# ROUGH BILINEAR SINGULAR grafakosl/preprints/Rough Bilinear Singular Integrals 29.pdf Seeger [28] in

date post

15-Jul-2020Category

## Documents

view

0download

0

Embed Size (px)

### Transcript of ROUGH BILINEAR SINGULAR grafakosl/preprints/Rough Bilinear Singular Integrals 29.pdf Seeger [28] in

ROUGH BILINEAR SINGULAR INTEGRALS

LOUKAS GRAFAKOS, DANQING HE, PETR HONZÍK

ABSTRACT. We study the rough bilinear singular integral, introduced by Coifman and Meyer [9],

TΩ( f ,g)(x)= p.v. ∫ Rn

∫ Rn |(y,z)|−2nΩ((y,z)/|(y,z)|) f (x−y)g(x−z)dydz,

when Ω is a function in Lq(S2n−1) with vanishing integral and 2≤ q≤∞. When q = ∞ we obtain boundedness for TΩ from Lp1(Rn)× Lp2(Rn) to Lp(Rn) when 1 < p1, p2 < ∞ and 1/p = 1/p1 + 1/p2. For q = 2 we obtain that TΩ is bounded from L2(Rn)×L2(Rn) to L1(Rn). For q between 2 and infinity we obtain the analogous boundedness on a set of indices around the point (1/2,1/2,1). To obtain our results we introduce a new bilinear technique based on tensor-type wavelet decompositions.

CONTENTS

1. Introduction 1 2. Estimates of Fourier transforms of the kernels 4 3. Boundedness: a good point 6 4. The diagonal part 10 5. The off-diagonal parts 14 6. Boundedness everywhere when q = ∞ 17 7. Boundedness of TΩ when Ω ∈ Lq(S2n−1) with 2≤ q < ∞ 22 References 23

1. INTRODUCTION

Singular integral theory was initiated in the seminal work of Calderón and Zygmund [3]. The study of boundedness of rough singular integrals of convolution type has been an active area of research since the middle of

The first author would like to acknowledge the support of the Simons Foundation and of the University of Missouri Research Board and Research Council. The second author was supported by NNSF of China (No. 11701583), and the Guangdong Natural Science Foundation (No. 2017A030310054). The third author was supported by the ERC CZ grant LL1203 of the Czech Ministry of Education.

2010 Mathematics Classification Number 42B20, 42B99. 1

2 LOUKAS GRAFAKOS, DANQING HE, PETR HONZÍK

the twentieth century. Calderón and Zygmund [4] first studied the rough singular integral

LΩ( f )(x) = p.v. ∫ Rn

Ω(y/|y|) |y|n

f (x− y)dy

where Ω is in L logL(Sn−1) with mean value zero and showed that LΩ is bounded on Lp(Rn) for 1 < p < ∞. The same conclusion under the less restrictive condition that Ω lies in H1(Sn−1) was obtained by Coifman and Weiss [10] and Connett [11]. The weak type (1,1) boundedness of LΩ when n = 2 was established by Christ and Rubio de Francia [7] and indepen- dently by Hofmann [20], both inspired by Christ’s work [6]. Additionally, in unpublished work, Christ and Rubio de Francia extended this result to all dimensions n ≤ 7. The weak type (1,1) property of LΩ was proved by Seeger [28] in all dimensions and was later extended by Tao [30] to situ- ations in which there is no Fourier transform structure. Several questions remain concerning the endpoint behavior of LΩ, such as if the condition Ω∈ L logL(Sn−1) can be relaxed to Ω∈H1(Sn−1), or merely Ω∈ L1(Sn−1) when Ω is an odd function. On the former there is a partial result of Ste- fanov [29] but not much is still known about the latter.

The bilinear counterpart of the rough singular integral linear theory is notably more intricate. To fix notation, we fix 1 < q ≤ ∞ and we let Ω in Lq(S2n−1) with

∫ S2n−1 Ωdσ = 0, where S2n−1 is the unit sphere in R2n.

Coifman and Meyer [9] introduced the bilinear singular integral operator associated with Ω by

(1) TΩ( f ,g)(x) = p.v. ∫ Rn

∫ Rn

K(x− y,x− z) f (y)g(z)dydz,

where f ,g are functions in the Schwartz class S(Rn),

K(y,z) = Ω((y,z)′)/|(y,z)|2n ,

and x′ = x/|x| for x ∈ R2n. General facts about bilinear operators can be found in [26, Chapter 13], [17, Chapter 7], and [27]. If Ω possesses some smoothness, i.e. if is a function of bounded variation on the cir- cle, Coifman and Meyer [9, Theorem I] showed that TΩ is bounded from Lp1(R)×Lp2(R) to Lp(R) when 1 < p1, p2, p < ∞ and 1/p = 1/p1+1/p2. In higher dimensions, it was shown Grafakos and Torres [19], via a bilinear T 1 condition, that if Ω a Lipschitz function on S2n−1, then TΩ is bounded from Lp1(Rn)× Lp2(Rn) to Lp(Rn) when 1 < p1, p2 < ∞, 1/2 < p < ∞, and 1/p = 1/p1 + 1/p2. But if Ω is rough, the situation is significantly more complicated, and the boundedness of TΩ remained unresolved until this work, except when in situations when it reduces to the uniform bound- edness of bilinear Hilbert transforms. If Ω is merely integrable function on

ROUGH BILINEAR SINGULAR INTEGRALS 3

S1, but is odd, the operator TΩ is intimately connected with the celebrated (directional) bilinear Hilbert transform

Hθ1,θ2( f1, f2)(x) = ∫ +∞ −∞

f1(x− tθ1) f2(x− tθ2) dt t

(in the direction (θ1,θ2)), via the relationship

TΩ( f1, f2)(x) = 1 2

∫ S2n−1

Ω(θ1,θ2)Hθ1,θ2( f1, f2)(x)d(θ1,θ2) .

The boundedness ofHθ1,θ2 was proved by Lacey and Thiele [22], [23] while the more relevant, for this problem, uniform in θ1,θ2 boundedness ofHθ1,θ2 was addressed by Thiele [31], Grafakos and Li [18], and Li [24]. Exploit- ing the uniform boundedness of Hθ1,θ2 , Diestel, Grafakos, Honzı́k, Si, and Terwilleger [13] showed that if n = 2 and the even part of Ω lies in H1(S1), then TΩ is bounded from Lp1(R)×Lp2(R) to Lp(R) when 1< p1, p2, p 1/2 in all dimensions. This breakthrough is a consequence of a new technique in this context. We build on the work of Duoandikoetxea and Rubio de Francia [15], but our key idea is to decompose the multiplier in terms of a tensor-type compactly-supported wavelet decomposition and to use combinatorial arguments to group the different pieces together, ex- ploiting orthogonality. One may speculate that other decompositions, such as certain types of frames or the φ -transform of Frazier and Jawerth [16], which has good almost-orthogonality estimates, could be used to attack this problem. But as the combinatorial aspect of our approach heavily depends on the fact that our basis is of tensor type, similar properties of other de- compositions will be necessary if they are to be adapted to the techniques of this article.

The main result of this paper is the following theorem.

Theorem 1. For all n≥ 1, if Ω ∈ L∞(S2n−1), then for TΩ defined in (1), we have

(2) ‖TΩ‖Lp1(Rn)×Lp2(Rn)→Lp(Rn) < ∞ whenever 1 < p1, p2 < ∞ and 1/p = 1/p1 +1/p2.

4 LOUKAS GRAFAKOS, DANQING HE, PETR HONZÍK

In the remaining sections we focus on the proof of this result while in the last section we focus on extensions to the case where Ω lies in Lq(S2n−1) for q < ∞.

Some remarks about our notation in this paper: For 1 < q < ∞ we set q′ = q/(q−1) and for q = ∞, we set ∞′ = 1. We denote the the norm of a bounded bilinear operator T from X×Y to Z by

‖T‖X×Y→Z = sup ‖ f‖X≤1

sup ‖g‖Y≤1

‖T ( f ,g)‖Z .

This notation was already used in (2). If x1,x2 are in Rn, then we denote the point (x1,x2) in R2n by ~x. We denote the set of positive integers by N and we set N0 =N∪{0}. In the sequel, multiindices in Z2n are elements of N2n0 . Finally, we adhere to the standard convention to denote by C a constant that depends only on inessential parameters of the problem.

2. ESTIMATES OF FOURIER TRANSFORMS OF THE KERNELS

Let us fix a q satisfying 1 < q ≤ ∞ and a function Ω ∈ Lq(Sn−1) with mean value zero. We fix a smooth function α in R+ such that α(t) = 1 for t ∈ (0,1], 0 < α(t)< 1 for t ∈ (1,2) and α(t) = 0 for t ≥ 2. For (y,z) ∈R2n and j ∈ Z we introduce the function

β j(y,z) = α(2− j|(y,z)|)−α(2− j+1|(y,z)|). We write β = β0 and we note that this is a function supported in [1/2,2]. We denote ∆ j the Littlewood-Paley operator ∆ j f = F−1(β j f̂ ). Here and throughout this paper F−1 denotes the inverse Fourier transform, which is defined via F−1(g)(x) =

∫ Rn g(ξ )e2πix·ξ dξ = ĝ(−x), where ĝ is the Fourier

transform of g. We decompose the kernel K as follows: we denote Ki = βiK and we set Kij = ∆ j−iK

i for i, j ∈ Z. Then we write

K = ∞

∑ j=−∞

K j,

where

K j = ∞

∑ i=−∞

Kij.

We also denote m j = K̂ j. Then the operator can be written as

TΩ( f ,g)(x) = ∑ j

∫ Rn

∫ Rn

K j(x− y,x− z) f (y)g(z)dydz =: ∑ j

Tj( f ,g)(x).

We have the following lemma whose proof is known (see for instance [14]) and is omitted.

ROUGH BILINEAR SINGULAR INTEGRALS 5

Lemma 2. Given Ω ∈ Lq(S2n−1), 0 < δ < 1/q′ and ~ξ = (ξ1,ξ2) ∈ R2n we have

|K̂0(~ξ )| ≤C‖Ω‖Lq min(|~ξ |, |~ξ |−δ ) and for all multiindices α in Z2n with α 6= 0 we have

|∂ α K̂0(~ξ )| ≤Cα‖Ω‖Lq min(1, |~ξ |−δ ) .

The following proposition is a consequence of the preceding lemma.

Proposition 3. Let 1 ≤ p1, p2 < ∞ and define p via 1/p = 1/p1 + 1/p2. Let Ω ∈ Lq(S2n−1), 1 < q ≤ ∞, 0 < δ < 1/q′, and for j ∈ Z consider the bilinear operator

Tj( f ,g)(x) = ∫ R2n

K j(x− y,x− z) f (y)g(z)dydz .

If both p1, p2 > 1, then Tj is bounded from Lp1(Rn)×Lp2(Rn) to Lp(Rn) with norm at most C‖Ω‖Lq 2(2n−δ ) j if j ≥ 0 and at most C‖Ω‖Lq 2−| j|(1−δ ) if j < 0. If at least one of p1 and p2 is equal to 1, then Tj maps Lp1(Rn)× Lp2(Rn) to Lp,∞(Rn) with a similar norm.

Proof. We prove the assertion by showing that the multiplier m j = K̂ j as- sociated with Tj satisfies the conditions of the Coifman-Meyer multi

Recommended

*View more*