ULTRASPHERICAL POLYNOMIALS
DARIUSZBURACZEWSKI,TERESAMARTINEZ,JOS
EL.TORREA,ANDROMANURBAN
Abstra t. We dene and investigate the Riesz transform asso iated to the dif-
ferential operator L
f() = f 00
() 2 otf 0
()+ 2
f(). We prove that it
an bedened as a prin ipal value, and that it isbounded on L p
([0;℄;dm
()),
dm
()=sin 2
d,forevery1<p<1andofweaktype(1,1).Thesamebound-
edness propertieshold for the maximaloperator of the trun ated operators. The
speedof onvergen eofthetrun atedoperatorsismeasuredintermsofthebound-
ednessinL p
(dm
),1<p<1andweaktype(1;1)oftheos illationand-variation
asso iated to them. Also, a multiplier theoremis proved toget the boundedness
ofthe onjugate fun tion studiedbyMu kenhoupt and Steinfor 1 <p <1asa
orollaryoftheresultsfortheRiesztransform. Moreover,wenda onditiononthe
weightvwhi hisne essaryand suÆ ientfortheexisten eofaweightusu hthat
theRiesztransformisboundedfromL p
(vdm
)intoL
p
(udm
):
1. Introdu tion
We onsider the ultraspheri al polynomials of type , with any > 0, P
n (x) de-
nedasthe oeÆ ientsintheexpansionofthegenerating fun tion(1 2x!+! 2
)
=
P
1
n=0
! n
P
n
(x)(see[17℄forfurtherdetails). ItisknownthatthesetfP
n
( os):n2Ng
isorthogonaland omplete inL 2
[0;℄ withrespe tto themeasure dm
():=sin 2
d:
Morepre isely,wehave
(1.1) Z
0 P
n
( os)P
m
( os)dm
()=Æ
n;m 2
1 2
() 2
(n+2)
(n+)n!
=Æ
n;m
=
n :
Itisalsoknownthat thefun tionsP
n
( os)areeigenfun tionsoftheoperatorL
(1.2) L
f()= f 00
() 2 otf 0
()+ 2
f();
withL
P
n
( os)=(n+) 2
P
n ( os).
In this paper we develop a general method for studying the properties of ertain
operators asso iated to the dierential operator L
, that appear similarly to the way
that lassi aloperatorsdowhen onsidering theLapla ianoperator 2
in[0;2℄. Given
anoperatorTweprodu eapartitionofT intoits\lo al"andits\global"partsa ording
totworegions: the\lo alregion",whereroughlyspeakingwe anusethattheLebesgue
measure d and dm
() are equivalent, and its omplementary, that we all \global
region"(see(3.2)and(3.3)). Thismethodwaswidelyused by manyauthors invarious
ontexts (see e.g. [4, 5, 9, 11, 14℄). The pro edure for the lo al partof the operators
2000Mathemati sSubje tClassi ation. 42C05,42C15.
Keywordsandphrases. Ultraspheri alpolynomials,Poissonintegral,Riesztransform,maximalop-
erator,weightedinequalities.
Theauthorswere partiallysupportedbyRTN Harmoni Analysisand Related Problems ontra t
HPRN-CT-2001-00273-HARP. Therst and fourth authors weresupported inpart byKBN grant1
P03A01826. These ondandthirdauthorswerepartiallysupportedbyBFMgrant2002-04013-C02-02.
onsistsin omparingthekerneloftheoperatorwiththekernelofthe lassi alanalogous
operatorinthetorus. Weshowthatthedieren ebetweenthemisboundedbya\good"
kernel,i.e.,akernelwhi hdenesaboundedoperatoronL p
(d)foreverypintherange
1 p 1. By a\heritage"theorem (Theorem 3.8), we provethat the lo al part of
the lassi al operator in the torus inherits the boundedness in L p
with respe t to the
Lebesguemeasure. Next, foroperators withkernelssupported in thelo al part(\lo al
operators"),weprove(seeLemma 3.9) that theboundedness withrespe tto Lebesgue
measureisequivalenttotheboundednesswithrespe ttodm
(). Thereforewe on lude
that thelo alpartofT is bounded in L p
(dm
):Intheglobalpartwejust usethat the
kerneloftheoperatorT isboundedbyapositiveandni ekernel.
InthispaperweapplytheabovemethodinordertostudytheRiesztransform(The-
orem2.14),themaximaloperator asso iatedtoit(Theorem 7.1)anditsos illationand
variation(Theorem 8.2). We alsoprovesomemultiplier theorem in order to showthat
theresults in [10℄ anbe obtainedfrom ours,see se tion 6. Wewould liketo thankS.
Medaforsomehelpful suggestions on erningthistheorem.
Forevery\ni e"fun tionf dened on[0;℄weasso iateitsultraspheri alexpansion
(1.3) f() 1
X
n=0 a
n P
n
( os); where a
n
=
n Z
0 f()P
n
( os)dm
():
Inthepaper[10℄ Mu kenhouptandStein studied ultraspheri alexpansionsfromahar-
moni analysis point of view. Forany fun tion asin (1.3)they dene the \harmoni "
extensionoff as
(1.4) f(r;)=
1
X
n=0 a
n r
n
P
n ( os)
andthe\ onjugateharmoni "seriesas
(1.5)
e
f(r;)= 1
X
n=1 2
n+2
a
n r
n
sinP
+1
n 1 ( os):
Then,bothfun tions satisfyCau hy{Riemannequations:
r
((rsin) 2
e
f) = r
2 1
(sin) 2
f;
((rsin) 2
e
f) = r 2+1
(sin) 2
r f:
Oneofthemainresultsin[10℄isthatiff 2L p
(dm
)for1<p<1,thenthereexists
theboundaryfun tion
~
f()of e
f(r;)su hthat
~
f()=lim
r!1 e
f(r;) in L p
(dm
) norm
andk
~
fk
L p
(dm)
A
p kfk
L p
(dm) .
~
f()is alledthe onjugatefun tion off. Con erning
the asep=1,theyshowthatthe\ onjugateharmoni "fun tionf()!
~
f(r;)satises
an appropriate substitute relation (see Corollary 2, page 43, in [10℄). As far as we
understand, this result doesnot lead in an obvious wayto the weak type (1;1) of the
onjugatefun tion
~
f().
The onne tion between the results in [10℄ and the ones ontained in this paper is
as follows. In se tion 6 we prove a multiplier theorem that allows us to obtain the
boundedness properties in L p
forp in the range1<p<1, of the onjugatefun tion
~
f()as a orollaryofourresultsfortheRiesz transformdened spe trallyin (1.7),and
a posteriori as a prin ipal value operator. Although we obtain indeed the weak type
f()!
~
f()ofMu khenhouptand Stein,whi hisdened onlyforf 2L p
,satises also
aweaktype(1;1) inequality.
Itiseasyto he kthatL
isformallyself-adjointonthespa eL 2
(dm
). Ouroperator
L
fa torizesasL
f() =(
+
2
)f(),see se tion 2,where
=
+2 ot is
formallyadjointto
;i.e.,h
f;gi
L 2
(m
)
= hf;
gi
L 2
(m
)
. Forafun tionf asin(1.3)
wedeneitsPoissonintegralasso iatedtoouroperatorL
asfollows
(1.6) Pf(e
t
;)=e t
p
L
f()= 1
X
n=0 a
n e
t(n+)
P
n ( os):
Wepro eednowtodeneoneofthemainobje tswearegoingtoinvestigateinthispaper,
theRiesz transform. Let us allpolynomialfun tion to anynitelinear ombination of
ultraspheri al polynomials,that is, any f of the form f = P
N
n=0 a
n P
n
. We dene the
Riesz transformof any polynomialfun tion f byspe tralte hniques,asit issuggested
in[15℄,asfollows:
R
f() =
(L
)
1=2
f()=
N
X
n=0 a
n
n+ P
n ( os)
!
(1.7)
= 2
N
X
n=1 a
n
n+ sinP
+1
n 1 ( os):
Theaboveequationfollowsfromtheformula
t P
n
(t)=2P
+1
n 1
(t), see[17℄.
It turns out that if for a given fun tion f and its Poisson integral Pf, we dene
analogouslyto(1.5)the onjugatePoissonintegral
R
f(r;)= 2
N
X
n=1 a
n
n+ r
n+
sinP
+1
n 1 ( os):
ThefollowingCau hy{Riemannequationsaresatised
Pf(e
t
;)=
t R
f(e
t
;);
t Pf(e
t
;)+ 2
Z
1
t
Pf(e s
;)ds=
R
f(e
t
;):
(1.8)
Ourrstmain result,Theorem2.13,saysthatinfa t ourRiesztransformonpolyno-
mialfun tions isgivenbyaprin ipalvalue,i.e.,foranyf apolynomialfun tion,
R
f()=p:v:
Z
0 R
(;)f()dm
()=lim
"!0 Z
j j>"
R
(;)f()dm
()a.e:
Also, we provethat theRiesz transform R
is bounded on L p
(dm
) and of weak type
(1,1)(seeTheorem2.14). Asa orollaryweobtainthatR
f(e
t
;)tendstoR
f in L
p
normfor1<p<1,ast goesto 0.
These ondobje tweareinterestedinisamaximalfun tion
R
f():=sup
">0
Z
j j>"
R
(;)f()dm
()
:
Theorem7.1saysthatR
isboundedonL p
(dm
)for1<p<1;andofweaktype(1,1).
Asa onsequen eofthis result,wegetthattheRiesztransformisaprin ipalvaluealso
for fun tions in L 1
(dm
) (Theorem 7.2). Given this, our next aim is measuring the
speed of onvergen e, something whi h has lassi ally been done by means of ertain
use the so alled os illation operator and -variation operator, and obtain that they
bothare bounded in L p
(dm
) for1<p<1;and that they satisfyaweak type(1;1)
inequalitywithrespe ttodm
(seese tion8forthepre isedenitionoftheoperatorsand
statementoftheresults). Asanintermediatestepintheproof ofthisresultsweobtain
the boundedness in L p
(d) for 1<p <1 and theweak type(1;1) for the os illation
and -variationasso iated to the lassi al onjugate fun tion in the torus. This result,
that webelieveofintrinsi interest, isobtainedby omparisonwith thesameoperators
fortheHilberttransformintheline.
Thelastpartofthispaperisdevotedtothefollowingproblem. Givenaxedoperator
T (inour asewe onsidertheRiesztransformR
)boundedinL p
(dm
)forsome1<p<
1;ndne essaryandsuÆ ient onditionsforaweightv in ordertohavetheexisten e
of anontrivialweight usu h that T be omes bounded from L p
(vdm
) into L p
(udm
):
Inthe aseof theRiesz transform onsidered in thispaperweprove(seeTheorem 9.3)
thatthe onditionisthat
Z
0 v(x)
1
p 1
dm
(x)<1:
Toobtainthisresult,we onsiderave tor-valuedextensionofR
totheoperatortaking
valuesintheBana hspa el p
. Thenweapplytheanalogousve tor-valuedversionofthe
te hniqueofsplittingtheoperatorintoitslo alandglobalparts,togettheboundedness
inL p
` p
oftheextension(seese tion3forthedenitions).
Guide to the paper. Thestru tureof thepaperisasfollows. Inse tion2westate
pre iselythenotationandallthene essarydenitions inorder tointrodu etheobje ts
ofourinterest. Inparti ular,theRiesztransformR
isdened. Inse tion3weintrodu e
a\generalma hinery"inordertohandlethe\lo al"and\global"partsoftheoperators.
Inse tion4westudythebehavioroftheRieszkernela ordingtothispartition,andin
se tion5weshowthattheRiesztransformR
isgivenasaprin ipalvalueforpolynomial
fun tions anditsboundednessproperties. Inse tion6weproveamultipliertheorem.
Se tion7isdevotedtostudythemaximalfun tionR
f()andprovethatthemaximal
operator f 7! R
f is bounded on every L p
(dm
), for 1 < p < 1 and of weak type
(1;1). Thenextse tion,se tion8dealswiththestudyoftheos illationand-variation
operators. Finally,inse tion9we onsiderweightedinequalitiesandanswerthequestion
mentionedabove(Theorem9.3).
2. The Riesz transform
2.1. Poissonkernel. Wearegoingtodes ribeheresomefundamentalpropertiesofour
PoissonkernelP. Byusing(1.4)and(1.6),observethat
(2.1) Pf(e
t
;)=e t
f(e t
;):
Theresults in [10℄ forf(r;) leadeasily to provethat thePoisson integralPf is given
byakernel,whi h anbe omputedexpli itly. Namely,forany2[0;℄wehave
(2.2) Pf(r;)=r
Z
0
P(r;;)f()dm
()
wherer=e t
and
(2.3) P(r;;)=
(1 r
2
) Z
0
sin 2 1
t
2
+1 dt:
Anothereasy onsequen eof (2.1)andTheorem2,se tion6,p.31in[10℄isthefollowing
theorem.
Theorem2.4. [Theorem 2,[10℄℄ Letf 2L p
(dm
);1p1. Then
(a) kPf(r;)k
L p
(dm
)
r
kfk
L p
(dm
)
;
(b) kPf(r;) fk
L p
(dm
)
!0asr!1;
( ) lim
r!1
Pf(r;)=f() almosteverywhere,
(d) ksup
r<1
jPf(r;)jk
L p
(dm
)
A
p kfk
L p
(dm
)
:
2.2. The Riesz Transform. Inthisse tion we omputethekernelof theRiesztrans-
form. Byusing(1.1),oneshowsthat
(2.5) kR
fk
L 2
(dm)
Ckfk
L 2
(dm) :
Withthesameproof,wegettheboundednessoftheoperatorforanyf 2L 2
(dm
)andso
thedenition(1.7)makessenseforeveryf 2L 2
(dm
). Observethatasimilarargument
givesthattheoperator(L
)
1=2
isalsowelldenedinL 2
(dm
)withtheparallelformula
to theoneabove. Inthefollowinglemmaweshowthat (L
)
1=2
is denedby akernel
almosteverywhere,andthattheRiesztransformhasanasso iatedkernelinthesenseof
Calderon-Zygmund. Thislaststatementmeans,forT alinearoperator,thatthereexists
afun tionk(;)su hthatforeveryf 2L 2
([0;℄;dm
)andoutsidethesupportoff,
Tf()= R
k(;)f()dm
().
Lemma2.6. Given f 2L 1
(dm
), for almost every2[0;℄,we havethat
(2.7) (L
)
1
2
f()= Z
0 W
(;)f()dm
(); where W
(;)= Z
1
0 r
1
P(r;;)dr:
Given f 2L 1
(dm
)and outsidethe supportof f,we havethat
(2.8) R
f()=
Z
0 R
(;)f()dm
(); where R
(;)= Z
1
0 r
1
P
(r;;)dr;
whereP
standsfor the partialderivative ofP with respe tto.
Proof. Toprovetherstformula,weusetheintegralformula
s a
= 1
(a) Z
1
0 e
ts
t a
dt
t
todenethenegativesquarerootofL
:
L 1=2
f() = 1
(1) Z
1
0 e
t p
L
f()t dt
t
= Z
1
0 e
t
f(e t
;)dt
= Z
1
0 Z
0 r
1
P(r;;)f()dm
()dr;
(2.9)
after applying the hange of variables r=e t
. We gethere(2.7) ifwe an hange the
order of integration in the formerintegral. Before we pro eed further let us introdu e
somenotation:
a = os os+sinsin ost= os( ) sinsin(1 ost);
(2.10)
D
r
= 1 2ra+r 2
:
Observe that a os( ), and that for negative a, D
r
1, therefore we always
have that D
r
1 os( ) 2
0. In parti ular, this implies that r
1
P(r;;)
1 2 +1
observethattheexpressionf(r;)= R
0
P(r;;)f()dm
() iswelldenedforalmost
every2[0;℄sin ebytheresultsin[10℄, namelyLemmas5.2 and5.3,pg. 28,wehave
thatjf(r;)j(jfj)
(), where
f
()= sup
h6=0;0+h
R
+h
jf()jdm
()
R
+h
dm
()
andthisfun tionisboundedin L p
(dm
)foreveryp2(1;1℄andofweaktype(1;1). In
parti ular, for f 2L 1
(dm
), it is nite foralmost every . Thus, we ansee that the
integrandin(2.9)belongstoL 1
(dm
dr)forf 2L 1
([0;℄;dm
).
On e we hange the order of integration in (2.9), to get the expression of R
(;)
we just have to be able to put the derivative
inside the integral. We prove it by
showingthatr
1
P
(r;;)f()2L 1
(dm
dr). Observethatj
aj2andthat then
jr
1
P
(r;;)j anbeboundedabovebya onstantforoutsidethesupportoff,what
givestheresult.
Remark2.11. One analsousetheCau hy{Riemannequations (1.8)toprovetheabove
lemma. Namely, inviewofequations
Pf(e
t
;)=
t R
f(e
t
;)andR
f(0;)=0,
weobtain
R
f(e
t
;)= Z
1
t
P
t
f()dt= Z
1
t Z
0 e
t
P
(e
t
;;)f()ddt:
Further,if does notbelongtothesupportoff,we anmake r=e t
andthenargueas
beforetojustifymaking r going to1toobtain (2.8).
Remark2.12. Observethatforanyboundedfun tionf,f
()kfk
L
1 isawelldened
fun tion,nitefor every2[0;℄. Thus,byrepeatingthe proofof (2.7), wehavethatit
holdsfor every 2[0;℄.
Next,letus stateherethemain results on erningtheRiesz transform,namelythat
itisaprin ipalvalue(Theorem2.13)anditsboundedness propertiesinL p
, 1<p<1,
andin weakL 1
(Theorem2.14). Wewillprovethemin se tion5.
Theorem 2.13. The Riesz transform R
is given as a prin ipal value on polynomial
fun tions, i.e.,for any polynomial fun tion f andfor almostevery 2[0;℄
R
f()=lim
"!0 Z
j j>"
R
(;)f()dm
():
Theorem 2.14. The Riesztransform R
isbounded on L p
(dm
), for 1<p<1 and
fromL 1
(dm
)intoL 1;1
(dm
).
Remark2.15. Givenp2(1;1)andf 2L p
(dm
)su hthatithasanasso iatedFourier
seriesexpansionf P
a
n P
n
( os), then
R
f()
X
2
a
n
n+ sinP
+1
n 1 ( os):
Indeed, onsider thelinear operators fromL p
(dm
) intoC given by
U
n
f =
2a
n
n+
;
V
n
f = ksinP
+1
n 1
( os)k 2
L 2
(dm) Z
0 R
f()sinP
+1
n 1
( os)dm
():
Let us note that U
n
is trivially bounded by (1.3) and Theorem 2.14, and that V
n is
boundedalsobyTheorem2.14 andthefa tthat sinP
+1
( os)arein L p
0
(dm
). Now,
observe that fsinP
+1
n 1
( os)g forman orthogonal systemin L 2
(dm
), and thatthere-
fore,for every polynomialfun tion f,U
n f =V
n
f. Sin epolynomial fun tionsaredense
inL p
(dm
)for everyp2(1;1), ourthesis holds.
3. General ma hinery
Westartthis se tionwithanappropriateforourpurposes overinglemma. First,we
onstru tafamilyofballsontheinterval[0;
2
℄. LetusxaverysmallÆ>0throughout
thepaper,anddeneasequen eof pointsandballsasfollows
0
=
2
;
i+1
=
i
1+Æ
; B
i
=
: 1
1+Æ
<
i
<1+Æ
;
Letus also onsider ballsB(
i
;Æ 0
)=f: (1+Æ 0
) 1
<
i
= <1+Æ 0
g. One an easily
extend(bysymmetrywith respe tto
2
)theabovefamilyto aset of balls overingthe
interval(0;),thatwillbedenotedin thesamewayfB
i g
1
i=0 .
Lemma3.1. Forthe olle tion of ballsfB
i g
1
i=0
denedabove wehave
(1) The olle tion fB
i g
1
i=0
of losedballs overs(0;).
(2) ThereexistsÆ
0
su hthatthe ballsB(
i
;Æ
0
)arepairwisedisjoint.
(3) Foreveryn1,if1+Æ
0
=(1+Æ) n
;thenthe olle tion fB(
i
;Æ
0
)g hasbounded
overlap.
(4) For every n1, if 1+Æ
0
=(1+Æ) n
; then there exist a onstant C depending
onlyon Æ,n,su hthatfor every measurable setEB(
i
;Æ
0
), wehavethat
1
C
2
i
jEjm
(E)C
2
i jEj:
Proof. By the symmetry of the onstru tion it is enough to prove this properties for
theballs with enter in [0;=2℄: Inorder to prove(2) itis enoughto takeÆ
0
satisfying
(1+Æ
0 )
2
<1+Æ. Infa t,wehave
(1+Æ
0 )
i+1
<
(1+Æ)
i+1
1+Æ
0
=
i
1+Æ
0 :
Weleavethedetails oftheremainingproofto thereader.
ConsideralinearoperatorT mappingthespa eofpolynomialfun tionsintothespa e
ofmeasurablefun tions on[0;℄,satisfyingthefollowingassumptions:
(a) T either extendsto abounded operator on L q
(d) for some1< q <1; or of
weaktype(1,1)withrespe ttod,
(b) thereexistsameasurablefun tionK,denedinthe omplementofthediagonal
in[0;℄[0;℄;su hthatforeveryfun tionf andalloutsidethesupportoff,
Tf()= Z
0
K(;)f()d;
( ) forall(;);6=,KveriesjK(;)jCj j 1
.
Let us denote by N
t
the region onstru ted in the following way: onsider the part of
theregionM
t
=f(;): 1
1+t
<
<1+tgunderthediagonalofthesquare[0;℄
2
with
negativeslope,anddeneN
t
tobethisregiontogetherwithitsre e tedwithrespe tto
thepoint(=2;=2). ItisnotdiÆ ultto seethat theregionsN
t1 andN
t2 , witht
1
=Æ
and1+t =(1+Æ) 3
havetheshapeshowninthenextpi ture.
6
-
2
2
=
(1+Æ ) 3
=
1+Æ
j
s o
6
=(1+Æ)
=(1+Æ) 3
Observethatwegetaregion ontainedinM
t
andsymmetri withrespe ttothepoint
(=2;=2). Thisregionwilldenethe\lo al"and \global"parts oftheoperators.
GivenanoperatorT satisfying(b),wedeneitsglobalandlo alpartsby
T
glob f()=
Z
0
K(;)(1 1
N
t
2
(;))f()d;
(3.2)
T
lo
f()=Tf() T
glob f():
(3.3)
Wewill need thefollowingtwolemmas, whose proofsare areformulationofthe proofs
ofLemmas3.2and3.3in [5℄usingonlypropertiesofthe overingfrom Lemma3.1.
Lemma 3.4. Let d be any positive measure in [0;℄, ff
j
gbe asequen e of fun tions
and dene f = P
j 1
B
j f
j
; where fB
j
: j 2 Ng is the olle tion of balls in Lemma 3.1.
Then,for all >0;
(3.5) f:jf()j>g X
j
f2B
j :jf
j
()j>=2g;
andfor any 1q<1
(3.6) kfk
L q
(d)
2
X
j Z
B
j jf
j ()j
q
d()
1=q
;
Proof. Observe that any point 2 [0;℄ belongs to at most two balls of the overing
fB
j
g. Then, the levelset f : jf()j > gis a subset of S
j f 2 B
j : jf
j
()j > =2g,
whi hgives(3.5). Toprove(3.6),just observethat
Z
0 jf()j
q
d() Z
0
(2max
j jf
j ()j1
B
j ())
q
d()2 q
Z
0 X
j jf
j ()j
q
1
B
j
()d():
Lemma3.7. LetT bealinearoperator mappingpolynomialfun tionsintothe spa eof
measurable fun tionson [0;℄ verifying (a). Given the overing fB
j g
1
, wedene the
operator
T 1
f()= X
j 1
Bj ()jT(1
B 0
j
f)()j; where B 0
j
=B(
j
;Æ 0
); 1+Æ 0
=(1+Æ) 2
:
Then, T 1
isalso bounded from L 1
(d) intoL 1;1
(d) or fromL q
(d) into L q
(d),1<
q<1; asthe asemightbe.
Proof. Toprovethat T 1
isofweaktype(1;1)withrespe tto difT is,observethat
jf2B
j : jT(1
B 0
j
f)()j>=2gj C
Z
B
j
jf()jd:
We nish if we apply Lemma 3.4 and the bounded overlapproperty from Lemma 3.1.
Theproofofstrongtype(q;q)isanalogous.
Thefollowingtheoremisoneofthemaintoolsintheproofsoftheresultsinthepaper.
Theorem3.8. Under the assumptions (a), (b) and ( )made on T above, the operator
T
lo
inherits fromT either L q
(d)-boundednessortheweaktype(1,1)withrespe ttothe
Lebesguemeasure, asthe asemight be.
Proof. Assumethat isin theballB
j
anddeneB 0
j
asinLemma 3.7. Then
T
lo
f()=Tf() T
glob f()
=T(f1
B 0
j
)()+T(f(1 1
B 0
j ))()
Z
0 (1 1
N
t
2
(;))K(;)f()d
=T(f1
B 0
j )()+
Z
0 (1
N
t
2
(;) 1
B 0
j
())K(;)f()d:
Bymultiplyingby1
Bj
andsummingoverj,wegetthatforany2[0;℄
jT
lo
f()jjT 1
(f)()j+T 2
(f)();
where T 1
is asin Lemma 3.7, and by that lemma itis bounded in L q
orof weak type
(1;1)as orresponds. Therefore,weonlyneedtoprovethatT 2
isbounded, where
T 2
(f)()= Z
0 X
j 1
Bj ()j1
Nt
2
(;) 1
B 0
j
()jjK(;)jjf()jd:
Thisoperatoris givenbythekernel
H(;)= X
j 1
Bj ()j1
Nt
2
(;) 1
B 0
j
()jjK(;)j;
whi hgivesaboundedoperator.Wewillstudythiskernelinea hofthesquares[0;=2℄
2
,
[0;=2℄[=2;℄,[=2;℄[0;=2℄and[=2;℄
2
. Intherstandfourthones,H(;)is
supportedin N
t2 nN
t1
. Indeed,letusassumerst(;)2[0;=2℄
2
. If(;)62N
t2 then
1
(1+Æ) 3
>
or
>(1+Æ) 3
:
Hen e, if 2 B
j
, then annot belong to B 0
j
and then j1
Nt
2
(;) 1
B 0
j
()j = 0. If
(;)2N
t1
and2B
j then
1
<
<1+Æ and
1
2
<
j
<(1+Æ) 2
;
whi himpliesthat2B 0
j andj1
N
t
2
(;) 1
Æ 0
B
j
()j=0. For(;)2[=2;℄
2
, weuse
that ea hterm 1
Bj ()j1
Nt
2
(;) 1
B 0
j
()j is symmetri with respe t to (=2;=2) and
wegetinthisregionthesameresult.
In theregion where (;) 2[0;=2℄[=2;℄, one an see bysimilar argumentsas
beforethat thesupport ofH(;)is ontainedin N
t1
. Sin ein that regionthereexists
a onstant " su h that j j >", there jK(;)j C
"
and by thenite overlapping
propertyfromLemma3.1,H(;)C1
N
t
1
(;). Bysymmetry,wegetthesamebound
intheregion[=2;℄[0;=2℄.
Sin ethepartoftheoperatorintheregions[0;=2℄[=2;℄and[=2;℄[0;=2℄is
triviallybounded inL p
(d) foreveryp(1p1),to provethat T 2
alsosatisesthis
property,itisenoughtoprovethat
sup
2[0;=2℄
Z
=2
0 1
Nt
2 nNt
1 (;)
j j
d<1; sup
2[0;=2℄
Z
=2
0 1
Nt
2 nNt
1 (;)
j j
d<1:
Letusprovetherstinequality,sin ethese ondonefollowsinanidenti alway. Tothis
end,write:
sup
Z
0 1
Nt
2 nNt
1 (;)
j j
dsup
Z
1+t
1
1+t
2 1
d+sup
Z
(1+t2)
(1+t1) 1
d
=log 1
1
1+t
1
1 1
1+t2 +log
1+t
2
1+t
1
<1:
Similarargumentsas theones shownintheaboveproofleadtothefollowinglemma.
Lemma 3.9. Assume that an operator S dened on polynomial fun tionssatises (b),
with a kernel supported in N
t2
. Then, strong type (p,p) for Lebesgue measure and m
measureareequivalent. Thesame holdsfor weak type(p,p), 1p1.
Proof. One an see that jSf()j P
j 1
B
j jS(f1
~
B
j
)()j, where B
j
are asin Lemma 3.1
and
~
B
j
=B(
j
;
~
Æ),for1+
~
Æ=(1+Æ) 4
,just pro eeding asin theproofofTheorem3.8.
Then, by using Lemma 3.4, and parts (3) and (4) from Lemma 3.1, one aneasily get
theresult.
In se tions 7 and 9 wewill have to onsider linear operators taking valuesin some
Bana h spa es. We denote by L p
B
() = L p
B
(;[0;℄), p < 1, the Bo hner-Lebesgue
spa e onsistingofallB-valued(strongly)measurablefun tions f denedon[0;℄su h
that kfk p
L p
B ()
= R
0 kf()k
p
B
d() < 1. For p =1 wewrite L 1
B
for the spa e of all
f su h thatkfk
L 1
B
=supesskf()k
B
<1. Similarly,thespa e L p;1
B
()=weak-L p
B ()
is formed by all B-valued fun tions f su h that kfk
L p;1
B ()
= sup
t>0
t(f 2 [0;℄ :
kf()k
B
>tg) 1=p
<1.
We note here that the \heritage" Theorem 3.8 remains valid in the ve tor valued
ase with the same proof. More pre isely, given B
1
; B
2
Bana h spa es, let T be a
linearoperator dened in thespa e ofpolynomialfun tions with B
1
-valued oeÆ ients
andtakingvaluesinthespa eofB
2
-valuedandstronglymeasurablefun tionson[0;℄;
satisfying onditions(a'),(b')and( ')whi hare,infa t,(a),(b)and( )withappropriate
(a') T extends to a bounded operator either from L q
B
1
(d) into L q
B
2
(d) for some
1<q<1; orfromL 1
B
1
(d)into L 1;1
B2 (d),
(b') there exists a L(B
1
;B
2
)-valued measurable fun tion K, dened in the omple-
mentof thediagonalin [0;℄[0;℄; su hthat foreveryfun tion f 2L 1
B1 and
outsidethesupportoff,
Tf()= Z
0
K(;)f()d:
( ') thefun tionK satiseskK(;)kCj j 1
. forall6=.
In this\ve torvaluedlanguage" we an obtainLemmas 3.4 and 3.7, Theorem 3.8 and
Lemma 3.9in asimilar way (just hangingabsolutevaluesbynorms). Forfurther use,
letusstatetheve torvaluedversionofTheorem3.8asfollows.
Theorem3.10. If T isan operator satisfying(a'), (b')and( '). Dene T
lo andT
glob
as in 3.3 and 3.2. Then T
lo
inherits from T either the L q
-boundedness or the weak
type (1,1) as the ase might be. Besides, the orresponding boundedness holds for both
Lebesgueandm
measures.
4. Estimates onthe Riesz kernel
4.1. The lo al part. In this se tion we study the behaviorof the kernel of theRiesz
transform,R
(;)relatedto thekernelofthe lassi al onjugatefun tionin thetorus,
C(;)= 1
2
1
tan(( )=2)
,when onsidered onthelo alpart. Wearegoingto provethat
thedieren e
jR
(;)(sin) 2
C
0
C(;)j;
where C
0
is appropriately hosenreal number, anbeestimated in thelo al region by
some\ni e"fun tionM(;)su hthatthelo aloperator R
0
M(;)f()disbounded
oneveryL p
(d); for1p1.
We will usethe notationfor a and D
r
introdu ed in (2.10), as well asthe following
ones
b =
a= sin os+ ossin ost= sin( ) ossin(1 ost);
r
= 1 2r os( )+r 2
;
r
= rsinsin:
Also,weshallwriteD,, insteadofD
1 ,
1 ,
1
. Asin[16℄wearegoingto makeuse
ofthefollowingsimpleidentity, whi hisvalidforeveryreal> 1
(4.1)
Z
1
0 r
(1 r 2
)
D +2
r
dr=
1
(+1)D +1
:
BytheargumentsintheproofofLemma2.6,(4.1)and(2.3),wemaysimplifythekernel
oftheRiesztransformasfollows
(4.2) R
(;)= 2
Z
0 bsin
2 1
t
D
+1 dt:
Dene
(4.3) N(;)=1+log
+
sinsin
1 os( )
:
Lemma4.4. There exist onstants C
0
andC>0su hthat in the lo al region we have
the following estimate
jR
(;)(sin) 2
C
0
C(;)jCM(;);
whereM(;)=N(;)(sin) 1
.
Proof. Observe that R
( ; ) = R
(;), C( ; ) = C(;) and
M( ; )=M(;). Therefore,ifwegetthe omparisonintheregion(;)2[0;=
2℄[0;℄wegetthesameinequalityintheregion[=2;℄[0;℄. Thus,wewillrestri t
ourselvestoprovetheinequalityfor(;)2[0;=2℄[0;℄. Inthelo alregionforthis
rangeof (;), weget thefollowingestimates, that will beused throughout theproof:
thereexistsa onstantC su h that
1
C
sinsinCsin; jsin( )jCsin:
Theproofoftheseinequalitiesistrivial,althoughtheargumentdierswhen2[0;=2℄
thanwhen2[=2;℄. Wewillalsohandleintegralsoftheform
I
= Z
2
0 t
(+
t 2
2 )
dt:
Byusingthe hangeofvariablest= 1
2
1
2
u,andtheformerinequalities,forany>0,
if=2+1and =+1,theaboveintegral anbeestimatedasfollows
I 2+1
+1
C(sinsin) (+1)
N(;):
Firstwegetrid ofapartofthekernel(4.2),byobservingthat
(sin)
2
2
Z
=2 b(sint)
2 1
D
+1 dt
C(sin) 2+1
Z
=2 (sint)
2 1
+1
t 2+2
dt C
sin
Next,letuspro eedwiththe omparisonof thekernelsin severalsteps.
Step1. Dene
R 1
(;)= 2
Z
=2
0 bt
2 1
D
+1 dt:
Thenusingthat jsin 2 1
t t 2 1
j=O(t 2+1
),weobtain
jR
(;) R 1
(;)j(sin) 2
C(sin) 2+1
I 2+1
+1
CM(;):
Step2. Dene
R 2
(;)= 2
Z
=2
0 bt
2 1
D
+1 dt;
where
D=2(1 os( )+sinsin t
2
2 ):
ThenbytheTaylorexpansion,fort2[0;℄.
1
D
+1 1
D
+1
C
sinsint 4
D
+2 :
Thus,wehave
jR 1
(;) R 2
(;)j(sin) 2
(sin) 2+3
I 2+3
+2
CM(;):
Step3. Dene
R 3
(;)= 2
Z
=2
0
sin( )t 2 1
+1
dt:
Then
jR 2
(;) R 3
(;)j(sin) 2
C(sin) 2+1
I 2+1
+1
CM(;):
Step4. Compute:
R 3
(;)(sin) 2
= (sin) 2
2
Z
2
0
sin( )t 2 1
D
+1
dt
= 2
sin( )(sin) 2
1
C Z
1
2
1
2
1
2 u
2 1
(1+u 2
)
+1 du
= C
0
sin
sin
C(;)+I;
where
jIjC
sin( )
Z
1
2
1
2
1
2 1
u 3
duCjsin( )j
1
C
sin
;
andnallynoti e
sin
sin
C(;) C(;)
C
sin :
Lemma4.5. The lo al operator denedfor all f 2L 1
(d) and2[0;℄as
Mf()= Z
0
M(;)f()d;
whereM(;)isdenedinLemma4.4isboundedon L p
(d) for 1p1.
Proof. Toprovethelemmaitisenoughto he kthefollowing onditions,for =(1+Æ) 3
sup
2[0;℄
Z
M(;)d<1; sup
2[0;℄
Z
M(;)d<1:
Bythe hangeofvariablesx=,wehavethat
sup
2[0;℄
Z
N(;)
sin
dC
1+ sup
2[0;℄
Z
log
+
( ) 2
d
C+C Z
1
log
+ x
(1 x) 2
dx
x
<1:
4.2. The global part. Using the symmetry of the Riesz kernel R
(;) = R
(
; ), wemay restri t ourselvesto onsider the global partof theRiesz transform
onlyfor2[0;=2℄. Observe(seethepi tureinpage8)thatfor2[0;=2℄,we angive
abound aboveforR
;glob
withthree integralsasfollows,where =(1+Æ) 3
:
jR
;glob
f()j Z
0 jR
(;)j(1 1
Nt
2
(;))jf()jdm
() (4.6)
= Z
f> g +
Z
f
>> g +
Z
[=2;℄
=I+II+III:
For the last integral, observe that there exists a " > 0 su h that in the interse tion
[0;=2℄[=2;℄ withtheglobalregion, j j>". Therefore, jR
(;)jC
"
as an
beeasilyseenfrom(4.2),andweget
III C
"
Z
[=2;℄
jf()jdm
()C
Æ kfk
L 1
(dm
)
;
thatis, thispartisboundedin allL p
(dm
),1p1. Letusnowhandletheintegral
I.If> and,2[0;=2℄,thensin( )sinandjbjsin( ). Wewilluseas
wellthatforandinthisregion,D=1 a=1 os( )C( ) 2
Csin( ) 2
.
Hen e,inthispartweobtainthat
jR
(;)j
Csin( )
(sin ( )) 2+2
= C
(sin) 2+1
:
Wewillstatetheboundednessofthisoperator inthefollowinglemma.
Lemma4.7. The operator
M 1
f()= Z
0 M
1
(;)f()dm
(); whereM 1
(;)= 1
(sin) 2+1
1
f> g\[0;=2℄
2(;
)
for >1,is of weak type(1,1) and strong type(p;p) for 1<p<1 with respe t tothe
measuredm
.
Proof. Firstweprovetheweaktype. Sin etheoperatorislinear,we anrestri tourselves
to provethe weak type (1;1) for positive fun tions with kfk
L 1
(dm
)
=1. In this ase,
jM 1
f()j 1
(sin) 2+1
. Takeany2. Then,
m
(f: M 1
f()>g)m
([0;℄) C
:
For>2,observethat 1
(sin) 2+1
>ifandonly if<ar sin (1=) 1=(2+1)
,sin ein
thatregionsinhasawelldenedin reasinginverse. Also,inthisregion(=2 "),
thereexistsC su hthat0<C< os. Therefore,
m
(f: M 1
f()>g) = m
(f: <ar sin (1=) 1=(2+1)
g)
1
C Z
ar sin (1=) 1=(2+1)
0
os(sin) 2
d= 1
C
Nowwehavetoprovethat M 1
is bounded onL 1
(dm
),andthentheresultwillfollow
fromMar inkiewi zinterpolationtheorem. Thisboundednessholds,sin e
sup
0=2 Z
2
0
(sin) 2
(sin) 2+1
1
f> g
(;)d<1:
To giveabound for II, observethat in theregion where > and , 2[0;=2℄,
wehavethat bCsin( )andalsojsin( )jsin. Sin ehereitalsoholdsthat
DCsin( ) 2
weobtain
jR
(;)j C
(sin) 2+1
:
Lemma4.8. The operator
M 2
f()= Z
0 M
2
(;)f()dm
(); whereM 2
(;)= 1
(sin) 2+1
1
f> g\[0;=2℄
2(;
)
for >1 isof weak type (1;1) and strong type (p;p) for 1<p<1 with respe t tothe
measuredm
.
Proof. Toproveweaktype(1,1)weargueasin thepreviouslemma,sin ein thisregion
sinCsin. These ondpartholds,be auseM 2
istheadjointofM 1
inL 2
(dm
).
5. Proofs of Theorems2.13 and2.14
5.1. Proof of Theorem 2.13. Letf beapolynomialfun tion. Then by thespe tral
denition ofR
(1.7)andRemark 2.12,we anwrite
R
f()=(
(L
)
1=2
)f()=(
[(L
)
1=2
f℄)()=
Z
0 W
(;)f()d
for every 2 [0;℄, sin e the integral exists for every 2 [0;℄ (by Lemma 2.6 and
Remark2.12).
Next step is adding and subtra ting the fun tion W(;) = 1
logjsin(( )=2)j,
thekernelof theoperator(
2
)
1
2
in thetorus. LetC(;)bethekernelofthe lassi al
onjugate fun tion in thetorus. Sin e W(;) is anintegrable fun tion in , wehave
that R
0
W(;)f()dm
() iswelldened forevery, andwe anwrite
R
f() =
Z
0 (W
(;)sin 2
C
0
W(;))f()d
(5.1)
+C
0
Z
0
W(;)f()d:
These ond termisaprin ipalvalue,sin eit isthe onjugatefun tion in thetorus. To
handle the rst term, observethat for 6=,
W
(;)= R
(;) (sin e for 6= ,
jr
1
P
(r;;)jCr
=(1 os( ) 2
)
+1
andthis fun tion is dr-integrable). Thus,
foralmostevery2[0;℄,wehavethat
(W(;)sin 2
C
0
W(;))=R
(;)sin 2
C
0 C(;):
Now,byusingLemma4.4and theargumentsofsubse tion4.2, wehavethat
jR
(;)sin 2
C
0 C(;)j
M(;)1
Nt
2
(;)+M 1
(;)(sin) 2
+M 2
(;)(sin) 2
+C1
N
t
2 (;);
wherethe onstantC omes fromtheboundedness forIII in(4.6)andtheboundedness
ofC(;)intheglobalregion. BytheproofsofLemmas4.5,4.7 and4.8, theright-hand
fun tionisd-integrable. Thus,byLebesgue'sdominated onvergen etheorem,therst
term in (5.1) is also the limitof the trun ated integrals, and thereforewe getthat for
almostevery2[0;℄,
R
f()=
= lim
"!0 Z
j j>"
(R
(;)sin 2
C
0
C(;))f()d+C
0 lim
"!0 Z
j j>"
C(;)f()d
= lim
"!0 Z
j j>"
R
(;)f()dm
():
5.2. ProofofTheorem2.14. Letuspointoutthatitisenoughtoprovethebounded-
nessinL p
orinweakL 1
foradensesub lassoffun tionsinthosespa es. Inthisse tion,
we onsider polynomialfun tionsf 2L p
forany1p<1,whi hformadensesubset
of L p
and for whi h Theorem 2.13holds. Next,wedene theglobal and lo al parts of
theRiesztransformR
a ordingto(3.2)and(3.3). Theboundednessoftheglobalpart
wasobtainedin subse tion4.2.
Considerthelo alpartR
;lo
. Wewrite
(5.2) R
;lo
=(R
;lo C
0 C
lo )+C
0 C
lo
;
where C denotesthe lassi al onjugatefun tion in thetorus. Letusobservethat ifan
operator T is given by a prin ipal value, then its lo al part is also a prin ipal value.
Thus, by Lemmas 4.4 and 4.5it followsthat R
;lo C
0 C
lo
is bounded on L p
(d) for
1p1. Letusre allthatby lassi alresultsandTheorem3.8,C
lo
isalsoboundedin
L p
(d),1<p<1andofweaktype(1;1). Thus,R
;lo
isboundedinL p
(d)1<p<1
andof weaktype(1;1),and byLemma 3.9it followsthatthesamestatementholds for
L p
(dm
())for1<p<1andfortheweaktype(1,1)withrespe ttodm
.
6. Comparison of the results forMu kenhoupt-Stein's onjugate
fun tion andthe Riesz transform
Theaimof thisse tionis givinganargumenttoderivetheboundednessin L p
(dm
)
for p 2 (1;1) obtained by Mu kenhoupt and Stein in [10℄ for the onjugate fun tion
dened there,asa orollaryof theboundednessobtainedin Theorem2.14fortheRiesz
transform in the same range of p's. First step is observing that Mu kenhoupt-Stein's
onjugate fun tion is dened, for a polynomial fun tion f() = P
N
n=0 a
n P
n
( os), as
~
f() = lim
r!1
~
f(r;). By using (1.5) and the expression for the Riesz transform of f
(1.7), we get that
~
f()=R
(T
m
f)(), where T
m
isthe multiplier operator dened for
polynomialfun tions as
T
m f()=
N
X
n=0 a
n m
n P
n
( os); m
0
=0; m
n
=
n+
n+2
for n1:
Therefore,ifweobtainthatthemultiplierT
m
isboundedinL p
(dm
)foreveryp2(1;1),
wewouldgettriviallytheboundednessofthe onjugatefun tioninL p
(sin epolynomial
fun tionsaredensein allL p
(dm
)). Thismultiplierfallsintothes opeofthemultiplier
theorem proved by Mu kenhoupt and Stein in [10℄, Theorem 10 in page 71. But this
theoremgivestheboundednessofT
m
onlyinarestri tedrangeofp's,soweneedanother
multipliertheorem. Wewillrestri tourselvestostatethemultipliertheoremjustforT
m ,
sin eitisnotourinterestat thispointtoproveageneraltheorem.
Lemma6.1. The multiplieroperator T
m
isboundedinL p
(dm
)for every p2(1;1).
Proof. Dene,forp2(1;1)thespa e
X
p
=(I
0 )L
p
(dm
);
where
0
is the ontinuousproje tion(orthogonal in L 2
(dm
)) onto the subspa egen-
erated by P
0
( os), whi h is a onstant (see[17℄). X
p
is a losed subspa e of L p
, the
subspa eoffun tions ofnullmean,andin parti ularaBana hspa eitself, wherepoly-
nomial fun tions of the form f() = P
K
k =1 a
k P
k
( os) are dense. Now, observe that
foranypolynomialfun tionf = P
K
k =0 a
k P
k
( os),T
m
f()=T
m (f a
0 P
0
( os)),and
thereforeitisenoughtogettheboundednessofT forthepolynomialfun tionsin X .
Let us onsider rst the ase p 2. To this end, let us observe that by spe tral
te hniques,wehavethat
T
m f()=
N
X
n=1 a
n
1
1+
n+
P
n
( os)=
1
1+L 1=2
f()=F(L 1=2
)f();
whereF(z)= 1
1+z
. Now,observethatthespe trumofL 1=2
inX
p
forp>2is
(L 1=2
)=
1
n+
: n1
$
z: jzj<
1
;
and that the fun tion F is holomorphi and admits a power series expansion F(z) =
P
1
k =0 a
k z
k
intheright-handset.Byusingwellknownresultsonoperatortheory,seefor
exampleTheoremVII.3.10in[3℄,weget
(6.2) T
m
=F(L 1=2
)=
1
X
k =0 a
k (L
1=2
)
k
:
Next step is giving a good bound for the norm of L 1=2
in X
p
. This is given by the
followinglemma
Lemma6.3. Foreveryp,2p<1,thereexists"
p
2(0;1)su hthatforeveryf 2X
p ,
kL 1=2
fk
L p
(dm)
C
+"
p kfk
L p
(dm) :
Letus postponethe proofof thislemma, in order to larifythereading ofthe proof
ofLemma6.1. By(6.2),wehavethat foranyf 2X
p
,2p<1,
kT
m fk
L p
(dm
)
1
X
k =0 ja
k jk(L
1=2
)
k
fk
L p
(dm
)
1
X
k =0 ja
k j
C
+"
p
k
kfk
L p
(dm)
Ckfk
L p
(dm)
;
sin etheseries onvergesabsolutelyforjzj<1=.
To prove the boundedness of T
m in X
p
for p 2 (1;2), we use the standard duality
argument.
Proofof Lemma 6.3. First,observethat foranyf 2X
2
polynomialfun tion,
ke t
p
L
fk 2
L 2
(dm)
=
K
X
k =1 a
k e
t(k +)
P
k ( os)
2
L 2
(dm)
e
2t(+1) K
X
k =1 ja
k j
2
=e
2t(+1)
kfk 2
L 2
(dm
)
:
Next, for any q > 2 and any fun tion in X
q
, by (a) in Theorem 2.4, we have that
ke t
p
L
fk
L q
(dm)
e t
kfk
L q
(dm)
. Takeanypsu hthat2<p<q,1=p="
p
=2+(1
"
p
)=q,thenbytheinterpolationtheorem,wehavethat
e t
p
L
Xp!L p
(dm)
C
e t
p
L
"p
X2!L 2
(dm)
e t
p
L
(1 "p)
Xq!L q
(dm)
Ce
t(+"p)
:
Finally,wegetthat
kL 1=2
k
X
p
!L p
(dm
)
Z
1
0
e t
p
L
X
p
!L p
(dm
)
dtC Z
1
0 e
t(+"
p )
dt= C
+"
p :
7. The maximal operator R
.
Denethefollowingmaximaloperators
R
f() = sup
"
jR
;"
f()j; where R
;"
f()= Z
j j>"
R
(;)f()d;
C
f() = sup
"
jC
"
f()j; where C
"
f()= Z
j j>"
C(;)f()d:
Themainresultof thisse tionistheboundednessof therstone. Theboundedness of
these ondoneisawellknownfa t,see[18℄.
Theorem7.1. R
is boundedonL p
(dm
)for 1<p<1, andof weak type (1,1) with
respe ttodm
.
Proof. TherststepissplittingR
asfollows,
R
f()=sup
"
jR
;"
f()jsup
"
Z
j j>"
jR
;lo
(;)(sin) 2
C
0 C
lo
(;)jjf()jd
+C
0 sup
"
Z
j j>"
C
lo
(;)f()d
+sup
"
Z
j j>"
jR
;glob
(;)jjf()jdm
():
The rst term is bounded aboveby the operator Mf(), as follows from Lemma 4.4.
Then,Lemma4.5givestheboundednessofthisterm. Thethirdfa tor anbeestimated
by the same arguments as in the proof of Theorem 2.14, see subse tion 4.2, namely,
gettinga parallel inequality to (4.6). Then, Lemmas 4.7 and 4.8 givethe boundedness
ofthis term. Thelo al partpartofthe lassi almaximaloperatorC
lo
isbounded asa
onsequen eofTheorem 3.10andtheboundedness ofC
lo
.
Asa orollaryoftheformerresult,by lassi alargumentswe anextendthedenition
oftheRiesztransformasaprin ipalvaluetoallfun tionsinL 1
(dm
).
Theorem7.2. Forany fun tion f 2L 1
(dm
)andalmost every 2[0;℄
R
f()=lim
"!0 Z
j j>"
R
(;)f()dm
():
8. Os illations
Bytheresultsinthepreviousse tions,weknowthatourRiesztransformisthelimitof
thetrun ated operatorsR
;"
almosteverywhere. Ingeneral,for everyfamily T =fT
"
g
su h that there exists Tf = lim
"!0 T
"
f, it is lassi al to measure the speed of this
onvergen ebymeansofexpressionsinvolvingdieren esofthetypejT
"
f T
"
0fj. Some
oftheoperatorsusedforthispurposearetheso alled os illationoperator,denedas
O(T)f(x)=
1
X
sup
t
i+1
"
i+1
<"
i
<t
i jT
"i+1
f(x) T
"i f(x)j
2
1=2
;
foraxedsequen eft
i
gde reasingtozero. Forin therange2<<1;wehavethe
so alled -variationoperator,
V
(T)f(x)=sup
f"ig
1
X
i=1 jT
"i+1
f(x) T
"i f(x)j
1=
;
where the supis takenoverall sequen es f"
i
gde reasing to zero. Classi alresults are
that when these operators are dened for the family of the trun ated operators of the
Hilbert transformor for thefamily of the Poisson integrals in R, they are bounded in
L p
(R ), 1<p<1andofweaktype(1;1)withrespe ttotheLebesguemeasure(see[2℄
and[7℄andthereferen estherein). Ouraim isprovingsimilarresultsfortheos illation
and -variation ofthefamily ofthetrun ated operatorsof theRiesz transform. Letus
all R
= fR
;"
g and C = fC
"
g to the families of the trun ated operators asso iated
to the Riesz transform and the lassi al onjugate fun tion, respe tively. First of all,
letus pointoutthat with theve torvaluedte hniques shown in [7℄, one anget easily
the boundedness of the os illationand -variationasso iated to the lassi al onjugate
fun tioninthetorusfromtheknownresultsfortheHilberttransformintheline. Letus
denote by H=fH
t
gthefamilyof thetrun atedoperators fortheHilbert transformin
R, thatis,
H
t f(x)=
Z
jx yj>t f(y)
x y dy:
Before pro eeding further, let us observe that for any family T = fT
"
g, O(T)f(x)
O 0
(T)f(x)foralmosteveryx,where
O 0
(T)f(x)=
1
X
i=1 sup
t
i+1
<st
i jT
ti+1
f(x) T
s f(x)j
2
1=2
:
Thus,itisenoughtoprovethedesiredboundednesspropertiesforthisoperatorinstead
of O. Denote by B = ` 2
L 1
(0;1)
the spa e of sequen es u = fu
i
g su h that for every
i, u
i
= u
i
(s)is afun tion in L 1
(0;1), with kuk
B
=kfku
i k
L 1gk
`
2. Then, if we put
U(T)f(x)=f(T
ti+1
f(x) T
s f(x))1
(t
i+1
;t
i
℄
(s)g,wehavethat
O 0
(T)f(x)=kf(T
ti+1
f(x) T
s f(x))1
(t
i+1
;t
i
℄ (s)gk
B
=kU(T)fk
B :
Also, let us all =
" = f"
i g : "
i
# 0
the set of sequen es de reasing to zero,
and B
= B
`
();
with in the range 2 < < 1; the spa e of bounded fun tions
v =v(")dened on ,whose valuesaresequen es in `
, v(")=fv
i
(")g, and su hthat
kvk
B
= sup
"2
kfv
i (")gk
`
. Then, if we put V(T)f(x) =fT
"i+1
f(x) T
"i
f(x)g, we
havethat
V
(T)f(x)=kfT
"i+1
f(x) T
"i f(x)gk
B
=kV(T)fk
B
:
Inthe asethat ea hT
t
hasan asso iatedkernelK
t
, thenU(T)andV(T)havealsoan
asso iatedkernel:
U(T)f(x) = f(T
ti+1
f(x) T
s f(x))1
(t
i+1
;t
i
℄ (s)g
= Z
(K
t
i+1
(x;y) K
s (x;y))1
(ti+1;ti℄
(s)
f(y)d(y);
and
V(T)f(x) = Z
K
"i+1
(x;y) K
"i (x;y)
f(y)d(y):
With all these preliminaries, it is lear that the boundedness of the os illation and -
variationoperators asso iatedto ertain family, namely, C, is equivalentto the bound-
ednessofU(C)andV(C). We anprovethisboundednessby omparisonwithU(H ) and
V(H ).
Proposition 8.1. Let us onsider f : [0;℄ ! R and extend it as 0 to the whole R.
Then, thereexistsa positive onstant C su hthat for every 2[0;℄ andp2[1;1℄ we
havethat
kU(C)f() 1
1
[0;℄
()U(H )f()k
B
Ckfk
L p
(d)
;
kV(C)f() 1
1
[0;℄
()V(H )f()k
B
Ckfk
L p
(d) :
Proof. WewillprovetheresultonlyforU,sin etheproofforV is ompletelyanalogous.
Inthis ase,wehavethat
U(C)f() 1
1
[0;℄
()U(H )f()
=
(R
ti+1
f() R
s f())1
(t
i+1
;t
i
℄ (s)
1
(H
ti+1
f() H
s f())1
(t
i+1
;t
i
℄ (s)
= Z
0 1
2
1
tan(( )=2)
1
( )=2
1
fti+1<j jsg
(j j)1
(ti+1;ti℄
(s)
f()d:
Noti ethat
1
tan(( )=2)
1
( )=2
1
fti+1<j jsg
(j j)1
(ti+1;ti℄
(s)
B
=
1
tan(( )=2)
1
( )=2
;
thereforebyMinkowski'sinequality,wehave
kU(C)f() 1
1
[0;℄
()U(H )f()k
B
C Z
0
1
tan(( )=2)
1
( )=2
jf()jd:
And now, for ; 2[0;℄, 2 [ ;℄ but both fun tions are odd. Therefore, we
onlyhaveto give abound for j1=tan(x=2) 1=(x=2)j forx 2[0;℄, orequivalently, for
j1=tanx 1=xj forx 2[0;=2℄. Inthis range, learlyj1=tanx 1=xjC, and we get
theresult.
Thefollowingtheoremstatesthemainresultsofthisse tion.
Theorem8.2. Forevery p2(1;1)thereexistsa onstantC su hthat
kO(R
)fk
L p
(dm
)
Ckfk
L p
(dm
)
; kV
(R
)fk
L p
(dm
)
Ckfk
L p
(dm
)
andalso
m
f: O(R
)f()>g
C
kfk
L 1
(dm
)
m
f: V
(R
)f()>g
C
kfk
L 1
(dm
)
Proof. Theproofofthistheorem anbedevelopedbyusingthesamete hniquesapplied
in theproofsofTheorems2.13and2.14,i.e.,splittingtheoperatorsintotheirlo al and
globalparts. Tothisend,deneforafamilyT =fT
t
gasabove,a tingonfun tionsover
[0;℄, itslo alandglobalpartsasT
lo
=fT
t;lo
gandT
glob
=fT
t;glob g.
Lemma8.3. Given afamily of operatorsT =fT
t
gon fun tionsdenedon [0;℄ su h
thatea hoperator T
t
isgiven by akernelK
t
,thenwe have
U(T)
lo
=U(T
lo
); U(T)
glob
=U(T
glob
) V(T)
lo
=V(T
lo
); V(T)
glob
=V(T
glob ):
Proof. Letusseetheprooffortheglobalpart,sin eforthelo alpartitfollowsfromits
denitionandthelinearityin T oftheoperators U andV:
U(T)
lo
=U(T) U(T)
glob
=U(T) U(T
glob
)=U(T T
glob
)=U(T
lo ):
Bydenition oftheglobalpartofanoperator,wehave
U(T)
glob
f() = Z
0
(L
ti+1
(;) L
s (;))1
(t
i+1
;t
i
℄ (s)
(1 1
Nt
2
(;))f()d;
= f(T
t
i+1
;glob
f() T
s;glob f())1
(ti+1;ti℄
(s)g=U(T
glob )f():
Asimilarproofholds forV.
Now,wewillprovethatthelo alpartofU(R
)andV(R
)dierfromthelo alpartof
U(C)andV(C),respe tively,inanoperatorwhi hisboundedinL p
(dm
)for1<p<1
andofweaktype(1;1).
Lemma 8.4. There exists apositive onstant C su hthat for every 2[0;℄ we have
that
kU(R
)
lo
f() C
0 U(C)
lo f()k
B
C Z
0
M(;)1
Nt
2
(;)jf()jd;
kV(R
)
lo
f() C
0 V(C)
lo f()k
B
C Z
0
M(;)1
Nt
2
(;)jf()jd
whereC
0
andM(;)areasin Lemma 4.4.
Proof. Sin e foreverytrun ated operatorweare awayfrom thediagonal, observethat
we anwrite
U(R
)
lo
f() C
0 U(C)
lo
f()=U(R
;lo
)f() C
0 U(R
lo )f()
=
(R
;ti+1;lo
f() R
;s;lo f())1
(ti+1;ti℄
(s)
C
0
(C
t
i+1
;lo
f() C
s;lo f())1
(ti+1;ti℄
(s)
= Z
0
[R
(;)(sin) 2
C
0 C(;)℄
1
fs<j jt
i+1 g
(j j)1
(t
i+1
;t
i
℄ (s)
1
Nt
2
(;)f()d:
Sin e
[R
(;)(sin) 2
C
0
C(;)℄1
fs<j jt
i+1 g
(j j)1
(t
i+1
;t
i
℄ (s)
B
=jR
(;)(sin) 2
C
0
C(;)j;
byMinkowski'sinequalityand Lemma4.4,wehavethat
kU(R
)
lo
f() C
0 U(C)
lo f()k
B
Z
0
M(;)1
N
t
2
(;)jf()jd:
ByProposition8.1,weknowtheboundednessofU(C)andV(C)inL p
(d)for1<p<
1 and weak type(1;1) with respe t to d. Now, one ansee that U(C)and V(C) are
ve torvaluedoperatorsasdenedinse tion3. Thus,bytheve torvaluedversionofthe
heritagetheorem,Theorem3.10,thelo alpartofbothoftheminheritstheboundedness
properties mentionedabovewithrespe tto d. ByLemmas 8.4 and4.5, this givesthe
boundedness ofthe lo al part ofU(R
) and V(R
)with respe tto d. Hen e,by the
ve tor valued version of Lemma 3.9, we obtain the boundedness of the lo al part of
U(R
)and V(R
)alsofor dm
. Inorder to proveTheorem 8.2, itis onlyleft toprove
thattheglobalpartsarealsoboundedinL p
(dm
). This anbeobtainedbyusingsimilar
argumentsasinLemma8.4tohandletheve torvaluedkernels,andthenpro eedingasin
theproofofTheorem2.14fortheglobalpart, on retelyobtainingtheparallelinequality
to4.6.
9. Weightedinequalities
Let us all weight to anystri tly positive measurable fun tion v. In this se tion we
will answerthe following question: nd ane essaryand suÆ ient onditionona given
weightv, fortheexisten eofaweightusu hthat theRiesztransformisboundedfrom
L p
(vdm
) intoL p
(udm
)for1<p<1. Wewill useawell knownte hniquebasedin
ideasbyRubiodeFran ia. Inparti ular,weshallneedthefollowingtheoremduetothis
author(see[12℄).
Theorem 9.1. Let (X;) be a measure spa e, B a Bana h spa e and T a sublinear
operator fromB intoL s
(X)su hthatthe following inequalityissatisedforsomes<p
andeverysequen eff
j gB
X
j jTf
j j
p
1=p
L s
(X)
C
p;s
X
j kf
j k
p
B
1=p
;
where C
p;s
isa onstant depending on pand s. Then there exists apositive fun tion u
su hthat u 1
2L s
p s
(X)and
Z
X
jTf()j p
u()d()
1=p
kfk
B :
ConsidertheextensionoftheRiesztransformR
totheoperatordenedforintegrable
fun tions f = ff
j g
j
taking values in ` p
, given by R
f = fR
f
j g
j
. For this extended
operator,wehavethefollowinglemma.
Lemma9.2. Letpbein therange1<p<1;then
R
: L
1
l p(m
)!L
1;1
l p
(m
)
Proof. Itisknownthatthe` p
-valuedextensionofthe lassi al onjugatefun tion inthe
torus,Cf =fCf
j g
j
;maps L 1
l
p(d)intoL 1;1
l p
(d) (see[1℄). Bytheve tor-valuedversion
oftheheritagetheorem,Theorem3.10,thelo alpartofthisextensionofCisalsoofweak
type (1;1). Therefore by using identity (5.2), Lemma 4.4 and Lemma 4.5 weget that
R
;lo
maps L 1
l
p(d)into L 1;1
l p
(d): Wehaveusedherethat theoperatorin Lemma 4.5
isgivenbyapositivekernelandthereforeit anbeextended to ` p
-valuedfun tions for
1p1. Anstraightforward` p
-valuedextensionofLemma3.9givestheboundedness
ofR
;lo
fromL 1
l p(m
)intoL 1;1
l p
(m
):
Asfortheglobalpartweobservethatsin eitismajorizedbyapositiveoperator(see