PII. S0161171201006548 http://ijmms.hindawi.com © Hindawi Publishing Corp.
ROUGH MARCINKIEWICZ INTEGRAL OPERATORS
HUSSAIN AL-QASSEM and AHMAD AL-SALMAN
(Received 16 January 2001)
Abstract.We study the Marcinkiewicz integral operator Mᏼf (x)=(−∞∞ ||y|≤2tf (x− ᏼ(y))(Ω(y)/|y|n−1) dy|2dt/22t)1/2, whereᏼis a polynomial mapping fromRnintoRd
andΩis a homogeneous function of degree zero onRnwith mean value zero over the
unit sphereSn−1. We prove anLpboundedness result ofMᏼfor roughΩ. 2000 Mathematics Subject Classification. 42B20, 42B15, 42B25.
1. Introduction. LetRn,n≥2 be then-dimensional Euclidean space andSn−1be the unit sphere inRn equipped with the induced Lebesgue measure. Consider the Marcinkiewics integral operator
µf (x)=
∞
−∞
Ft(x) 2dt
22t
1/2
, (1.1)
where
Ft(x)=
|x−y|≤2tf (y)
Ω(x−y)
|x−y|n−1dy, (1.2)
andΩis a homogeneous function of degree zero which has the following properties:
Ω∈L1Sn−1,
Sn−1Ω
ydσy=0. (1.3)
WhenΩ∈Lipα(Sn−1),(0< α≤1), Stein proved theLpboundedness ofµ(f )for all 1< p≤2. Subsequently, Benedek, Calderón, and Panzone proved theLpboundedness ofµ(f )for all 1< p <∞under the conditionΩ∈C1(Sn−1)(see [2]).
The authors of [3] were able to prove the following result for the more general class of operators
µPf (x)=
∞
−∞
FP ,t(x) 2dt
22t
1/2
, (1.4)
where
FP ,t(x)=
|y|≤2tf
x−P (|y|)y Ω(y)
|y|n−1dy (1.5)
andP is a real-valued polynomial onRand satisfiesP (0)=0.
Theorem1.1(see [3]). Letα >0, andΩ∈Vα(n). Then the operatorµP is bounded
In [1], Al-Salman and Pan studied the singular integral operator
TΩ,ᏼf (x)=p.v.
Rnf
x−ᏼ(y)Ω
y
|y|n dy, (1.6)
whereᏼ=(P1, . . . , Pd):Rn→Rd is a polynomial mapping,d≥1,n≥2. The authors of [1] proved thatTΩ,ᏼis bounded inLp(Rd)whenever(2+2α)/(1+2α) < p <2+2α
andΩ∈Wα(n). HereWα(n)is a subspace ofL1(Sn−1)and its definition as well as the definition ofVα(n)will be reviewed inSection 2. It was shown in [1] thatWα(n)= Vα(n),if n=2 and it is a proper subspace ofVα(n)ifn≥3.
Our purpose in this paper is to study theLpboundedness of the class of operators
Mᏼf (x)=
∞
−∞
Fᏼ,t(x)2dt 22t
1/2
, (1.7)
where
Fᏼ,t(x)=
|y|≤2tf
x−ᏼ(y) Ω(y)
|y|n−1dy. (1.8) Our main result in this paper is the following theorem.
Theorem 1.2. Letα >0, and Ω∈Wα(n). Then the operator Mᏼ is bounded in Lp(Rd)for(2α+2)/(2α+1) < p <2+2α. The bound ofM
ᏼf is independent of the coefficients of{Pj}.
By [1, Theorem 3.1] andTheorem 1.2we have the following corollary.
Corollary 1.3. Letα >0, Ω∈Vα(2) andᏼ:R2→Rd. Then Mᏼ is bounded in Lp(Rd)for(2α+2)/(2α+1) < p <2+2α. The bound ofM
ᏼ is independent of the coefficients of{Pj}.
2. Preparation. We start this section by recalling the following definition from [1].
Definition2.1. Forα >0,N≥1, letᐂ(n, N) = Nm=1ᐂ(n, m)and letWα(N, n)be the subspace ofL1(Sn−1)defined by
Wα(N, n)=
Ω∈L1Sn−1:
Sn−1Ω
ydσy=0, Mα(N, n) <∞
, (2.1)
where
Mα(N, n)
=max
Sn−1 Ω
y
log 1 |P (y)|
1+α
dσy:P∈ᐂ(n, N) withP =1
.
(2.2)
Forα >0, we defineWα(n)to be
Wα(n)= ∞
N=1
Wα(N, n). (2.3)
Hereᐂ(n, m)is the space of all real-valued homogeneous polynomials onRnwith degree equal tomand with norm·defined by
|α|=m aαyα
=
|α|=m
aα. (2.4)
Now we need to recall the following results.
Lemma2.2(see van der Corput [7]). Supposeφandψare real-valued and smooth
in(a, b), and that|φ(k)(t)| ≥1for allt∈(a, b). Then the inequality
abe−iλφ(t)ψ(t) dt≤Ck|λ|−1/k
|ψ(b)|+
b
a
ψ(t)dt
, (2.5)
holds when
(i) k≥2, or
(ii) k=1andφis monotonic.
The boundCkis independent ofa,b,φ, andλ.
Lemma2.3(see [7]). Letᏼ=(P1, . . . , Pd)be a polynomial mapping fromRnintoRd.
Letdeg(ᏼ)=max1≤j≤ddeg(Pj). SupposeΩ∈L1(Sn−1)and
µΩ,ᏼf (x)=sup
h>0
1
hn
|y|<hf
x−ᏼ(y)Ωydy. (2.6)
Then for every1< p≤ ∞, there exists a constantCp>0which is independent ofΩand
the coefficients of{Pj}such that
µΩ,ᏼfp≤CpΩL1(Sn−1)fp (2.7)
for everyf∈Lp(Rd).
To each polynomial mappingᏼ=(P1, . . . , Pd):Rn→Rd with
degᏼ=max
1≤j≤ddegPj=N, d≥1, n≥2, (2.8)
we define a family of measures
ϑtl, λlt:l=0,1, . . . , N, t∈R
(2.9)
as follows.
For 1≤j≤d, 0≤l≤N letPj=
|α|≤NCjαyα and let Ql=(Ql1, . . . , Qld) where Ql
j=
|α|≤lCjαyα.
Now for 0≤l≤Nandt∈R, letϑl
t andλlt be the measures defined in the Fourier transform side by
ϑltˆ
(ξ)=
|y|≤2te
−2π iξ·Ql(y)Ωy |y|n−1
dy 2t ,
λltˆ
(ξ)=
|y|≤2te
−2π iξ·Ql(y)Ωy |y|n−1
dy 2t .
The maximal functions(ϑl)∗defined by
ϑl∗(f )(x)=sup t∈R
λl
t∗f (x), (2.11)
forl=0,1, . . . , N.
For later purposes, we need the following definition.
Definition 2.4. For each 1≤l≤N, letNl= |{α∈Nn:|α| =l}|and let {α∈
Nn :|α| =l} = {α
1, . . . , αNl}. For each 1≤l≤N, define the linear transformations Lαlj:Rd→RandL
l:Rd→RNl by
Lαlj(ξ)= d
i=1
Ci,α
jyαj
ξi, j=1, . . . , Nl,
Ll(ξ)=
Lα1
l (ξ), . . . , L αNl l (ξ)
.
(2.12)
To simplify the proof of our result we need the following lemma.
Lemma2.5. Let{σtl:l=0,1, . . . , N, t∈R}be a family of measures such thatσt0=0
for allt∈R. LetDl:Rn→Rd,l=0,1, . . . , N be linear transformations. Suppose that
for allt∈Randl=0,1, . . . , N,then σl
t≤C(l),
σl t
ˆ
(ξ)≤C Mα
logc2ltDl(ξ)1+α,
σl t
ˆ
(ξ)−σtl−1
ˆ
(ξ)≤C2ltDl(ξ).
(2.13)
Then there exists a family of measures{νtl:l=1, . . . , N}t∈Rsuch that
νl
t≤C(l),
νtlˆ
(ξ)≤C Mα logc2ltD
l(ξ) 1+α,
νl tˆ
(ξ)≤C2ltDl(ξ),
σN t =
N
l=1 νl
t.
(2.14)
Proof. By [5, Lemma 6.1], for eachl=1, . . . , Nchoose two nonsingular linear
trans-formations
Al:Rr (l) →Rd, Bl:Rd →Rd, (2.15)
such that
Alπr (l)d Bl(ξ)≤Dl(ξ)≤NAlπr (l)d Bl(ξ), ξ∈Rd, (2.16)
Now chooseη∈C0∞(R)such thatη(t)=1 for|t| ≤1/2 andη(t)=0 for|t| ≥1. Let ϕ(t)=φ(t2)and let
νl
tˆ
(ξ)=σl t
ˆ (ξ)
l<j≤N
ϕ2tjA
jπr (j)d Bj(ξ)
−σtl−1
ˆ
(ξ)
l−1<j≤N
ϕ2tjA
jπr (j)d Bj(ξ)
(2.17)
with the conventionj∈∅aj=1, 1≤l≤N.
Hence, one can easily see that {σtl:l=1, . . . , N, t∈R} is the desired family of measures.
Now for the boundedness of the maximal functions(ϑl)∗,l=0,1, . . . , N, we have the following lemma whose proof is an easy consequence ofLemma 2.3, polar coordinates and Hölder’s inequality:
Lemma2.6. Forl=1, . . . , Nandp∈(1,∞), there exists a constantCp,lwhich is
inde-pendent of the coefficients of the polynomial components of the mappingQlsuch that
ϑl∗f
p≤Cp,lfp. (2.18)
3. Boundedness of some square functions. For a nonnegativeC∞radial function
ΦonRnwith
supp(Φ)⊂
x∈Rn:1
2≤ |x| ≤2
,
∞
0
Φ(t)
t dt=1, (3.1)
and for a linear transformationL:Rn→Rd, define the functionsψt,t∈Rby ˆψ t(y)=
Φ(2tL(y)).
For a family of measures{σt}t∈R, real numberuandl∈N, letJlu(f )be the square function defined by
Jlu(f )(x)=
∞
−∞
σt∗ψl(t+u)∗f (x) 2
dt 1/2
. (3.2)
For such a square function we have the following theorem.
Theorem3.1. If{σt}t∈Ris a family of measures such that the corresponding
maxi-mal function
σ∗(f )(x)=sup t∈R
σt∗f (x) (3.3)
is bounded onLp(Rd)for every1< p <∞, then
Jl
u(f )LpRd≤Cp,l
σ∗
(p/2)sup t∈R
σtfLpRd (3.4)
for every1< p <∞. HereCp,lis a constant that depends only onpand the dimension
Proof. If supt∈Rσt = ∞, then the inequality holds trivially. Thus we may assume
that supt∈Rσt<∞. In this case we follow a similar argument as in [4]. Letp >2 and q=(p/2). Choose a nonnegative functionv∈Lq+withvq=1 such that
Jl u(f )
2 p=
Rd
∞
−∞
σt∗ψl(t+u)∗f (x) 2
dt
v(x)dx. (3.5)
Thus it is easy to see that
Jl u(f )
2 p≤sup
t∈R
σt∞ −∞
Rd
ψl(t+u)∗f (z)2
σ∗(v)(−z)dz dt
≤sup t∈R
σt
Rd[g(f )]
2(z)σ∗(v)(−z) dz,
(3.6)
where
g(f )(x)=
∞
−∞
ψl(t+u)∗f (x)2 dt
1/2
. (3.7)
Now sinceRdψt(x) dx=0, it is well known that
g(f )p≤Cpfp ∀1< p <∞ (3.8)
with constantCpthat depends only onpand the dimension of the underlying space. Thus by (3.6) and Hölder’s inequality we have
Jl u(f )
2 p≤sup
t∈R
σtg(f )2pσ∗(u)q
≤Cp2sup t∈R
σtσ∗
(p/2)f2p.
(3.9)
Hence our result follows by taking the square root on both sides. The casep <2 follows by duality.
4. Proof of the main theorem. Let α >0, Ω∈Wα(n). Let ᏼ=(P1, . . . , Pd) be a polynomial mapping fromRnintoRdwith degᏼ=max1
≤j≤ddegPj=N, whered≥1 andn≥2. For 0≤l≤NletNl,Ql,νl
t,λlt, andLlbe as inSection 3. The first step in our proof is to show that eachϑl
t,l=1, . . . , Nsatisfies the hypothe-ses ofLemma 2.5, that is,
ϑl
t≤C(l), (4.1)
ϑl tˆ
(ξ)≤C Mα logc2ltL
l(ξ)1+α
, (4.2)
ϑl tˆ
(ξ)−ϑlt−1ˆ
(ξ)≤C2ltLl(ξ). (4.3)
One can easily see that (4.1) holds trivially. Using the cancellation property ofΩ, it is easy to see that (4.3) holds. Thus, we need only to verify (4.2). To see that, we notice that
ϑl tˆ
(ξ)≤
Sn−1 Ω
y 1
0e
Now the quantityξ·Ql(2tlr y)can be written in the form
ξ·Ql2tlr y=2tlrlλGly+ξ·R2tr y, (4.5)
whereQlis a homogeneous polynomial of degreelwithGl =1,Ris a polynomial of degree at mostl−1 in the variabler,
λ= Nl
j=1
Lαj
l (ξ)≥NlLl(ξ) (4.6)
andα1, . . . , αNl are the constants that appeared inSection 2. Thus by van der Corput lemma, we have
01e−2π iξ·Ql(2tr y)
dr≤Cmin1,2tlL
l(ξ)Gly−1/l
(4.7)
and hence
01e−2π iξ·Ql(2tr y)
dr≤C
logcGly−11+α
logc2tlLl(ξ)1+α , (4.8)
whereC is a constant independent oft andξ. SinceΩ∈Wα(n), the estimate (4.2) follows.
ByLemma 2.5, there exists a family of measures{νtl:l=1, . . . , N, t∈R}such that
νl
t≤C(l), (4.9)
νl tˆ
(ξ)≤C Mα logc2ltL
l(ξ)
1+α, (4.10)
νl tˆ
(ξ)≤C2ltLl(ξ), (4.11)
ϑtN= N
l=1
νtl. (4.12)
Also byLemma 2.6and the definition ofνtl(see the proof ofLemma 2.5), we have
νl∗f
p≤Cp,lfp ∀1< p <∞. (4.13)
Now one can easily see that
2−tFᏼ,t(x)=ϑNt ∗f (x)= N
l=1
νtl∗f (x). (4.14)
Therefore,
Mᏼfp≤ N
l=1
Ml
ᏼfp, (4.15)
where
Mᏼlf (x)=
∞
−∞ νl
t∗f (x) 2
dt 1/2
Thus to show the boundedness ofMᏼf, it suffices to show that Ml
ᏼfp≤Cp,lfp (4.17)
forp∈((2+2α)/(1+2α),2+2α), and for alll=1, . . . , N.
To show (4.17), we proceed as follows: letΦandψtbe as inSection 3. Then
Ml
ᏼf (x)=log 2l
∞
−∞ −∞∞ νl
t∗ψl(t+u)∗f (x) du
2dt
1/2
≤log 2l
∞
−∞S l
uf (x) du,
(4.18)
where
Sulf (x)=
∞
−∞ νl
t∗ψl(t+u)∗f (x) 2
dt 1/2
. (4.19)
Now by (4.13) andTheorem 3.1, we have
Sl
ufp≤Cpfp (4.20)
for allp∈(1,∞)and forl=1, . . . , N which in turn implies that
1
−1
Sl
ufpdu≤2Cpfp ∀p∈(1,∞). (4.21)
On the other hand, ifu≥1, by the estimate (4.11) we have
Sl uf
2 2=
Rd ∞
−∞ νl
t∗ψl(t+u)∗f (x)2dt dx
=
∞
−∞
Rd
Φ2lt+luL l(ξ)
2 νl
tˆ
(ξ)2f (ξ)ˆ 2dξ dt
≤22l−2lu
Rd
f (ξ)ˆ 2log(2 l/|L
l(ξ)|)−u
log(1/2l|L l(ξ)|)−u
dt
dξ
=2 log 2l22l−2luf2 2.
(4.22)
Thus
Sl uf2≤
2 log 2l2l−luf2. (4.23)
By interpolating between (4.20) and (4.23) we get
Sl
ufp≤Cp,l2θl−θlufp (4.24)
for all 1< p <∞and for someθ=θ(p) >0. Hence we have
∞
1
Sl
Finally, if u <−1, by the estimate (4.10) and similar argument as in the case of u≤1, we get
Sl
uf2≤Cl(|u|)−1−αf2. (4.26)
By interpolating between (4.26) and anyp ∈(1,∞) in (4.20), we get that, ifp∈ ((2+2α)/(1+2α),2+2α)there existsβ >0 such that
Sl
ufp≤Cp(|u|)−βfp, (4.27)
which implies that
−1
−∞ Sl
ufpdu≤Cpfp (4.28)
forp∈((2+2α)/(1+2α),2+2α).
Hence by combining (4.18), (4.21), (4.25), and (4.28) we get (4.17).
Acknowledgement. The publication of this paper was supported by Yarmouk
University Research Council.
References
[1] A. Al-Salman and Y. Pan,Singular integrals with rough kernels, to appear in Canad. Math. Bull.
[2] A. Benedek, A. P. Calderón, and R. Panzone, Convolution operators on Banach space valued functions, Proc. Nat. Acad. Sci. U.S.A. 48 (1962), 356–365.MR 24#A3479. Zbl 103.33402.
[3] J. Chen, D. Fan, and Y. Pan,A note on a Marcinkiewics integral operator, to appear in Math. Nachr.
[4] J. Duoandikoetxea and J. L. Rubio de Francia,Maximal and singular integral operators via Fourier transform estimates, Invent. Math.84(1986), no. 3, 541–561.MR 87f:42046. Zbl 568.42012.
[5] D. Fan and Y. Pan,Singular integral operators with rough kernels supported by subvarieties, Amer. J. Math.119(1997), no. 4, 799–839.MR 99c:42029. Zbl 0899.42002.
[6] L. Grafakos and A. Stefanov, Lp bounds for singular integrals and maximal singular integrals with rough kernels, Indiana Univ. Math. J. 47 (1998), no. 2, 455–469.
MR 99i:42019. Zbl 913.42014.
[7] E. M. Stein,Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Inte-grals, Princeton Mathematical Series, vol. 43, Princeton University Press, New Jersey, 1993.MR 95c:42002. Zbl 821.42001.
Hussain Al-Qassem: Department of Mathematics, Yarmouk University, Irbid, Jordan E-mail address:husseink@yu.edu.jo