• Aucun résultat trouvé

Sur un opérateur bilinéaire discret singulier

N/A
N/A
Protected

Academic year: 2022

Partager "Sur un opérateur bilinéaire discret singulier"

Copied!
5
0
0

Texte intégral

(1)

Contents lists available atScienceDirect

C. R. Acad. Sci. Paris, Ser. I

www.sciencedirect.com

Harmonic analysis

On a discrete bilinear singular operator

Sur un opérateur bilinéaire discret singulier

Dong Dong

DepartmentofMathematics,UniversityofIllinoisatUrbana–Champaign,1409W.GreenStreet,Urbana,IL61801,USA

a rt i c l e i n f o a b s t ra c t

Articlehistory:

Received27September2016 Acceptedafterrevision15March2017 Availableonline29March2017 PresentedbyYvesMeyer

We prove that for a large class of functions P and Q, the discrete bilinear operator TP,Q(f,g)(n)=

m∈Z\{0}f(nP(m))g(nQ(m))m1 isboundedfroml2×l2intol1+, forany

(0,1].

©2017Académiedessciences.PublishedbyElsevierMassonSAS.Allrightsreserved.

ré s u m é

Nousmontrons, quepour unegrande classe defonctions P et Q,l’opérateur bilinéaire discret TP,Q(f,g)(n)=

m∈Z\{0}f(nP(m))g(nQ(m))m1 est borné de l2×l2 dans l1+,∞,pourtout

(0,1].

©2017Académiedessciences.PublishedbyElsevierMassonSAS.Allrightsreserved.

1. Introduction

TheHilberttransform(HTforshort)isdefinedby

H

(

f

)(

x

) =

f

(

x

t

)

dt

t

,

f

S

(R),

whereS(Rn

)

,n∈N,istheSchwartzspaceonRn.Itwasprovedin1928([21])thatHTisboundedonLp forp

(

1

,

)

.An interestinggeneralizationofHTistheso-calledHTalongcurves:

HC

(

f

)(

x

) =

f

(

x

γ (

t

))

dt

t

,

f

S

( R

n

).

Here

γ

:R→Rn isawell-behavedcurve.The Lp boundednessofHC hasbeenobtainedforvarious curves.See[22]fora comprehensivesurvey and[2]forageneralization ofHC tothenon-translation-invariantsetting.When

γ

isapolynomial withintegercoefficients,thereisadiscreteversionofHC definedby

E-mailaddress:ddong3@illinois.edu.

http://dx.doi.org/10.1016/j.crma.2017.03.010

1631-073X/©2017Académiedessciences.PublishedbyElsevierMassonSAS.Allrightsreserved.

(2)

HdisC

(

f

)(

n

) =

m∈Z\{0}

f

(

n

γ (

m

))

1

m

,

f

D

( Z

n

),

whereD

(Z

n

)

isthespaceofcompactlysupportedcomplex-valuedfunctionsdefinedonZn.Ontheonehand,HdisC hasmany applicationsinergodictheory([6,16–18,20]),butontheotherhandthisdiscreteoperatorismoresubtletohandlethanits continuous counterpart HC,asmanynumbertheoreticaltoolsare involved. HdisC wasatfirstproved tobe boundedonlp forp

(

32

,

3

)

([23]).Thisrestrictedrangewasextendedtothefullrange

(

1

,

∞)alongtimelater([7,15]).

AnotherdirectionofgeneralizingHTistoconsideritsbilinear analogue,whichissignificantlymoredifficulttoanalyze sincePlancherelTheoremisunavailableinthebilinearsetting.ThebilinearHilberttransform(BHTforshort)canbedefined as

B

(

f

,

g

)(

x

) =

f

(

x

t

)

g

(

x

+

t

)

dt

t

,

f

,

g

S

( R ).

Itwasabout70yearsafterthefirstproofoftheboundednessofHTthatLaceyandThiele([9,10])obtainedtheLp estimates forBHT.Veryrecently,Lp estimatesforBHTalongcurves

BC

(

f

,

g

)(

x

) =

f

(

x

t

)

g

(

x

γ (

t

))

dt

t

,

f

,

g

S

( R ),

werealsoestablishedwhen

γ

isapolynomial([14]).Notethat BC isanaturalbilinearversionofHC. Followingthedevelopmentofthelinearcase,inthispaperweconsiderthediscreteversionofBC,thatis,

BdisC

(

f

,

g

)(

n

) =

m∈Z\{0}

f

(

n

m

)

g

(

n

P

(

m

))

1

m

,

f

,

g

D

(Z),

where P isapolynomialwithintegercoefficients. Thisoperatorcanalsobe viewedasabilinearanalogueofHdisC .As HdisC ishardertohandlethan HC,itisreasonabletoexpectthatprovingtheboundednessofBdisC shouldbemoredifficultthan that of BC. As astarting pointof thelong journey ofinvestigation on BdisC ,in thisarticlewe show thel2×l2l1+,∞

boundednessof BdisC (Theorem 1.1).

Wewillstudyanoperatorthatismoregeneralthan BdisC (see (1.1)).Giventwofunctions P and Q thatmap ZintoZ, define

AP,Q

:=

(

m1

,

m2

)( Z \ {

0

} )

2

:

P

(

m1

)

Q

(

m1

) =

P

(

m2

)

Q

(

m2

)

.

Wesaythatthepairoffunctions

(

P

,

Q

)

satisfiescondition

()

ifthereareconstants D1 andD2suchthat ||mm1|

2|≤D1forall

(

m1

,

m2

)

AP,Q andforeachm1∈Z,thereareatmostD2 pairs

(

m1

,

m2

)

intheset AP,Q.

Theorem1.1.GiventwofunctionsP andQ thatmapZintoZ,let

TP,Q

(

f

,

g

)(

n

) :=

m∈Z\{0}

f

(

n

P

(

m

))

g

(

n

Q

(

m

))

1

m

,

f

,

g

D

(

Z

).

(1.1)

Assumethat

(

P

,

Q

)

satisfiescondition

()

.Thenforany

(

0

,

1],thereisaconstantCdependingonlyon ,D1andD2suchthat

TP,Q

(

f

,

g

)

l1+,∞

C

f

l2

g

l2

.

(1.2)

Remarks.(1).Condition

()

ismild. Apairofpolynomials withintegercoefficients

(

P

,

Q

)

satisfiescondition

()

aslong as PQ isnotconstant.Notethat D1 dependsonthecoefficientsof P and Q,sodoesC inthe theorem.It isnatural toexpectthatthisdependencecanberemoved,asuniformestimatesexistforrelatedoperators([3,4,11,12,14,23,24]).We shallnotpursuitthishere.

(2).WeconjecturethatatleastforsomespecialpairsofP andQ (forexample,P

(

t

)

=tandQ

(

t

)

=t2),TP,Q isbounded fromlp×lqintolr,wherep

,

q

(

1

,

∞), 1p+1q=1r.Thisproblemisverydifficultandcurrentlyoutofreach.

(3). A useful operator related with TP,Q is the corresponding maximal operator TP,Q

(

f

,

g

)(

n

)

=supM∈[1,)|M1 ×

M

m=1 f

(

nP

(

m

))

g

(

nQ

(

m

))

|,whichisatfirstprovedtobeboundedfroml2×l2 tolr whenr

>

1 ([5]).ByusingHölder inequalityandboundednessofthecorrespondingdiscretelinearmaximalfunction fsupM∈[1,)]M1

M

m=1|f

(

nP

(

m

))

| (see,forexample,[1,8,19,26]),we canprovethat TP,Q isboundedfromlp×lq intolr,wheneverp

,

q

(

1

,

∞), 1p+1q=1r, andr

>

1 (seep. 75in[25]forasimilartrick).Whethertherestrictionr

>

1 canbedroppedisstillunknown.

(4).See[13]foradiscussionaboutanergodicanalogueofTP,Q.

TherestofthepaperisdevotedtotheproofofTheorem 1.1.

(3)

2. ProofofTheorem 1.1

We will use AB todenote thestatement that AC B forsome positive constant C.When theimplied constant C dependsonr,wewrite ArB.AlltheconstantsmaydependonD1 andD2 (appearedinthedefinitionofcondition

()

), but thisdependence will be suppressed since D1 and D2 are oftenfixed inapplications. AB is short for AB and BA.Foranysetofintegers E,|E|and

χ

E willbe usedtodenotethecountingmeasureandtheindicatorfunctionofE,

receptively.

Let P andQ beapairoffunctionssatisfyingcondition

()

.Fornotationalconvenience,wewillsimplywrite T forTP,Q andr:=1+

.Forany

λ >

0 and f

,

gD

(

Z

)

,define thelevelset = {n∈Z: |T

(

f

,

g

)(

n

)

|

> λ

}.Ourgoalistoprovethe followinglevelsetestimate

|

|

r

1

λ

r

,

whenever

f

l2

=

g

l2

=

1

.

(2.3)

Wefirstwrite T=

m∈Z\{0}f

(

nP

(

m

))

g

(

nQ

(

m

))

m1 asabilinearmultiplieroperator.RecalltheFouriertransformfor any fD

(

Z

)

isdefinedby ˆf

(ξ )

:=

m∈Zf

(

m

)

e2πiξm.Hence

T

(

f

,

g

)(

n

) =

T

T

ˆ

f

(ξ )

g

ˆ ( η )

e2πi(ξ+η)n

σ (ξ, η )

d

ξ

d

η ,

whereTistheunitcircleand

σ

istheperiodicmultiplier(a.k.a. symbol)givenby

σ (ξ, η ) =

m∈Z\{0} 1

me2πi(P(m)ξ+Q(m)η)

.

Then we decomposedyadicallythesymbol

σ

asfollows.Pickan oddfunction

ρ

S

(

R

)

supportedintheset {x: |x|∈

(

12

,

2

)

}withthepropertythat

1 x

=

j=0

1 2j

ρ

x

2j

for anyx

∈ R

with

|

x

| ≥

1

.

Sothesymbol

σ

canbewrittenas

σ (ξ, η )

=

j=0

σ

j

(ξ, η )

,where

σ

j

(ξ, η ) :=

1 2j

m∈Z

ρ

m

2j

e2πi(P(m)ξ+Q(m)η)

.

Correspondingly T=

j=0Tj,where

Tj

(

f

,

g

)(

n

) =

T

T

ˆ

f

(ξ )

g

ˆ ( η )

e2πi+η)n

σ

j

(ξ, η )

d

ξ

d

η

=

1 2j

m∈Z

ρ

m

2j

f

(

n

P

(

m

))

g

(

n

Q

(

m

)).

By thesupportof

ρ

andHölderinequality,itis easyto seeTj

(

f

,

g

)

l1fl2gl2.Sowe havethefollowing levelset estimateforeach Tj.

Lemma2.1.Foranyf

,

gD

(Z)

withl2-norm1, j∈N,and

λ >

0,wehave

|{

n

∈ Z : |

Tj

(

f

,

g

)(

n

)| > λ}|

1

λ .

This lemma says that each single Tj is under good (and uniform) control. The difficulty is how to get the desired estimatesforthesumofTj’s.Inthefollowing,wewillapplytheideaoftheT T method.

Defineanauxiliaryfunctionh

(

n

)

=TT((ff,,gg)()(nn))

χ

E

λ

(

n

)

.Itiseasytoverifythat

λ

2

|

Eλ

|

2

n∈Z

T

(

f

,

g

)(

n

)

h

(

n

)

2

.

(2.4)

ByFubinitheoremandthedefinitionofFouriertransform,

(4)

n∈Z

T

(

f

,

g

)(

n

)

h

(

n

) =

T

T

ˆ

f

(ξ )

g

ˆ ( η ) σ (ξ, η )

h

ˆ (−(ξ + η ))

d

ξ

d

η

=

T

T

ˆ

f

η )

g

ˆ ( η ) σ η , η )

h

ˆ (ξ )

d

ξ

d

η .

ApplytheCauchy–SchwarzinequalityandthePlancherelTheorem,andweget

n∈Z

T

(

f

,

g

)(

n

)

h

(

n

)

2

B

|

|,

(2.5)

where

B

:=

sup

ξT

T

| σ η , η )|

2d

η .

Combining(2.4)and(2.5),weseethat|Eλ|≤λB2.Hence,toprove(2.3),itsufficestoobtaintheestimate

B

r

λ

2r

.

(2.6)

Tocontrol B,wemakeuseofthedyadicdecompositionof

σ

,aimingforsomecancellations.Forany

ξ

T,

T

| σ η , η )|

2d

η =

T

j=0

σ

j

η , η )

2

d

η

j1,j2=0

1 2j1

1 2j2

m1,m2∈Z

ρ

m1

2j1

ρ

m2

2j2

χ

AP,Q

(

m1

,

m2

).

(2.7)

Bycondition

()

, ||mm1|

2|≤D1 forall

(

m1

,

m2

)

AP,Q.The supportof

ρ

forces|m1|2j1 and |m2|2j2.Thesefacts show that|j1j2|1.Alsonotethatforeachm1,thereareonlyboundednumberofm2’ssuchthat

(

m1

,

m2

)

AP,Q.Thus(2.7) implies

B

=

sup

ξ∈T

T

| σ η , η ) |

2d

η

j=0

1 2j

.

When

λ

1, asr

(

1

,

2], trivially B

λ

2r andwe are done.Let M= [(2r

)

log2λ1]+1,where[x] denotestheinteger partof x.Inthe case

λ <

1,since

j=M+121j r

λ

2r,the abovemethod stillgives thedesiredestimate for

j=M+1Tj, theoperatorassociatedwiththesymbol

j=M+1

σ

j.ItremainstocontrolthelevelsetoftheoperatorM

j=0Tj for

λ <

1.

ApplyingLemma 2.1,wehave

⎧ ⎨

n

∈ Z :

M j=0

Tj

(

f

,

g

)(

n

) > λ

⎫ ⎬

M2

λ

r

1

λ

r

,

whereweusedthefactsr

>

1 and

λ <

1 inthelastinequality.ThisfinishestheproofofTheorem 1.1.

References

[1]J.Bourgain,Pointwiseergodictheoremsforarithmeticsets,withanappendixbytheauthor,HarryFurstenberg,YitzhakKatznelsonandDonaldS.

Ornstein,Publ.Math.IHÉS69(1989)5–45.

[2]M.Christ,A.Nagel,E.M.Stein,S.Wainger,SingularandmaximalRadontransforms:analysisandgeometry,Ann.ofMath.(2)150 (2)(1999)489–577.

[3]L.Grafakos,X.Li,UniformboundsforthebilinearHilberttransforms.I,Ann.ofMath.(2)159 (3)(2004)889–933.

[4]L.Grafakos,X.Li,Thediscasabilinearmultiplier,Amer.J.Math.128 (1)(2006)91–119.

[5]Y.Hu,X.Li,DiscreteFourierrestrictionassociatedwithSchrödingerequations,Rev.Mat.Iberoam.30 (4)(2014)1281–1300.

[6]A.Ionescu,E.M.Stein,A.Magyar,S.Wainger,DiscreteRadontransformsandapplicationstoergodictheory,ActaMath.198 (2)(2007)231–298.

[7]A.Ionescu,S.Wainger,LpboundednessofdiscretesingularRadontransforms,J.Amer.Math.Soc.19 (2)(2006)357–383,publishedelectronically:

24 October2005.

[8]B.Krause,Polynomialergodicaveragesconvergerapidly:variationsonatheoremofBourgain,arXiv:1402.1803.

[9]M.Lacey,C.Thiele,LpestimatesonthebilinearHilberttransformfor2<p<,Ann.ofMath.(2)146 (3)(1997)693–724.

[10]M.Lacey,C.Thiele,OnCalderón’sconjecture,Ann.ofMath.(2)149 (2)(1999)475–496.

[11]X.Li,UniformboundsforthebilinearHilberttransforms.II,Rev.Mat.Iberoam.22 (3)(2006)1069–1126.

[12]X.Li,Uniformestimatesforsomeparaproducts,N.Y.J.Math.14(2008)145–192.

(5)

[13]X.Li,BilinearHilberttransformsalongcurvesI:themonomialcase,Anal.PDE6 (1)(2013)197–220.

[14]X.Li,L.Xiao,UniformestimatesforbilinearHilberttransformandbilinearmaximalfunctionsassociatedtopolynomials,Amer.J.Math.138 (4)(2016) 907–962.

[15]M.Mirek,SquarefunctionestimatesfordiscreteRadontransforms,arXiv:1512.07524.

[16]M.Mirek,E.M.Stein,B.Trojan,lp(Zd)-estimatesfordiscreteoperatorsofRadontype:maximalfunctionsandvector-valuedestimates,arXiv:1512.

07518.

[17] M.Mirek,E.M.Stein,B.Trojan,lp(Zd)-estimatesfordiscreteoperatorsofRadontype:variationalestimates,Invent.Math.(2017),http://dx.doi.org/10.

1007/s00222-017-0718-4,inpress,arXiv:1512.07523.

[18]M.Mirek,B.Trojan,Cotlar’sergodictheoremalongtheprimenumbers,J.FourierAnal.Appl.21 (4)(2015)822–848.

[19]M.Mirek,B.Trojan,Discretemaximalfunctionsinhigherdimensionsandapplicationstoergodictheory,Amer.J.Math.138 (6)(2016)1495–1532.

[20] M.Mirek,B.Trojan,P.Zorin-Kranich,Variationalestimatesforaveragesandtruncatedsingularintegralsalongtheprimenumbers,Trans.Amer.Math.

Soc.(2017),https://doi.org/10.1090/tran/6822,inpress,arXiv:1410.3255.

[21]M.Riesz,Surlesfonctionsconjuguées,Math.Z.27 (1)(1928)218–244.

[22]E.M.Stein,S.Wainger,Problemsinharmonicanalysisrelatedtocurvature,Bull.Amer.Math.Soc.84 (6)(1978)1239–1295.

[23]E.M.Stein,S.Wainger,DiscreteanaloguesofsingularRadontransforms,Bull.Amer.Math.Soc.(N.S.)23 (2)(1990)537–544.

[24]C.Thiele,Auniformestimate,Ann.ofMath.(2)156 (2)(2002)519–563.

[25]C.Thiele,WavePacketAnalysis,CBMSRegionalConferenceSeriesinMathematics,vol. 105,2006.

[26]P.Zorin-Kranich,Variationestimatesforaveragesalongprimesandpolynomials,J.Funct.Anal.268 (1)(2015)210–238.

Références

Documents relatifs

Sj¨ olin , P., A counter-example concerning maximal estimates for solutions to equations of Schr¨ odinger type, Indiana Univ.. Sj¨ olin , P., Spherical harmonics and maximal

Our tool to prove convergence in the Hilbert transform setting is the oscillation inequality (1.5), an idea first employed in ergodic theory in the pioneering work of Bourgain on

L’accès aux archives de la revue « Rendiconti del Seminario Matematico della Università di Padova » ( http://rendiconti.math.unipd.it/ ) implique l’accord avec les

The main new ingredient in the proofs of Theorems 1.1, 1.2 and 1.4 is a bilinear estimate in the context of Bourgain’s spaces (see for instance the work of Molinet, Saut and

But Legendre formula also suffers from a congenital disease: if we ignore the error terms, the supposedly main term it gives is incorrect in view of the prime number theorem ∗.. 5

We discuss necessary and sufficient conditions for the validity of bilinear estimates, based on -L 2 norms in space and time, of derivatives of products of solutions.. Also, we

exfension theorem invafiably fails to hold in fhese spaces (i.e., there always are isomedries between subspaces that admit no extension to the whole space).. some

Motivated by the so–called matrix A 2 conjecture, which asks for the exact growth of the matrix weighted norm of the Hilbert transform acting on vector functions in terms of the