
limN1
kethf gTHORN ethfN gN THORNk _BBsc 1 _BBsc11 frac14 0
For this sequence of functions there exists a sequence of unique smooth global solu-tions cN of (1) with this initial data Furthermore the cN converge to c as follows
limN1
kc cNkCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0
Also c is the only solution which may be obtained as a limit (in the above sense) ofsolutions to (1) with regularizations of ethf gTHORN as initial data Finally c retains anyextra smoothness inherent in the initial data That is if ethf gTHORN also has finite_HHs _HHs1 norm for sc lt s then so does c at fixed time and one has the followingestimate
kckCeth _HHsTHORNCeth1THORNeth _HHs1THORN Ckethf gTHORNk _HHs _HHs1 eth7THORN
In a straightforward way the function spaces we iterate in allow us to show thefollowing scattering result
Theorem 12 Using the same notation as above we have that there exists data setsethf gTHORN such that if c is the solution to the homogeneous wave equation with the
Global Regularity for NLWE 1509
ORDER REPRINTS
corresponding initial data the following asymptotics hold
limt1kcthorn ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth8THORN
limt1kc ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth9THORN
Furthermore the scattering operator retains any additional regularity inherent inthe initial data That is if ethf gTHORN has finite _HHs norm then so does ethf gTHORN andthe following asymptotics hold
limt1kcthorn ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth10THORN
limt1kc ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth11THORN
2 PRELIMINARY NOTATION
For quantities A and B we denote by AB to mean that A C B for somelarge constant C The constant C may change from line to line but will alwaysremain fixed for any given instance where this notation appears Likewise we usethe notation A B to mean that 1
C B A C B We also use the notation
A B to mean that A 1C B for some large constant C This is the notation we will
use throughout the paper to break down quantities into the standard cases A Bor A B or B A and AB or B A without ever discussing which constantswe are using
For a given function of two variables etht xTHORN 2 R R3 we write the spatial andspace-time Fourier transform as
ffetht xTHORN frac14Z
e2pixxfetht xTHORNdx
~ffetht xTHORN frac14Z
e2piethttthornxxTHORN fetht xTHORNdt dx
respectively At times we will also write Ffrac12f frac14 ~ff For a given set of functions of the spatial variable only we denote byWethf gTHORN the
solution of the homogeneous wave equation with Cauchy data ethf gTHORN If F is afunction on spacendashtime we will denote by WethFTHORN the function W Feth0THORN tFeth0THORNeth THORN
Let E denote any fundamental solution to the homogeneous wave equation ieone has the formula ampE frac14 d We define the standard Cauchy parametrix for thewave equation by the formula
amp1F frac14 E F WethE FTHORN
1510 Sterbenz
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Cauchy problem for the system (1) is globally well posed with scattering (GWPS)That is given initial data
feth0THORN frac14 f tfeth0THORN frac14 g eth4THORN
we wish to describe how much smoothness and decay ethf gTHORN needs to possess in orderfor there to exist a unique global solution to the system (1) with this given initialdata We also wish to show that the solutions we construct depend continuouslyon the initial data and are asymptotic to solutions of the linear part of (1) We willdescribe shortly in what sense we will require these notions to hold
Our main motivation here is to be able to prove global well-posedness for non-linear wave equations of the form (1) in a context where the initial data may not bevery smooth and furthermore does not possess enough decay at space-like infinityto be in L2 Also we would like to understand how this can be done in situationswhere the equations being considered contain no special structure in the non-linearity For instance this is of interest in discussing the problem of small dataglobal well-posedness for the MaxwellndashKleinndashGordon and YangndashMills equationswith the Lorentz gauge enforced instead of the more regular Coulomb gauge Thisprovides a significant point of departure from earlier works on the global existencetheory of non-linear wave equations which for the general case requires precisecontrol on the initial data in certain weighted Sobolev spaces (see Klainerman 1985)or else requires the non-linearity to have some specific algebraic structure (perhapscoming from a gauge transformation) which allows one to exploit lsquolsquonull formrsquorsquoidentities or apply standard Strichartz estimates directly to the equation beingconsidered (see Tao 2001 Tataru 1998)
From the point of view of homogeneity we are lead directly to considerations ofthe low regularity properties of Eq (1) as follows By a simple scaling argumenta onecan see that the most efficient L2 based regularity assumption possible on the initialdata involves sc frac14 n
2 s derivatives Again by scale invariance and looking at unitfrequency initial datab one can see that if we are to impose only an L2 smallness con-dition on the initial data which contains no physical space weights then sc frac14 n
2 s isin fact the largest amount of derivatives we may work with This leads us to considerthe question of GWPS for initial data in the homogeneous Besov spaces _BBscp forvarious values of p In this work we will concentrate solely on the case p frac14 1 Thisis the strongest scale and translation invariant control on the initial data possibleand will be crucial for the kind of non-linearities we work with here In fact it doesnot seem possible to push any type of global regularity for equations of the type (1)which contain derivatives in the non-linearity down to the scale invariant Sobolevspace _HHsc frac14 _BBsc2 unless the equations under consideration possess a great deal ofspecial structure in the non-linearity This has been done for the wave-maps
aIn conjunction with finite time blowup for large data This phenomena is known to happenfor higher dimensional equations with derivative non-linearities even in the presence ofpositive conserved quantities (see eg Cazenave et al 1998)bThat is initial data sets where the Fourier transform is supported in the unit frequency
annulus fx 12 lt jxj lt 2g
Global Regularity for NLWE 1507
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
equations (see Tao 2001) and more recently for the MaxwellndashKleinndashGordonequations in eth6thorn 1THORN and higher dimensions with the help of the Coulomb gauge(see Rodnianski preprint) Both of these results depend crucially on the fact thatthe underlying gauge group of the equations is compact
In recent years there has been much progress in our understanding of the lowregularity local theory for general non-linear wave equations of the form (1) Inthe lower dimensional setting ie when n frac14 2 3 4 it is known from counterexam-ples of Lindblad (see Lindblad 1996) that there is ill posedness for initial data inthe Sobolev space Hs0 where s0 sc thorn 5n
4 Intimately connected with this phenom-ena is the failure of certain space-time estimates for the linear wave equation knownas Strichartz estimates Specifically one does not have anything close to an L2ethL4THORNestimate in these dimensions Such an estimate obviously plays a crucial role (viaDuhamelrsquos principle) in the quadratic theory However using the Strichartz esti-mates available in these dimensions along with Picard iteration in certain functionspaces one can show that the Lindblad counterexamples are sharp in that thereis local well-posedness for initial data in the spaces Hs when sc thorn 5n
4 lt s (see forexample Klainerman and Selberg 2002)
In the higher dimensional setting ie when the number of spatial dimensions isn frac14 5 or greater one does have access to Strichartz estimates at the level of L2ethL4THORN(see Keel and Tao 1998) and it is possible to push the local theory down to HscthornEwhere 0 lt E is arbitrary (see Tataru 1999)
In all dimensions the single most important factor which determines the localtheory as well as the range of validity for Strichartz estimates is the existence of freewaves which are highly concentrated along null directions in Minkowski spaceThese waves known as Knapp counterexamples resemble a single beam of lightwhich remains coherent for a long period of time before dispersing For a specialclass of non-linearities known as lsquolsquonull structuresrsquorsquo interactions between thesecoherent beams are effectively canceled and one gains an improvement in the localtheory of equations whose nonlinearities have this form (see for example Klainermanand Machedon 1993 Klainerman and Selberg 2002)
In both high and low dimensional settings the analysis of certain null structuresspecifically non-linearities containing the Q0 null fromc has led to the proof that thewavendashmaps model equationsd are well posed in the scale invariant lsquo1 Besov space _BB
n21
(see Tataru 1998 2001) While the proof of this result is quite simple for high dimen-sions it relies in an essential way on the structure of the Q0 null form In fact there isno direct way to extend the proof of this result to include the less regular nonlinea-rities of the form fHf or for that matter the Qij null formse which show up in theequations of gauge field theory However the high dimensional non-linear interac-tion of coherent waves is quite weak (eg giving the desired range of validity forStrichartz estimates) and one would expect that it is possible to prove local wellposedness for quadratic equations with initial data in the scale invariant lsquo1 Besov
cThis is defined by the equation Q0ethfcTHORN frac14 afacdNot the rough schematic we have listed here but rather equations of the formampf frac14 GethfTHORNQ0ethffTHORNeThese are defined by QijethfcTHORN frac14 ifjc jfic
1508 Sterbenz
ORDER REPRINTS
space without resorting to any additional structure in the nonlinearity For n frac14 5dimensions it may be that this is not quite possible although we provide no convin-cing evidence except for the fact that there is no obvious way to add over our loca-lized estimates in that dimension in order to obtain a full set of estimates that worksin all of spacendashtime Fourier space In fact every estimate we prove here leads to alogarithmic divergence in the distance to the cone in Fourier space for the case ofeth5thorn 1THORN dimensions so in this sense our argument breaks down completely in thatregime However for n frac14 6 and higher dimensions we will prove that in fact no nullstructure is needed for there to be well posedness in _BBsc1 This leads to the statementof our main result which is as follows
Theorem 11 (Global Well Posedness) Let 6 n be the number of spatial dimen-sions For any of the generic equations listed above YM WM or MD let ethf gTHORN be a(possibly vector valued) initial data set Let sc frac14 n
2 s be the corresponding L2
scaling exponent Then there exists constants 0 lt E0C such that if
kethf gTHORNk _BBsc 1 _BBsc11 E0 eth5THORN
there exits a global solution c which satisfies the continuity condition
kckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN Ckethf gTHORNk _BBsc 1 _BBsc11 eth6THORN
The solution c is unique in the following sense There exists a sequence of smoothfunctions ethfN gN THORN such that
limN1
kethf gTHORN ethfN gN THORNk _BBsc 1 _BBsc11 frac14 0
For this sequence of functions there exists a sequence of unique smooth global solu-tions cN of (1) with this initial data Furthermore the cN converge to c as follows
limN1
kc cNkCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0
Also c is the only solution which may be obtained as a limit (in the above sense) ofsolutions to (1) with regularizations of ethf gTHORN as initial data Finally c retains anyextra smoothness inherent in the initial data That is if ethf gTHORN also has finite_HHs _HHs1 norm for sc lt s then so does c at fixed time and one has the followingestimate
kckCeth _HHsTHORNCeth1THORNeth _HHs1THORN Ckethf gTHORNk _HHs _HHs1 eth7THORN
In a straightforward way the function spaces we iterate in allow us to show thefollowing scattering result
Theorem 12 Using the same notation as above we have that there exists data setsethf gTHORN such that if c is the solution to the homogeneous wave equation with the
Global Regularity for NLWE 1509
ORDER REPRINTS
corresponding initial data the following asymptotics hold
limt1kcthorn ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth8THORN
limt1kc ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth9THORN
Furthermore the scattering operator retains any additional regularity inherent inthe initial data That is if ethf gTHORN has finite _HHs norm then so does ethf gTHORN andthe following asymptotics hold
limt1kcthorn ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth10THORN
limt1kc ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth11THORN
2 PRELIMINARY NOTATION
For quantities A and B we denote by AB to mean that A C B for somelarge constant C The constant C may change from line to line but will alwaysremain fixed for any given instance where this notation appears Likewise we usethe notation A B to mean that 1
C B A C B We also use the notation
A B to mean that A 1C B for some large constant C This is the notation we will
use throughout the paper to break down quantities into the standard cases A Bor A B or B A and AB or B A without ever discussing which constantswe are using
For a given function of two variables etht xTHORN 2 R R3 we write the spatial andspace-time Fourier transform as
ffetht xTHORN frac14Z
e2pixxfetht xTHORNdx
~ffetht xTHORN frac14Z
e2piethttthornxxTHORN fetht xTHORNdt dx
respectively At times we will also write Ffrac12f frac14 ~ff For a given set of functions of the spatial variable only we denote byWethf gTHORN the
solution of the homogeneous wave equation with Cauchy data ethf gTHORN If F is afunction on spacendashtime we will denote by WethFTHORN the function W Feth0THORN tFeth0THORNeth THORN
Let E denote any fundamental solution to the homogeneous wave equation ieone has the formula ampE frac14 d We define the standard Cauchy parametrix for thewave equation by the formula
amp1F frac14 E F WethE FTHORN
1510 Sterbenz
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
equations (see Tao 2001) and more recently for the MaxwellndashKleinndashGordonequations in eth6thorn 1THORN and higher dimensions with the help of the Coulomb gauge(see Rodnianski preprint) Both of these results depend crucially on the fact thatthe underlying gauge group of the equations is compact
In recent years there has been much progress in our understanding of the lowregularity local theory for general non-linear wave equations of the form (1) Inthe lower dimensional setting ie when n frac14 2 3 4 it is known from counterexam-ples of Lindblad (see Lindblad 1996) that there is ill posedness for initial data inthe Sobolev space Hs0 where s0 sc thorn 5n
4 Intimately connected with this phenom-ena is the failure of certain space-time estimates for the linear wave equation knownas Strichartz estimates Specifically one does not have anything close to an L2ethL4THORNestimate in these dimensions Such an estimate obviously plays a crucial role (viaDuhamelrsquos principle) in the quadratic theory However using the Strichartz esti-mates available in these dimensions along with Picard iteration in certain functionspaces one can show that the Lindblad counterexamples are sharp in that thereis local well-posedness for initial data in the spaces Hs when sc thorn 5n
4 lt s (see forexample Klainerman and Selberg 2002)
In the higher dimensional setting ie when the number of spatial dimensions isn frac14 5 or greater one does have access to Strichartz estimates at the level of L2ethL4THORN(see Keel and Tao 1998) and it is possible to push the local theory down to HscthornEwhere 0 lt E is arbitrary (see Tataru 1999)
In all dimensions the single most important factor which determines the localtheory as well as the range of validity for Strichartz estimates is the existence of freewaves which are highly concentrated along null directions in Minkowski spaceThese waves known as Knapp counterexamples resemble a single beam of lightwhich remains coherent for a long period of time before dispersing For a specialclass of non-linearities known as lsquolsquonull structuresrsquorsquo interactions between thesecoherent beams are effectively canceled and one gains an improvement in the localtheory of equations whose nonlinearities have this form (see for example Klainermanand Machedon 1993 Klainerman and Selberg 2002)
In both high and low dimensional settings the analysis of certain null structuresspecifically non-linearities containing the Q0 null fromc has led to the proof that thewavendashmaps model equationsd are well posed in the scale invariant lsquo1 Besov space _BB
n21
(see Tataru 1998 2001) While the proof of this result is quite simple for high dimen-sions it relies in an essential way on the structure of the Q0 null form In fact there isno direct way to extend the proof of this result to include the less regular nonlinea-rities of the form fHf or for that matter the Qij null formse which show up in theequations of gauge field theory However the high dimensional non-linear interac-tion of coherent waves is quite weak (eg giving the desired range of validity forStrichartz estimates) and one would expect that it is possible to prove local wellposedness for quadratic equations with initial data in the scale invariant lsquo1 Besov
cThis is defined by the equation Q0ethfcTHORN frac14 afacdNot the rough schematic we have listed here but rather equations of the formampf frac14 GethfTHORNQ0ethffTHORNeThese are defined by QijethfcTHORN frac14 ifjc jfic
1508 Sterbenz
ORDER REPRINTS
space without resorting to any additional structure in the nonlinearity For n frac14 5dimensions it may be that this is not quite possible although we provide no convin-cing evidence except for the fact that there is no obvious way to add over our loca-lized estimates in that dimension in order to obtain a full set of estimates that worksin all of spacendashtime Fourier space In fact every estimate we prove here leads to alogarithmic divergence in the distance to the cone in Fourier space for the case ofeth5thorn 1THORN dimensions so in this sense our argument breaks down completely in thatregime However for n frac14 6 and higher dimensions we will prove that in fact no nullstructure is needed for there to be well posedness in _BBsc1 This leads to the statementof our main result which is as follows
Theorem 11 (Global Well Posedness) Let 6 n be the number of spatial dimen-sions For any of the generic equations listed above YM WM or MD let ethf gTHORN be a(possibly vector valued) initial data set Let sc frac14 n
2 s be the corresponding L2
scaling exponent Then there exists constants 0 lt E0C such that if
kethf gTHORNk _BBsc 1 _BBsc11 E0 eth5THORN
there exits a global solution c which satisfies the continuity condition
kckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN Ckethf gTHORNk _BBsc 1 _BBsc11 eth6THORN
The solution c is unique in the following sense There exists a sequence of smoothfunctions ethfN gN THORN such that
limN1
kethf gTHORN ethfN gN THORNk _BBsc 1 _BBsc11 frac14 0
For this sequence of functions there exists a sequence of unique smooth global solu-tions cN of (1) with this initial data Furthermore the cN converge to c as follows
limN1
kc cNkCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0
Also c is the only solution which may be obtained as a limit (in the above sense) ofsolutions to (1) with regularizations of ethf gTHORN as initial data Finally c retains anyextra smoothness inherent in the initial data That is if ethf gTHORN also has finite_HHs _HHs1 norm for sc lt s then so does c at fixed time and one has the followingestimate
kckCeth _HHsTHORNCeth1THORNeth _HHs1THORN Ckethf gTHORNk _HHs _HHs1 eth7THORN
In a straightforward way the function spaces we iterate in allow us to show thefollowing scattering result
Theorem 12 Using the same notation as above we have that there exists data setsethf gTHORN such that if c is the solution to the homogeneous wave equation with the
Global Regularity for NLWE 1509
ORDER REPRINTS
corresponding initial data the following asymptotics hold
limt1kcthorn ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth8THORN
limt1kc ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth9THORN
Furthermore the scattering operator retains any additional regularity inherent inthe initial data That is if ethf gTHORN has finite _HHs norm then so does ethf gTHORN andthe following asymptotics hold
limt1kcthorn ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth10THORN
limt1kc ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth11THORN
2 PRELIMINARY NOTATION
For quantities A and B we denote by AB to mean that A C B for somelarge constant C The constant C may change from line to line but will alwaysremain fixed for any given instance where this notation appears Likewise we usethe notation A B to mean that 1
C B A C B We also use the notation
A B to mean that A 1C B for some large constant C This is the notation we will
use throughout the paper to break down quantities into the standard cases A Bor A B or B A and AB or B A without ever discussing which constantswe are using
For a given function of two variables etht xTHORN 2 R R3 we write the spatial andspace-time Fourier transform as
ffetht xTHORN frac14Z
e2pixxfetht xTHORNdx
~ffetht xTHORN frac14Z
e2piethttthornxxTHORN fetht xTHORNdt dx
respectively At times we will also write Ffrac12f frac14 ~ff For a given set of functions of the spatial variable only we denote byWethf gTHORN the
solution of the homogeneous wave equation with Cauchy data ethf gTHORN If F is afunction on spacendashtime we will denote by WethFTHORN the function W Feth0THORN tFeth0THORNeth THORN
Let E denote any fundamental solution to the homogeneous wave equation ieone has the formula ampE frac14 d We define the standard Cauchy parametrix for thewave equation by the formula
amp1F frac14 E F WethE FTHORN
1510 Sterbenz
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
space without resorting to any additional structure in the nonlinearity For n frac14 5dimensions it may be that this is not quite possible although we provide no convin-cing evidence except for the fact that there is no obvious way to add over our loca-lized estimates in that dimension in order to obtain a full set of estimates that worksin all of spacendashtime Fourier space In fact every estimate we prove here leads to alogarithmic divergence in the distance to the cone in Fourier space for the case ofeth5thorn 1THORN dimensions so in this sense our argument breaks down completely in thatregime However for n frac14 6 and higher dimensions we will prove that in fact no nullstructure is needed for there to be well posedness in _BBsc1 This leads to the statementof our main result which is as follows
Theorem 11 (Global Well Posedness) Let 6 n be the number of spatial dimen-sions For any of the generic equations listed above YM WM or MD let ethf gTHORN be a(possibly vector valued) initial data set Let sc frac14 n
2 s be the corresponding L2
scaling exponent Then there exists constants 0 lt E0C such that if
kethf gTHORNk _BBsc 1 _BBsc11 E0 eth5THORN
there exits a global solution c which satisfies the continuity condition
kckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN Ckethf gTHORNk _BBsc 1 _BBsc11 eth6THORN
The solution c is unique in the following sense There exists a sequence of smoothfunctions ethfN gN THORN such that
limN1
kethf gTHORN ethfN gN THORNk _BBsc 1 _BBsc11 frac14 0
For this sequence of functions there exists a sequence of unique smooth global solu-tions cN of (1) with this initial data Furthermore the cN converge to c as follows
limN1
kc cNkCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0
Also c is the only solution which may be obtained as a limit (in the above sense) ofsolutions to (1) with regularizations of ethf gTHORN as initial data Finally c retains anyextra smoothness inherent in the initial data That is if ethf gTHORN also has finite_HHs _HHs1 norm for sc lt s then so does c at fixed time and one has the followingestimate
kckCeth _HHsTHORNCeth1THORNeth _HHs1THORN Ckethf gTHORNk _HHs _HHs1 eth7THORN
In a straightforward way the function spaces we iterate in allow us to show thefollowing scattering result
Theorem 12 Using the same notation as above we have that there exists data setsethf gTHORN such that if c is the solution to the homogeneous wave equation with the
Global Regularity for NLWE 1509
ORDER REPRINTS
corresponding initial data the following asymptotics hold
limt1kcthorn ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth8THORN
limt1kc ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth9THORN
Furthermore the scattering operator retains any additional regularity inherent inthe initial data That is if ethf gTHORN has finite _HHs norm then so does ethf gTHORN andthe following asymptotics hold
limt1kcthorn ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth10THORN
limt1kc ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth11THORN
2 PRELIMINARY NOTATION
For quantities A and B we denote by AB to mean that A C B for somelarge constant C The constant C may change from line to line but will alwaysremain fixed for any given instance where this notation appears Likewise we usethe notation A B to mean that 1
C B A C B We also use the notation
A B to mean that A 1C B for some large constant C This is the notation we will
use throughout the paper to break down quantities into the standard cases A Bor A B or B A and AB or B A without ever discussing which constantswe are using
For a given function of two variables etht xTHORN 2 R R3 we write the spatial andspace-time Fourier transform as
ffetht xTHORN frac14Z
e2pixxfetht xTHORNdx
~ffetht xTHORN frac14Z
e2piethttthornxxTHORN fetht xTHORNdt dx
respectively At times we will also write Ffrac12f frac14 ~ff For a given set of functions of the spatial variable only we denote byWethf gTHORN the
solution of the homogeneous wave equation with Cauchy data ethf gTHORN If F is afunction on spacendashtime we will denote by WethFTHORN the function W Feth0THORN tFeth0THORNeth THORN
Let E denote any fundamental solution to the homogeneous wave equation ieone has the formula ampE frac14 d We define the standard Cauchy parametrix for thewave equation by the formula
amp1F frac14 E F WethE FTHORN
1510 Sterbenz
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
corresponding initial data the following asymptotics hold
limt1kcthorn ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth8THORN
limt1kc ckCeth _BBsc 1THORNCeth1THORNeth _BBsc11THORN frac14 0 eth9THORN
Furthermore the scattering operator retains any additional regularity inherent inthe initial data That is if ethf gTHORN has finite _HHs norm then so does ethf gTHORN andthe following asymptotics hold
limt1kcthorn ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth10THORN
limt1kc ckCeth _HHsTHORNCeth1THORNeth _HHs1THORN frac14 0 eth11THORN
2 PRELIMINARY NOTATION
For quantities A and B we denote by AB to mean that A C B for somelarge constant C The constant C may change from line to line but will alwaysremain fixed for any given instance where this notation appears Likewise we usethe notation A B to mean that 1
C B A C B We also use the notation
A B to mean that A 1C B for some large constant C This is the notation we will
use throughout the paper to break down quantities into the standard cases A Bor A B or B A and AB or B A without ever discussing which constantswe are using
For a given function of two variables etht xTHORN 2 R R3 we write the spatial andspace-time Fourier transform as
ffetht xTHORN frac14Z
e2pixxfetht xTHORNdx
~ffetht xTHORN frac14Z
e2piethttthornxxTHORN fetht xTHORNdt dx
respectively At times we will also write Ffrac12f frac14 ~ff For a given set of functions of the spatial variable only we denote byWethf gTHORN the
solution of the homogeneous wave equation with Cauchy data ethf gTHORN If F is afunction on spacendashtime we will denote by WethFTHORN the function W Feth0THORN tFeth0THORNeth THORN
Let E denote any fundamental solution to the homogeneous wave equation ieone has the formula ampE frac14 d We define the standard Cauchy parametrix for thewave equation by the formula
amp1F frac14 E F WethE FTHORN
1510 Sterbenz
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Explicitly one has the identity
damp1Famp1Fetht xTHORN frac14 Z t
0
sineth2pjxjetht sTHORNTHORN2pjxj
bFFeths xTHORNds eth12THORN
For any function F which is supported away from the light cone in Fourierspace we shall use the following notation for division by the symbol of the waveequation
X1F frac14 E F
Of course the definition of X1 does not depend on E so long as F is supported awayfrom the light cone in Fourier space for us this will always be the case when thisnotation is in use Explicitly one has the formula
F X1F etht xTHORN frac14 1
4p2etht2 jxj2THORNeFFetht xTHORN
3 MULTIPLIERS AND FUNCTION SPACES
Let j be a smooth bump function (ie supported on the set jsj 2 such thatj frac14 1 for jsj 1) In what follows it will be a great convenience for us to assumethat j may change its exact form for two separate instances of the symbol j (evenif they occur on the same line) In this way we may assume without loss of generalitythat in addition to being smooth we also have the idempotence identity j2 frac14 j Weshall use this convention for all the cutoff functions we introduce in the sequel
For l 2 f2j j 2 Zg we denote the dyadic scaling of j by jlethsTHORN frac14 jethslTHORN Themost basic Fourier localizations we shall use here are with respect to the space-timevariable and the distance from the cone Accordingly for ld 2 f2j j 2 Zg we formthe LittlewoodndashPaley type cutoff functions
sletht xTHORN frac14 j2lethjetht xTHORNjTHORN j12lethjetht xTHORNjTHORN eth13THORN
cdetht xTHORN frac14 j2dethjtj jxjTHORN j12dethjtj jxjTHORN eth14THORN
We now denote the corresponding Fourier multiplier operator via the formulasfSluSlu frac14 sl~uu and gCduCdu frac14 cd~uu respectively We also use a multi-subscript notation todenote products of the above operators eg Sld frac14 SlCd We shall use the notation
Sld frac14Xdd
Sld eth15THORN
to denote cutoff in an OethdTHORN neighborhood of the light cone in Fourier space At timesit will also be convenient to write Sld frac14 Sl Slltd We shall also use the notationSld etc to denote the multiplier Sld cutoff in the half space t gt 0
Global Regularity for NLWE 1511
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
The other type of Fourier localization which will be central to our analysis is thedecomposition of the spatial variable into radially directed blocks of various sizesTo begin with we denote the spatial frequency cutoff by
plethxTHORN frac14 j2lethjxjTHORN j12lethjxjTHORN eth16THORN
with Pl the corresponding operator For a given parameter d l we now decom-pose Pl radially as follows First decompose the the unit sphere Sn1 Rn intoangular sectors of size d
l dl with bounded overlap (independent of d) These
angular sectors are then projected out to frequency l via rays through the originThe result is a decomposition of suppfplg into radially directed blocks of sizel d d with bounded overlap We enumerate these blocks and label thecorresponding partition of unity by bold It is clear that things may be arranged sothat upon rotation onto the x1-axis each bold satisfies the bound
jN1 b
oldj CNl
N jNi b
oldj CNd
N eth17THORN
In particular each Bold is given by convolution with an L1 kernel We shall also
denote
Sold frac14 Bo
lethldTHORN12Sld Sold frac14 Bo
lethldTHORN12Sld
Note that the operators Sold and Sold are only supported in the region wherejtj jxj
We now use these multipliers to define the following dyadic norms which will bethe building blocks for the function spaces we will use here
kukpX
12lp
frac14Xd22Z
dp
2kSldukpL2 (lsquolsquoclassicalrsquorsquo Hsd) eth18THORN
kukYl frac14 l1kampSlukL1ethL2THORN (Duhamel) eth19THORN
kukZlfrac14 l
2n2
Xd
Xo
kSolduk2L1ethL1THORN
12
(outer block) eth20THORN
Notice that the (semi) norms X12
lp and Yl are only well defined modulo measuressupported on the light cone in Fourier space Because of this it will be convenientfor us to include an extra L1ethL2THORN norm in the definition of our function spaces Thisrepresents the inclusion in the above norms of solutions to the wave equation withL2 initial data Adding everything together we are led to define the followingfixed frequency (semi) norms
Fl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN
eth21THORN
Unfortunately the above norm is alone not strong enough for us to be able to iterateequations of the form (1) which contain derivatives This is due to a very specific
1512 Sterbenz
4_LPDE29_09amp10_R3
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Low High frequency interaction in quadratic non-linearities Fortunately thisproblem has been effectively handled by Tataru (1999) based on ideas fromKlainerman and Machedon (1996) and Klainerman and Tataru (1999) What isnecessary is to add some extra L1ethL1THORN norms on lsquolsquoouter blockrsquorsquo regions of Fourierspace This is the essence of the norm (20) above which is a slight variant of thatwhich appeared in Tataru (1999) This leads to our second main dyadic norm
Gl frac14 X12
l1 thorn Yl
Sl L1ethL2THORN Zl eth22THORN
Finally the spaces we will iterate in are produced by adding the appropriate numberof derivatives combined with the necessary Besov structures
kuk2Fs frac14Xl
l2skuk2Fl eth23THORN
kukGs frac14Xl
lskukGl eth24THORN
Due to the need for precise microlocal decompositions of crucial importance tous will be the boundedness of certain multipliers on the components (18)ndash(19) of ourfunction spaces as well as mixed Lebesgue spaces We state these as follows
Lemma 31 (Multiplier Boundedness)
(1) The following multipliers are given by L1 kernels l1HSl Sold Sold and
ethldTHORNX1Sold In particular all of these are bounded on every mixedLebesgue space LqethLrTHORN
(2) The following multipliers are bounded on the spaces LqethL2THORN for1 q 1 Sld and Sld
Proof of Lemma 31 (1) First notice that after a rescaling the symbol for themultiplier l1HSl is a C1 bump function with Oeth1THORN support Thus its kernel is inL1 with norm independent of l
For the remainder of the operators listed in (1) above it suffices to work withethldTHORNX1Sold The boundedness of the others follows from a similar argument Welet w denote the symbol of this operator cut off in the upper resp lower half planeAfter a rotation in the spatial domain we may assume that the spatial projection ofw is directed along the positive x1 axis Now look at wthorneths ZTHORN with coordinates
s frac14 1ffiffiffi2
p etht x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0
It is apparent that wthorneths ZTHORN has support in a box of dimension lffiffiffiffiffiffild
p ffiffiffiffiffiffild
p d with sides parallel to the coordinate axis and longest side in
Global Regularity for NLWE 1513
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
the Z1 direction and shortest side in the s direction Furthermore a directioncalculation shows that one has the bounds
jNZ1wthornj CNl
N jNZ0 w
thornj CN ethldTHORNN=2 jNs w
thornj CNdN
Therefore we have that wthorn yields an L1 kernel A similar argument works for thecutoff function w using the rotation
s frac14 1ffiffiffi2
p ethtthorn x1THORN
Z1 frac14 1ffiffiffi2
p ethtthorn x1THORN
Z0 frac14 x0 amp
Proof of Lemma 31 eth2THORN We will argue here for Sld The estimates for the othersfollow similarly If we denote by Ketht xTHORN the convolution kernel associated with Sldthen a simple calculation shows that
e2pitjxjcKKetht xTHORN frac14Z
e2pittcetht xTHORNdt
where suppfcg is contained in a box of dimension l l d with sides alongthe coordinate axis and short side in the t direction Furthermore one has theestimate
jNt cj CNd
N
This shows that we have the bound
kcKKkL1t ethL1
x THORN 1
independent of l and d Thus we get the desired bounds for the convolutionkernels amp
As an immediate application of the above lemma we show that the extra Zl
intersection in the Gl norm above only effects the X12
l1 portion of things
Lemma 32 (Outer Block Estimate on Yl) For 5 lt n one has the following uniforminclusion
Yl 13 Zl eth25THORN
1514 Sterbenz
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Proof of (25) It is enough to show that
Xo
kX1Solduk2L1ethL1THORN
12
ln42
d
l
n54
kSlukL1ethL2THORN
First using a local Sobolev embedding we see that
kBo
lethldTHORN12X1SoldukL1ethL1THORN l
nthorn14 d
n14 kX1SoldukL1ethL2THORN
Therefore using the boundedness Lemma 31 it suffices to note that by Minkowskirsquosinequality we can bound
Xo
ZkSolduethtTHORNkL2
x
2
dt
0 1A12
Z X
o
kSolduethtTHORNk2L2x
12
dt
kSldukL1ethL2THORN
The last line of the above proof showed that it is possible to bound a squaresum over an angular decomposition of a given function in L1ethL2THORN It is also clear that
this same procedure works for the X12
l1 spaces because one can use Minkowskirsquosinequality for the lsquo1 sum with respect to the cone variable d This fact will be of greatimportance in what follows and we record it here as
Lemma 33 (Angular Reconstruction of Norms) Given a test function u andparameter d l one can bound
Xo
kBolduk2
X12l1Yl
12
kukX
12l1Yl
eth26THORN
4 STRUCTURE OF THE Fk SPACES
The purpose of this section is to clarify some remarks of the previous section andwrite down two integral formulas for functions in the Fl space This material is allmore or less standard in the literature (see eg Foschi and Klainerman 2000 Tataru1998 1999) and we include it here primarily because the notation will be useful forour scattering result Our first order of business is to write down a decomposition forfunctions in the Fl space
Lemma 41 (Fl Decomposition) For any ul 2 Fl one can write
ul frac14 uXlthorn u
X1=2
l1thorn uYl eth27THORN
Global Regularity for NLWE 1515
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
where uXlis a solution to the homogeneous wave equation u
X1=2
l1is the Fourier
transform of an L1 function and uYl satisfies
uYleth0THORN frac14 tuYleth0THORN frac14 0
Furthermore one has the norm bounds
1
CkulkFl
ku
XlkL1ethL2THORN thorn ku
X1=2
l1kX
12l1
thorn kuYlkYl
CkulkFl eth28THORN
We now show that the two inhomogeneous terms on the right hand side of (27)can be written as integrals over solutions to the wave equation with L2 data Thisfact will be of crucial importance to us in the sequel The first formula is simply arestatement of (12)
Lemma 42 (Duhamelrsquos Principle) Using the same notation as above for any uYl one can write
uYlethtTHORN frac14 Z t
0
jDxj1 sinetht sTHORNjDxj
ampuYlethsTHORNds eth29THORN
Likewise one can write the uX
1=2
l1portion of the sum (27) as an integral over
modulated solutions to the wave equation be foliating Fourier space by forwardand backward facing light-cones
Lemma 43 (X12
l1 Trace Lemma) For any uX
1=2
l1 let u
X1=2
l1
denote its restriction to the
frequency half space 0 lt t Then one can write
uX
1=2
l1
ethtTHORN frac14Z
e2pitseitjDxjuls ds eth30THORN
where uls is the spatial Fourier transform of fuu restricted to the sth translate of theforward or backward light-cone light cone in Fourier space ie
cuulsethxTHORN frac14Z
detht s jxjTHORNfuuetht xTHORNdtIn particular one has the formulaZ
kulskL2ds kuX
1=2
l1
kX
12l1
eth31THORN
5 STRICHARTZ ESTIMATES
Our inductive estimates will be based on a method of bilinear decompositionsand local Strichartz estimates as in the work Tataru (1999) We first state thestandard Strichartz from which the local estimates follow
1516 Sterbenz
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Lemma 51 (Homogeneous Strichartz Estimates see Keel and Tao 1998) Let5 lt n s frac14 n1
2 and suppose u is a given function of the spatial variable only Thenif 1
qthorn s
r s
2 and1qthorn n
rfrac14 n
2 g the following estimate holds
keitjDxjPlukLqt ethLr
xTHORN lgkPlukL2 eth32THORN
Combining the L2ethL2ethn1THORNn3 THORN endpoint of the above estimate with a local Sobolev in the
spatial domain we arrive at the following local version of (32)
Lemma 52 (Local Strichartz Estimate) Let 5 lt n then the following estimateholds
keitjDxjBo
lethldTHORN12ukL2
t ethL1x THORN l
nthorn14 d
n34 kBo
lethldTHORN12ukL2 eth33THORN
Using the integral formulas (29) and (30) we can transfer the above estimates tothe Fl spaces
Lemma 53 (Fl Strichartz Estimates) Let 5 lt n and set s frac14 n12 Then if 1
qthorn s
r s
2
and 1qthorn n
rfrac14 n
2 g the following estimates hold
kSlukLqethLrTHORN lgkukFl eth34THORN
Xo
kSolduk2L2ethL1THORN
12
lnthorn14 d
n34 kukFl
eth35THORN
Proof of Lemma 53 It suffices to prove the estimate (34) as the estimate (35)follows from this and a local Sobolev embedding combined with the re-summingformula (26) Using the decomposition (41) and the angular reconstruction formula(26) it is enough to prove (34) for functions u
X1=2
l1and uYl Using the integral formula
(30) we see immediately that
kuX
1=2
l1kLqethLrTHORN
X
ZkeitjDxjulskLqethLrTHORN ds
lgX
ZkulskL2 ds
lgkuX
1=2
l1kX
12l1
For the uYl portion of things we can chop the function up into a fixed number ofspacendashtime angular sectors using L1 convolution kernels Doing this and using Ra
Global Regularity for NLWE 1517
4_LPDE29_09amp10_R3_102804
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
to denote an operator from the set fI ijDxj1g we estimate
kuYlkLqethLrTHORN l1Xa
kauYlkLqethLrTHORN
l1Xa
ZkeitjDxjeisjDxjRaampuYleths xTHORN
kLqt ethLr
xTHORN ds
lgl1Xa
ZkeisjDxjRaampuYleths xTHORNkL2
xds
lg kuYlkYl amp
A consequence of (34) is that we have the embedding
X12
l1 13 L1ethL2THORN
Using a simple approximation argument along with uniform convergence we arriveat the following energy estimate for the Fs and Gs spaces
Lemma 54 (Energy Estimates) For space-time functions u one has the followingestimates
kukCeth _HHsTHORNCeth1THORNeth _HHs1THORN kukFs eth36THORN
kukCeth _BBsTHORNCeth1THORNeth _BBs1THORN kukGs eth37THORN
Also by duality and the estimate (5) we have that
lX1L1ethL2THORN 13 lN1X1
2
l1 13 X12
l1 eth38THORN
This proves shows
Lemma 55 (L2 Estimate for Yl) The following inclusion holds uniformly
d12SldethYlTHORN 13 L2ethL2THORN eth39THORN
in particular by dyadic summing one has
d12SldethFlTHORN 13 L2ethL2THORN
6 SCATTERING
It turns out that our scattering result Theorem 12 is implicitly contained in thefunction spaces Fs and Gs That is there is scattering in these spaces independentlyof any specific equation being considered Therefore to prove Theorem 12 it willonly be necessary to show that our solution to (1) belongs to these spaces
1518 Sterbenz
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Using a simple approximation argument it suffices to deal with things at fixedfrequency In what follows we will denote the space _HH1 1
t ethL2THORN to be the Banachspace with fixed time energy norm
kuethtTHORNk2_HH11t ethL2THORN frac14 kuethtTHORNk2_HH1 thorn ktuethtTHORNk2L2
Because the estimates in Theorem 12 deal with more than one derivative we willshow that
Lemma 61 (Fl Scattering) For any function ul 2 Fl there exists a set of initialdata ethf
l gl THORN 2 PlethL2THORN lPlethL2THORN such that the following asymptotic holds
limt1kulethtTHORN Wethfthorn
l gthornl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth40THORN
limt1kulethtTHORN Wethf
l gl THORNethtTHORNk _HH11
t ethL2THORN frac14 0 eth41THORN
Proof of Lemma 61 Using the notation of Sec 4 we may write
ul frac14 uXlthorn uthorn
X1=2
l1
thorn uX
1=2
l1
thorn uYl
We now define the scattering data implicitly by the relations
Wethfthornl g
thornl THORNethtTHORN frac14 u
XlZ 1
0
jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Wethfl g
l THORNethtTHORN frac14 u
XlZ 0
1jDxj1 sin jDxjetht sTHORNeth THORNampuYlethsTHORNds
Using the fact that ampuYl has finite L1ethL2THORN norm it suffices to show that one has the
limits
limt1
kuthornX
1=2
l1
ethtTHORN thorn uX
1=2
l1
ethtTHORNk _HH11t ethL2THORN frac14 0
Squaring this we see that we must show the limits
limt1
ZjDxjuthorn
X1=2
l1
ethtTHORN jDxjuX
1=2
l1
ethtTHORN frac14 0 eth42THORN
limt1
Ztu
thornX
1=2
l1
ethtTHORN tuX
1=2
l1
ethtTHORN frac14 0 eth43THORN
Wersquoll only deal here with the limit (42) as the limit (43) follows from a virtuallyidentical argument Using the trace formula (30) along with the Plancherel theorem
Global Regularity for NLWE 1519
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
we compute
ethLHSTHORNeth42THORNfrac14 lim
t14p2
Ze2pitethjxjjxjTHORNjxj2
Ze2pits1 duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORNds1 ds2 dx
By (31) we have the bounds
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN
ds1 ds2 L1x
l2 kuthornX
1=2
l1
kX
12l1
kuX
1=2
l1
kX
12l1
This shows that the function
HtethxTHORN frac14 jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORN ds1 ds2
is bounded point-wise by an L1 function uniformly in t Therefore by the dominatedconvergence theorem it suffices to show that we in fact have that limt1 Ht frac14 0To see this notice that by the above bounds in conjunction with Fubinirsquos theoremwe have that for almost every fixed x the integral
jxj2Z duthornls1thorns2
uthornls1thorns2ethxTHORN duls2uls2ethxTHORN ds2
is in L1s1 The result now follows from the Riemann Lebesgue Lemma Explicitly one
has that for almost every fixed x the following limit holds
limt1
HtethxTHORN frac14 limt1
jxj2Z
e2pits1 duthornls1thorns2uthornls1thorns2
ethxTHORN duls2uls2ethxTHORNds1 ds2 frac14 0 amp
7 INDUCTIVE ESTIMATES I
Our solution to (1) will be produced through the usual procedure of Picard itera-tion Because the initial data and our function spaces are both invariant with respectto the scaling (3) any iteration procedure must effectively be global in time There-fore we shall have no need of an auxiliary time cutoff system as in the worksKlainerman and Machedon (1995) and Tataru (1999) Instead we write (1) directlyas an integral equation
f frac14 Wethf gTHORN thornamp1NethfDfTHORN eth44THORN
By the contraction mapping principle and the quadratic nature of the nonlinearityto produce a solution to (44) which satisfies the regularity assumptions of our maintheorem it suffices to prove the following two sets of estimates
Theorem 71 (Solution of the Division Problem) Let 5 lt n then the F and G
spaces solve the division problem for quadratic wave equations in the sense that
1520 Sterbenz
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
for any of the model systems we have written above YM WM or MD one has thefollowing estimates
kamp1NethuDvTHORNkGsc kukGsckvkGsc eth45THORN
kamp1NethuDvTHORNkFs kukGsc kvkFs thorn kukFskvkGsc eth46THORN
The remainder of the paper is devoted to the proof of Theorem 71 In whatfollows we will work exclusively with the equation
f frac14 Wethf gTHORN thornamp1ethfHfTHORN eth47THORN
In this case we set sc frac14 n22 The proof of Theorem 71 for the other model equations
can be achieved through a straightforward adaptation of the estimates we give hereIn fact after the various derivatives and values of sc are taken into account the proofin these cases follows verbatim from estimates (49) and (50) below
Our first step is to take a LittlewoodndashPaley decomposition of amp1ethuHvTHORN withrespect to space-time frequencies
amp1ethuHvTHORN frac14Xmi
amp1ethSm1uHSm2vTHORN eth48THORN
We now follow the standard procedure of splitting the sum (48) into three piecesdepending on the cases m1 m2 m2 m1 and m2 m1 Therefore due to the lsquo1 Besovstructure in the F spaces in order to prove both (45) and (46) it suffices to show thetwo estimates
kamp1ethSm1uHSm2vTHORNkGl l1m
n2
1kukFm1kvkFm2
m1 m2 eth49THORN
kamp1ethSmuHSlvTHORNkGl m
n22 kukGm
kvkFl m l eth50THORN
Notice that after some weight trading the estimates (45) and (46) follow from (50) inthe case where m2m1
Proof of (49) It is enough if we show the following two estimates
kSlethSm1uHSm2vTHORNkL1ethL2THORN mn2
1kukFm1kvkFm2
m1 m2 eth51THORN
kSlamp1ethSm1uHSm2vTHORNkL1ethL2THORN l1mn2
1kukFm1kvkFm2
m1 m2 eth52THORN
In fact it suffices to prove (51) To see this notice that one has the formula
Slamp1
G frac14 WethE SlGTHORN SlWethE GTHORN
Global Regularity for NLWE 1521
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Thus after multiplying by Sl we see that
Sl Slamp1
G frac14 PlWethE SlGTHORN SlWethE GTHORN
frac14 WethE SlPlGTHORN SlWethE PlGTHORN
frac14 Sl Slamp1
PlG
Therefore by the (approximate) idempotence of Sl one has
Slamp1G frac14 Slamp1SlGthorn Sl Slamp1
G
frac14 Slamp1SlGthorn Sl Slamp1
PlG
Thus by the boundedness of Sl on the spaces L1ethL2THORN the energy estimate one canbound
kSlamp1GkL1ethL2THORN l1 kSlGkL1ethL2THORN thorn kPlGkL1ethL2THORN
We now use the fact that the multipliers Sl and Pl are both bounded on the spaceL1ethL2THORN to reduce things to the estimate
kSm1uHSm2vkL1ethL2THORN kSm1ukL2ethL4THORNkHSm2vkL2ethL4THORN
mn24
1 mnthorn24
2 kukFm1kvkFm2
Taking into account the bound m1 m2 the claim now follows amp
Next wersquoll deal with the estimate (50) For the remainder of the paper we shallfix both l and m and assume they such that m l for a fixed constant We nowdecompose the product SlethSmuHSlvTHORN into a sum of three pieces
SlethSmuHSlvTHORN frac14 Athorn Bthorn C eth53THORN
where
A frac14 SlethSmuHSlcmvTHORNB frac14 SlcmethSmuHSlltcmvTHORNC frac14 SlltcmethSmuHSlltcmvTHORN
Here c is a suitably small constant which will be chosen later It will be needed tomake explicit a dependency between some of the constants which arise in a specificfrequency localization in the sequel We now work to recover the estimate (50) foreach of the three above terms separately
1522 Sterbenz
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Proof of (50) for the Term A Following the remarks at the beginning of the proofof (49) it suffices to compute
kSlethSmuHSlcmvTHORNkL1ethL2THORN l kSmukL2ethL1THORNkSlcmvkL2ethL2THORN
lmn12 kukFm
ethcmTHORN12kvkFl
c1lmn22 kukFm
kvkFl
For a fixed c we obtain the desired result amp
We now move on to showing the inclusion (50) for the B term above In thisrange we are forced to work outside the context of L1ethL2THORN estimates This is thereason we have included the L2ethL2THORN based X
12
l1 spaces This also means that we willneed to recover Zl norms by hand (because they are only covered by the Yl spaces)However because this last task will require a somewhat finer analysis than what wewill do in this section we content ourselves here with showing
Proof of the X12
l1 SlethL1ethL2THORNTHORN Estimates for the Term B Our first task will be dealwith the energy estimate which we write as
kSlamp1SlcmethSmuHSlltcmvTHORNkL1ethL2THORNmn22 kukFm
kvkFl eth54THORN
For G supported away from the light-cone in Fourier space we have the identity
Slamp1SlG frac14 X1SlGWethX1PlSlGTHORN
Therefore by using the energy estimate for the X12
l1 space this allows us to estimate
kSlamp1SlGkL1ethL2THORN kX1SlGkL1ethL2THORN thorn kWethX1PlSlGTHORNkL1ethL2THORN
kX1SlGkL1ethL2THORN
kX1SlGkX
12l1
Therefore in order to prove the estimate (54) we are left with estimating the term B
in the X12
l1 space To do this for a fixed distance d from the cone we compute that
kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN d1 kSmukL2ethL1THORNkSlvkL1ethL2THORN
d1mn22 kvkFm
kukFl
Global Regularity for NLWE 1523
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Summing d12 times this last expression over all cm d yieldsX
cmd
d12kX1SldSlcmethSmuHSlltcmvTHORNkL2ethL2THORN
Xcmd
md
12
mn22 kvkFm
kukFl
For a fixed c we obtain the desired result amp
8 INTERLUDE SOME BILINEAR DECOMPOSITIONS
To proceed further it will be necessary for us to take a closer look at theexpression
SoldethSmuHSlltcmvTHORN cm d eth55THORN
as well as the C term from line (53) above which we write as the sum
C frac14 SlltcmethSmuHSlltcmvTHORN frac14 CI thorn CII thorn CIII
where
CI frac14Xdltcm
SldethSmdu HSldvTHORN
CII frac14Xdltcm
SlltdethSmdu HSldvTHORN
CIII frac14Xdm
SlltminfcmdgethSmdu HSlltminfcmdgvTHORN
Wersquoll begin with a decomposition of CI and CII The CIII term is basically thesame but requires a slightly more delicate analysis All of the decompositions wecompute here will be for a fixed d The full decomposition will then be given by sum-ming over the relevant values of d Because our decompositions will be with respectto Fourier supports it suffices to look at the convolution product of the correspond-ing cutoff functions in Fourier space In what follows wersquoll only deal with the CI
term It will become apparent that the same idea works for CII Therefore withoutloss of generality we shall decompose the product
sthornld smd sthornld
eth56THORN
To do this we use the standard device of restricting the angle of interaction inthe above product It will be crucial for us to be able to make these restrictionsbased only on the spatial Fourier variables because we will need to reconstructour decompositions through square-summing For etht0 x0THORN 2 suppfsmdg and
1524 Sterbenz
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
etht xTHORN 2 suppfsthornldg we compute that
OethdTHORN frac14 jt0 thorn tj jx0 thorn xjfrac14 jx0j thorn jxj thorn OethdTHORN jx0 thorn xj
frac14OethdTHORN thorn jx0j thorn jxj jx0 thorn xj
Using now the fact that d lt cm and m lt cl to conclude that jx0j m and jxj l wesee that one has the angular restriction
mY2x0x
jx0j thorn jxj jx0 thorn xjfrac14OethdTHORN
In particular we have that Yx0x ffiffidm
q This allows us to decompose the product (56)
into a sum over angular regions with O
ffiffidm
q spread The result is
Lemma 81 (Wide Angle Decomposition) In the ranges stated for the CI termabove one can write
sthornldethsmd sthornldTHORN frac14X
o1 o2 o3
jo1o2 jethd=mTHORN12
jo1o3 jethd=mTHORN12
bo1
llethdmTHORN12
sthornld so2
md
bo3
llethdmTHORN12
sthornld
eth57THORN
for the convolution of the associated cutoff functions in Fourier space
We note here that the key feature in the decomposition (57) is that the sum is(essentially) diagonal in all three angles which appear there (o1o2o3) It is usefulto keep in mind the diagram (Fig 1)
Figure 1 Spatial supports in the wide angle decomposition
Global Regularity for NLWE 1525
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
We now focus our attention on decomposing the convolution
sthornlminfcmdgsmd sthornlminfcmdg
eth58THORN
If it is the case that d m then the same calculation which was used to produce (57)works and we end up with the same type of sum However if we are in the case whered m we need to compute things a bit more carefully in order to ensure that we maystill decompose the multiplier smd using only restrictions in the spatial variable Todo this we will now assume that things are set up so that cm d It is clear thatall the previous decompositions can be made so that we can reduce things to thisconsideration If we now take etht0 x0THORN 2 suppfsmdg and etht xTHORN 2 suppfsthornlminfcmdggwe can use the facts that t0 frac14 OethdTHORN jx0j t frac14 OethcmTHORN thorn jxj and jxj m to computethat
OethcmTHORN frac14t0 thorn t
jx0 thorn xj
frac14OethdTHORN jx0j thorn OethcmTHORN thorn jxj jx0 thorn xj
eth59THORN
where the term OethdTHORN in the above expression is such that jOethdTHORNj d In fact onecan see that the equality (59) forces OethdTHORN lt 0 on account of the fact thatethjx0j thorn jxj jx0 thorn xjTHORN gt 0 and the assumption jOethcmTHORN thorn OethdTHORNj d In particularthis means that we can multiply smd in the product (58) by the cutoff sjtjltjxj withouteffecting things This in turn shows that we may decompose the product (58) basedsolely on restriction of the spatial Fourier variables just as we did to get the sum inLemma 81
We now return to the CI term For the sequel we will need to know whatthe contribution of the factor Sldv to the following localized product is
So1
ldethSmdu HSldvTHORN
Using Lemma 81 we see that we may write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md bo3
llethdmTHORN12
sld
eth60THORN
where jo1 o2j jo3 o2j ffiffidm
q However this can be refined significantly To
see this assume that the spatial support of so1
ld lies along the positive x1 axis Wersquolllabel this block by b
o1
lethldTHORN12 Because we are in the range where
ffiffiffiffiffiffimd
p ffiffiffiffiffiffild
p and
furthermore because for every x 2 suppfbo3
llethdmTHORN12
g and x0 2 suppx0 fso2
mdg the sum
xthorn x0 must belong to suppfbo1
lethldTHORN12g we in fact have that x itself must belong to a
1526 Sterbenz
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
block of size l ffiffiffiffiffiffild
p ffiffiffiffiffiffild
p This allows us to write
Lemma 82 (Small Angle Decomposition) In the ranges stated for the CI termabove we can write
so1
ldethsmd sldTHORN frac14 so1
ld
so2
md so3
ld
eth61THORN
where jo1 o3j ffiffidl
q and jo1 o2j
ffiffidm
q
It is important to note here that if one were to sum the expression (60) over o1the resulting sum would be (essentially) diagonal in o3 but there would be many o1
which would contribute to a single o2 This means that the resulting sum would notbe diagonal in o2 as was the case for the sum (57) It is helpful to visualize thingsthrough the Fig 2
Our final task here is to mention an analog of Lemma 82 for the term (55) Herewe can frequency localize the factor Slltcm in the product using the fact that one hasm c
12
ffiffiffiffiffiffild
p The result is
Lemma 83 (Small Angle Decomposition for the Term B) In the ranges stated forthe B term above we can write
so1
ldethsm slcmTHORN frac14 so1
ld
sm b
o3
lethldTHORN12slcm
eth62THORN
where jo1 o3j ffiffidl
q
Finally we note here the important fact that in the decomposition (62) abovethe range of interaction in the product forces d m This completes our list ofbilinear decompositions
Figure 2 Spatial supports in the small angular decomposition
Global Regularity for NLWE 1527
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
9 INDUCTIVE ESTIMATES II REMAINDER OF THELowHigh)High FREQUENCY INTERACTION
It remains for us is to bound the term B from line (55) in the Zl space as well asshow the inclusion (50) for the terms CI ndash CIII from line (8) We do this now proceed-ing in reverse order
Proof of Estimate (50) for the CIII Term To begin with we fix d Using the remarksat the beginning of the proof of (49) we see that it is enough to show that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN lX
o
kSomduk2L1ethL1THORN
12
kvkFl
eth63THORN
To accomplish this we first use the wide angle decomposition (57) on the left handside of (63) This allows us to compute using a CauchyndashSchwartz that
kSlltminfcmdgethSmdu HSlltminfcmdgvTHORNkL1ethL2THORN
Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL1ethL1THORN kHBo3
llethdmTHORN12
SlltminfcmdgvTHORNkL1ethL2THORN
lXo
kSomduk2L1ethL1THORN
12 X
o
kBo
llethdmTHORN12SlltminfcmdgvTHORNk2L1ethL2
xTHORN
12
lXo
kSomduk2L1ethL1THORN
12
kvkFl
Summing over d now yields the desired estimate amp
Proof of (50) for the CII Term Again fixing d and using the angular decomposi-tion Lemma 81 we compute that
kSlltdethSmdu HSldvTHORNkL1ethL2THORN
lXo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvkL2ethL2THORN
lXo
kSomduk2L2ethL1THORN
12
kSldvkL2ethL2THORN
lmn22
d
m
n54
kukFmkvkFl
1528 Sterbenz
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
This last expression can now be summed over d using the condition d lt cm toobtain the desired result amp
Proof of (50) for the CI Term This is the other instance where we will have to relyon the X
12
l1 space Following the same reasoning used previously we first bound
kX1SldethSmdu HSldvTHORNkL2ethL2THORN
d1Xo2 o3
jo3o2 jethd=mTHORN12
kSo2
mdukL2ethL1THORN kBo3
llethdmTHORN12
SldvTHORNkL1ethL2THORN
d1Xo
kSomduk2L2ethL1THORN
12
Xo
kBo
llethdmTHORN12SldvTHORNk2L1ethL2
xTHORN
12
d12m
n22
d
m
n54
kukFmkvkFl
Multiplying this last expression by d12 and then using the condition d lt cm to sum
over d yields the desired result for the X12
l1 space part of estimate (50) It remainsto prove the Zl estimate Here we use the second angular decomposition Lemma 82to compute that for fixed d
Xo1
kX1So1
ldethSmdu HSldvTHORNk2L1ethL1THORN
12
ethldTHORN1X
o1 o2 o3
o1o3ethd=lTHORN12
o1o2ethd=mTHORN12
kSo1
ld
So2
mdu HSo3
ldv
k2L1ethL1THORN
0BBBBBB
1CCCCCCA
12
d1 supo
kSomdukL2ethL1THORN Xo
kSoldvk2L2ethL1THORN
12
d
m
n54 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying this last expression by leth2nTHORNeth2THORN and summing over d using the condition
d lt l m yields the desired result amp
Proof of the Zl Embedding for the B Term The pattern here follows that of thelast few lines of the previous proof Fixing d we use the decomposition Lemma 83
Global Regularity for NLWE 1529
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
to compute that
Xo
kN1SoldethSmu HSlcmvTHORNk2L1ethL1THORN
12
ethldTHORN1Xo1 o3
jo1o3 jethd=lTHORN12
kSo1
ld
Smu HBo3
lethldTHORN12Slcmv
k2L1ethL1THORN
0BB1CCA
12
d1kSmukL2ethL1THORN Xo
kBo
lethldTHORN12Slcmvk2L2ethL1THORN
12
md
12 d
l
n54
mn22 l
n22 kukFm
kvkFl
Multiplying the last line above by a factor of leth2nTHORNeth2THORN and using the conditions d lt l
and cm lt d m we may sum over d to yield the desired result amp
ACKNOWLEDGMENT
This work was conducted under NSF grant DMS-0100406
REFERENCES
Bournaveas N (1996) Local existence for the MaxwellndashDirac equations in threespace dimensions Comm Partial Differential Equations 21(5ndash6)693ndash720
Cazenave T Shatah J Shadi Tahvildar-Zadeh A (1998) Harmonic maps of thehyperbolic space and development of singularities in wave maps andYangndashMills fields Ann Inst H Poincare Phys Theor 68(3)315ndash349
Foschi D Klainerman S (2000) Bilinear space-time estimates for homogeneouswave equations Ann Sci cole Norm Sup (4) 33(2)211ndash274
Keel M Tao T (1998) Endpoint Strichartz estimates Am J Math 120(5)955ndash980
Klainerman S (1985) Uniform decay estimates and the Lorentz invariance of theclassical wave equation Comm Pure Appl Math 38(3)321ndash332
Klainerman S Machedon M (1993) Space-time estimates for null forms and thelocal existence theorem Comm Pure Appl Math 46(9)1221ndash1268
Klainerman S Machedon M (1995) Smoothing estimates for null forms andapplications Duke Math J 81(1)99ndash103
Klainerman S Machedon M (1996) Estimates for null forms and the spaces HsdInt Math Res Notices 17853ndash865
1530 Sterbenz
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
ORDER REPRINTS
Klainerman S Tataru D (1999) On the optimal local regularity for YangndashMillsequations in R4thorn1 J Am Math Soc 12(1)93ndash116
Klainerman S Selberg S (2002) Bilinear estimates and applications to nonlinearwave equations Commun Contemp Math 4(2)223ndash295
Lindblad H (1996) Counterexamples to local existence for semi-linear waveequations Am J Math 118(1)1ndash16
Rodnianski I Tao T Global regularity for the MaxwellndashKleinndashGordon equationin high dimensions Preprint
Tao T (2001) Global regularity of wave maps I Small critical Sobolev norm in highdimension Int Math Res Notices 6299ndash328
Tataru D (1998) Local and global results for wave maps I Comm PartialDifferential Equations 23(9ndash10)1781ndash1793
Tataru D (2001) On global existence and scattering for the wave maps equationAm J Math 123(1)37ndash77
Tataru D (1999) On the equationampu frac14 jHuj2 in 5thorn 1 dimensionsMath Res Lett6(5ndash6)469ndash485
Received October 2003Accepted January 2004
Global Regularity for NLWE 1531
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details
Request PermissionOrder Reprints
Reprints of this article can also be ordered at
httpwwwdekkercomservletproductDOI101081PDE200037764
Request Permission or Order Reprints Instantly
Interested in copying and sharing this article In most cases US Copyright Law requires that you get permission from the articlersquos rightsholder before using copyrighted content
All information and materials found in this article including but not limited to text trademarks patents logos graphics and images (the Materials) are the copyrighted works and other forms of intellectual property of Marcel Dekker Inc or its licensors All rights not expressly granted are reserved
Get permission to lawfully reproduce and distribute the Materials or order reprints quickly and painlessly Simply click on the Request Permission Order Reprints link below and follow the instructions Visit the US Copyright Office for information on Fair Use limitations of US copyright law Please refer to The Association of American Publishersrsquo (AAP) website for guidelines on Fair Use in the Classroom
The Materials are for your personal use only and cannot be reformatted reposted resold or distributed by electronic means or otherwise without permission from Marcel Dekker Inc Marcel Dekker Inc grants you the limited right to display the Materials only on your personal computer or personal wireless device and to copy and download single copies of such Materials provided that any copyright trademark or other notice appearing on such Materials is also retained by displayed copied or downloaded as part of the Materials and is not removed or obscured and provided you do not edit modify alter or enhance the Materials Please refer to our Website User Agreement for more details