Abstract
We study the bilinear Marcinkiewicz integral operator associated with a rough, integrable kernel on . For every , we prove that the operator maps boundedly into whenever , , satisfies the mean-zero condition. The proof uses a parameter-dependent product wavelet decomposition adapted to the square-function structure. To the best of our knowledge, this is the first boundedness result for the bilinear Marcinkiewicz integral operator with rough angular kernels, without any angular smoothness assumption.
Keywords:
bilinear Marcinkiewicz integral; rough kernels; Lp boundedness; square functions; wavelet decomposition MSC:
42B20; 42B25
1. Introduction
Singular integrals and Littlewood–Paley square functions are fundamental objects in harmonic analysis. Their boundedness depends on the interplay between cancellation, kernel singularity, and angular regularity. Since the foundational work of Calderón and Zygmund [1,2], the study of integral operators with rough kernels has attracted considerable attention.
A classical example is the Marcinkiewicz integral. Let , and let satisfy
The Marcinkiewicz integral associated with is defined by
Stein [3] proved that this operator is bounded on for and is of weak type when with . Benedek, Calderón, and Panzone [4] subsequently established boundedness for all under the assumption . Later developments substantially weakened the angular regularity requirements. Ding, Fan, and Pan [5] proved boundedness for all when , and Al-Salman, Al-Qassem, Cheng, and Pan [6] obtained the same conclusion under the condition . Here the notation , for , means that
where . The case is denoted by .
The Marcinkiewicz integral is closely related to the rough Calderón–Zygmund singular integral
where p.v. denotes the Cauchy principal value. The operators and involve the same angular cancellation (1), but their radial structures differ. The kernel inside the truncated integral defining has order , whereas the singular integral kernel has order . Thus the former has a weaker singularity inside each truncation, while the outer square-function norm introduces an additional scale-dependent structure.
For , Calderón and Zygmund [2] established boundedness for when . The space is strictly contained in , the kernel class in the result for cited above. Connett [7] and Ricci and Weiss [8] obtained the same conclusion under the condition . Thus the condition yields boundedness for both and throughout the range . Grafakos and Stefanov [9] established bounds for under a logarithmic integrability condition on the kernel. Chen, Fan, and Pan [10] obtained related bounds for the Marcinkiewicz integral for kernels in the Grafakos–Stefanov class. These similar results further illustrate the connection between the two operators and motivate the investigation of their bilinear counterparts.
Let , and let satisfy
The bilinear rough singular integral associated with is defined by
where p.v. denotes the Cauchy principal value and . Coifman and Meyer [11] introduced this operator and established boundedness results under regularity assumptions. In dimension , they obtained boundedness for and , when has bounded variation on . For Lipschitz kernels in higher dimensions, the multilinear Calderón–Zygmund theory of Grafakos and Torres [12] yields the boundedness of from to throughout the range
The rough-kernel case poses additional challenges. Grafakos, He, and Honzík [13] proved that, for mean-zero with , is bounded from to , and also from to for exponent triples in a neighborhood of . When , they further established the boundedness of throughout the same full exponent range stated in (3).
Their proof begins with a dyadic decomposition of into operators denoted by in their paper. For , the Coifman–Meyer multiplier theorem provides summable bounds for . For large positive j, the decaying estimate is obtained by decomposing the multiplier of into compactly supported product wavelets and exploiting combinatorial grouping and frequency orthogonality. To reach the full exponent range for bounded kernels, they also verified that, for , the kernel of satisfies the bilinear Calderón–Zygmund size and Hölder continuity estimates with kernel constant
Combining these kernel estimates with the bound, they applied the multilinear Calderón–Zygmund theorem of Grafakos and Torres [12] to obtain
Since can be chosen arbitrarily small, interpolation with the decaying estimate yields sufficient decay for large positive j. Together with the bounds for , this allows the operators to be summed throughout the full exponent range.
Subsequent work weakened the angular integrability assumptions in the bilinear setting and extended the theory to multilinear operators. In the bilinear case, He and Park [14] proved boundedness of for and , assuming that has mean zero and . For rough m-linear singular integrals with mean-zero angular kernels in , Grafakos, He, Honzík, and Park [15] established boundedness in the optimal open range of input exponents for , and Dosidis and Slavíková [16] extended the boundedness result to . When , these results imply that maps boundedly into whenever has mean zero and belongs to for some .
Motivated by these results and the relationship between singular integrals and Marcinkiewicz integrals, we study the corresponding bilinear square function. For satisfying (2), define
where
The square function is the pointwise norm of the bilinear map . This formulation relates to Hart’s vector-valued theory [17], which covers output exponents below one for discrete bilinear square functions. Hart’s framework imposes size and local Hölder-type difference estimates on the vector-valued kernel ([17], Definition 5.1). The weighted theory of Bu, Fu, and Zhang [18] imposes similar kernel size and difference conditions for non-convolution bilinear Marcinkiewicz integrals.
In what follows, we study for mean-zero kernels with . For finite q, this assumption allows essentially unbounded kernels and therefore does not guarantee the pointwise kernel size bounds required by Hart ([17], Equation (5.2)) and by Bu, Fu, and Zhang ([18], Definition 4). Even for , angular jumps can violate the kernel difference estimates required in both frameworks. Thus neither framework applies directly to the full class of kernels considered here.
From the Fourier multiplier perspective, has the dilation form of the bilinear square functions studied by Si, Xue, and Yabuta [19]. Let and set . For , the kernel of is , whose multiplier is . Since , we obtain
The weighted theorem of Si, Xue, and Yabuta ([19], Theorem 1.1) assumes a family of a priori unweighted weak-type estimates for the square function. In particular, applying their theorem at requires an unweighted bound.
In this paper, we prove the following estimate for directly from the angular condition and spherical cancellation, without imposing any of the additional hypotheses discussed above.
Theorem 1.
The restriction arises in the Plancherel step, where we use the embedding to control the norm of the unit-scale kernel piece . As for the mapping exponents, the scalar-valued bilinear Calderón–Zygmund argument for does not extend the range of boundedness for , which is sublinear in each input.
The proof adapts the product wavelet method to the mixed norm . We first introduce a continuous frequency decomposition and express the truncated kernel as a sum of rescaled annular kernels. The resulting multiplier is separated into a part where both frequency variables are bounded away from zero and two parts where one frequency variable is small. Writing , with and , separates the local parameter dependence from the dyadic scale.
A key ingredient in our proof is the local parameter-dependent multiplier estimate in Lemma 4. This estimate is similar in form to the fixed-symbol criterion of Grafakos, He, and Slavíková ([20], Theorem 1.3 with ). Both estimates use bounds L and A for the norm of the symbol and the norms of its derivatives, respectively. With these bounds uniform in , applying their criterion to each gives an bound of order . By Minkowski’s inequality, this controls the weaker mixed norm. Our lemma yields the bound needed for the square-function argument, with operator norm at most . We expand the symbols in a common product wavelet basis, view the coefficient functions as elements of , and group the wavelet indices into rows, columns, and matching sets to exploit frequency orthogonality. When one frequency variable is small, maximal-function bounds, orthogonality, and bounded overlap of frequency annuli provide the decay needed for summation.
The paper is organized as follows. Section 2 presents the frequency decomposition and reduces Theorem 1 to two estimates for the localized operators. Section 3 proves Proposition 1. Section 4 establishes Proposition 2 and proves the local parameter-dependent multiplier estimate (Lemma 4).
Throughout the paper, C denotes a positive constant that is independent of the essential variables and may vary from line to line. Its dependence on relevant parameters will be specified when necessary.
2. Preliminaries for the Proof of Theorem 1
Set
Then
where , and the convolution is taken on .
Choose a nonnegative function such that and
For , define the Fourier multiplier operator on by
For , set
Thus,
By the change of variables , we have
Consequently,
This gives the decomposition
Define the bilinear multiplier operator by
Using the preceding decomposition, we obtain
By Minkowski’s integral inequality,
Applying Minkowski’s integral inequality once more, we obtain, for ,
For and a family of measurable functions on , define
This is a norm when and a quasi-norm when . To prove Theorem 1, we will establish the following two propositions.
Proposition 1.
Proposition 2.
The conclusion of Theorem 1 then follows from (4) and the preceding two propositions. These propositions will be proved in the next two sections.
3. Proof of Proposition 1
We first state a lemma whose proof follows the same argument as that of the Coifman–Meyer multiplier theorem (see Theorem 7.5.3 of [21]).
Lemma 1.
Let be a family of functions on satisfying
for every and all multi-indices with . Let and satisfy . Define
Then there exists a constant such that
To prove Proposition 1, it suffices to verify the derivative estimates required by Lemma 1, with the appropriate dependence on s. We first record the following estimates, which follow by adapting the Fourier transform argument in ([22], p. 550).
Lemma 2.
Let , where , satisfy the mean-zero condition (2), and let . Then
Moreover, for every nonzero multi-index ,
By (6), the scaling identity , and the support condition , we obtain
where
For each fixed , the set F is an interval of length 2, and is comparable to on F.
To estimate the derivatives of , set
For with , Leibniz’s rule gives
Since the derivatives of are bounded and supported in the annulus , it follows that
If , then (7) yields
where
Indeed, on F,
Since , we have
If , then (6) gives
Combining (8), (9), (10) and (11), we obtain, for every with ,
This estimate, together with Lemma 1, completes the proof of Proposition 1.
4. Proof of Proposition 2
Define
and, for , set
The dyadic decomposition of the unit ball gives
Taking Fourier transforms, we obtain
Using polar coordinates, we have
Consequently, for ,
Moreover, the same argument as that used to establish (6) and (7) gives, for every multi-index , including ,
For , let and define
Equivalently,
We have the following estimates.
Lemma 3.
For every ,
Moreover, for every integer ,
Proof.
Let be a smooth radial function on such that , for , and for . Define
and
Let . Then
The support conditions imply
For example, on the support of ,
We will establish
and
These two estimates imply Proposition 2.
Indeed, assuming (20) and (21), we obtain
By (12) and the definition of ,
Set . By the definition of ,
Therefore,
Using (22) and the translation invariance of integration in t, we obtain
For ,
Moreover,
Consequently,
To prove (20) and (21), we first state the following lemma, whose proof will be given at the end of this section.
Lemma 4.
Let , and let be an integer. Suppose that , , is a jointly measurable family of symbols such that for almost every u. Assume that
and
Define
Then
The constant is independent of the locations and sizes of the supports of the symbols. Consequently, the operator extends to a bounded bilinear map from into .
4.1. Estimate for
4.2. Estimates for
The support conditions for and are symmetric in and , and the same argument applies to both terms after interchanging the roles of f and g. We therefore only prove
Let
To further decompose , we use compactly supported product wavelets with sufficiently high regularity. The existence of such wavelets was established by Daubechies [23] and discussed in Meyer’s book [24]. Their construction is also presented in Triebel [25].
Lemma 5
([25], pp. 8–9, (1.55)–(1.60)). For any fixed , there exist real compactly supported functions satisfying , and for , we have , and, if for , we define
where belongs to the set
then the family of functions
forms an orthonormal basis of .
Let . We use the product wavelets in Lemma 5 with regularity at least and vanishing moments through order T. We choose the generating functions and with support of diameter approximately 1. Wavelets with the same dilation factor have a uniformly bounded overlap N independent of .
As in [13], the wavelets at each scale can be divided into at most groups such that the wavelets in each group have disjoint supports and share the same index . For , let denote one such group. The supports of its wavelets have diameters about . Therefore, the wavelet expansion of is
where
By the definition of ,
Thus, for every multi-index , Leibniz’s rule gives
where is viewed as a function on that is independent of . Since , the derivatives of the cutoff are uniformly bounded. Hence, by (16),
Moreover, and its derivatives are supported in the region . It follows that, for any fixed ,
In particular,
Using the derivative bounds in (30) for and the vanishing moments of the wavelets, we apply the argument of Lemma 7 in [13] to obtain
For , write
where and We accordingly write . Define
and set
By (31),
The decomposition (29) can therefore be written as
where
By (18), for every ,
If , the support of the corresponding wavelet intersects this region. Since its support has diameter at most , there is a fixed constant , independent of j, u, and , such that
On the other hand, for a fixed constant depending only on the generating functions, is contained in the cube
Consequently, there are at most possible values of k for which some is nonzero.
We next establish the pointwise bound
where M is the Hardy–Littlewood maximal operator on . For fixed and , write
Then
and hence
The compact support and sufficient regularity of imply
Therefore, the convolution kernel satisfies
and thus
Since , this gives
Splitting the integral into and , , yields
The constant is independent of k, h, and .
Applying the Cauchy–Schwarz inequality, we obtain
For fixed , , and k, the supports of the functions appearing in this sum are pairwise disjoint. Consequently, Plancherel’s theorem and (32) give
Combining this estimate with (34), the boundedness of M, and the bound for the number of relevant indices k, we conclude that
The analogous estimate for follows by interchanging the two frequency variables and the roles of f and g. Together, these estimates establish (21).
4.3. Proof of Lemma 4
Set . We use the wavelet construction and notation introduced above. Applying Lemma 5 with , we obtain fixed generating functions satisfying
Recall that each group introduced above has a fixed product type . For this type, write
The normalized factors defined above are then
and the wavelet with translation indices is
Choose such that , and fix an integer . For , let consist of the wavelets in whose translation indices satisfy
Each pair determines a unique pair of residue vectors , so the nonempty sets partition into at most subgroups.
Within a fixed subgroup, if , then , so for some coordinate i. Since
and , the boxes corresponding to k and have disjoint projections onto the ith coordinate axis. Hence and have disjoint supports. The same argument shows that and have disjoint supports whenever within the subgroup.
The number of original groups is uniformly bounded in , and d depends only on the wavelet system fixed in terms of n and T. Thus the number of refined groups at each scale is bounded by an integer . After adding empty groups if necessary, we index the refined groups at each scale by and continue to denote them by . We also retain the notation for the set of translation-index pairs in the refined group .
Using the wavelet basis introduced above, we expand for almost every as
where
Since , Parseval’s identity gives
where the sum runs over the entire wavelet basis.
For , set
The functions and are uniformly bounded. Within the fixed group, their supports are pairwise disjoint for distinct first indices and distinct second indices, respectively. Consequently, Plancherel’s theorem yields
where the sums run over the distinct indices occurring in the group.
In addition, (36) gives that
Define
Set
We claim that
The estimate is immediate if . We may therefore assume .
For , define
These sets partition the pairs with nonzero coefficients. Moreover,
so
Fix a positive integer . For each first-coordinate index k, let
Thus, indexes the rows containing at least pairs. Since distinct rows are disjoint,
and hence
For , define
The supports of the Fourier transforms are pairwise disjoint. Plancherel’s theorem therefore gives
Using the triangle inequality in H and Cauchy–Schwarz, first in x and then in k, we obtain
where we used (37) and (40).
Remove all pairs belonging to these rows, and write
For each second-coordinate index l, define
Thus, indexes the remaining columns containing at least pairs. Since distinct columns are disjoint,
and therefore
Interchanging the roles of the two coordinates in the row estimate gives
After removing these columns, denote the residual set by
Every row and every column of contains at most pairs. We show that this set can be partitioned into at most subsets such that distinct pairs and in the same subset satisfy and We call subsets with this property matching sets.
If , the residual set is empty. Suppose that . Since is finite, enumerate its pairs and place them successively into initially empty sets. When placing a pair , exclude every set that already contains a pair with first index k or second index l. Apart from itself, its row contains at most pairs, and its column contains at most pairs. Thus, at most previously placed pairs can prevent a set from being used, and at most that many sets are excluded. Since , an available set always exists.
This procedure places every pair into exactly one set and ensures that no set contains two pairs sharing a first or second index. Discarding the empty sets gives the desired partition into at most matching sets.
For each matching set , the triangle inequality in H and Cauchy–Schwarz yield
Here each first index and each second index occurs at most once in . Integrating in x and applying (37), we obtain
Since the residual set is partitioned into at most matching sets, the triangle inequality gives
Combining the estimates (41), (42), and (43), we obtain
Choose
Since , we have
Consequently,
Substituting this bound into the preceding estimate and using , we obtain
The sets partition the pairs in with nonzero coefficients. Since , the triangle inequality and summation over r prove (39).
To estimate each wavelet coefficient , first consider a wavelet of type . Writing and , the corresponding wavelet has the form
There is a coordinate such that . The vanishing moments of and a change of variables imply
For every multi-index with , we have . Factoring the integral into one-dimensional integrals therefore gives
In particular, every polynomial of total degree at most has zero integral against .
Choose . For almost every u, define the Taylor polynomial of degree by
The preceding cancellation gives
Taylor’s formula with integral remainder gives
Since for , it follows that
Since the support of has diameter at most , then
Moreover, the definition of and the change of variables give
Consequently, combining (44) with the preceding two estimates, we obtain
The type occurs only at and is estimated directly, without using cancellation. For these scaling functions,
Hence, for almost every ,
Thus, the bound also holds for this type.
Since has measure one, the coefficient estimates for all types imply
5. Conclusions
In this paper, we establish the boundedness of the bilinear Marcinkiewicz integral for mean-zero kernels , . The proof combines a product wavelet decomposition with a local parameter-dependent bilinear multiplier estimate adapted to the square-function structure.
A natural question is whether, for every mean-zero kernel , the bilinear Marcinkiewicz integral satisfies the estimate
throughout the range
For rough bilinear singular integrals , Grafakos, He, and Honzík [13] obtained estimates throughout this range for bounded kernels by combining their initial boundedness estimate with bilinear Calderón–Zygmund kernel estimates and interpolation. This scalar-valued argument does not apply directly to , whose definition involves an outer norm in the truncation parameter. An extension of that approach would require estimates adapted to the corresponding Hilbert space-valued operator. To the best of our knowledge, boundedness of throughout the above exponent range remains unresolved for mean-zero .
Beyond the bounded-kernel case, a further problem is to determine the admissible combinations of q, , and for which
holds for every mean-zero , with . In particular, it remains to determine how the integrability of the angular kernel restricts the range of boundedness.
Author Contributions
Conceptualization, W.H. and D.F.; methodology, W.H. and D.F.; formal analysis, W.H.; investigation, W.H. and D.F.; writing—original draft preparation, W.H.; writing—review and editing, D.F. All authors have read and agreed to the published version of the manuscript.
Funding
This research was supported by the National Natural Science Foundation of China (No. 12571109, 12071437).
Institutional Review Board Statement
Not applicable.
Informed Consent Statement
Not applicable.
Data Availability Statement
Data are contained within the article.
Acknowledgments
During the preparation of this manuscript, the authors used ChatGPT-6 to polish the language of the initial draft. The authors have reviewed and edited the output and take full responsibility for the content.
Conflicts of Interest
The authors declare no conflicts of interest.
References
- Calderón, A.P.; Zygmund, A. On the existence of certain singular integrals. Acta Math. 1952, 88, 85–139. [Google Scholar] [CrossRef] [Scilit]
- Calderón, A.P.; Zygmund, A. On singular integrals. Amer. J. Math. 1956, 78, 289–309. [Google Scholar] [CrossRef] [Scilit]
- Stein, E.M. On the functions of Littlewood-Paley, Lusin, and Marcinkiewicz. Trans. Amer. Math. Soc. 1958, 88, 430–466. [Google Scholar] [CrossRef] [Scilit]
- Benedek, A.; Calderón, A.P.; Panzone, R. Convolution operators on Banach space valued functions. Proc. Nat. Acad. Sci. USA 1962, 48, 356–365. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Ding, Y.; Fan, D.; Pan, Y. Lp-boundedness of Marcinkiewicz integrals with Hardy space function kernels. Acta Math. Sin. (Engl. Ser.) 2000, 16, 593–600. [Google Scholar] [CrossRef] [Scilit]
- Al-Salman, A.; Al-Qassem, H.; Cheng, L.C.; Pan, Y. Lp bounds for the function of Marcinkiewicz. Math. Res. Lett. 2002, 9, 697–700. [Google Scholar] [CrossRef] [Scilit]
- Connett, W.C. Singular integrals near L1. In Harmonic Analysis in Euclidean Spaces, Part 1; American Mathematical Society: Providence, RI, USA, 1979; pp. 163–165. [Google Scholar]
- Ricci, F.; Weiss, G. A characterization of H1(Σn−1). Proc. Symp. Pure Math. 1979, 35, 289–294. [Google Scholar]
- Grafakos, L.; Stefanov, A. Lp bounds for singular integrals and maximal singular integrals with rough kernels. Indiana Univ. Math. J. 1998, 47, 455–469. [Google Scholar] [CrossRef] [Scilit]
- Chen, J.; Fan, D.; Pan, Y. A note on a Marcinkiewicz integral operator. Math. Nachr. 2001, 227, 33–42. [Google Scholar] [CrossRef] [Scilit]
- Coifman, R.R.; Meyer, Y. On commutators of singular integrals and bilinear singular integrals. Trans. Amer. Math. Soc. 1975, 212, 315–331. [Google Scholar] [CrossRef] [Scilit]
- Grafakos, L.; Torres, R.H. Multilinear Calderón-Zygmund theory. Adv. Math. 2002, 165, 124–164. [Google Scholar] [CrossRef] [Scilit]
- Grafakos, L.; He, D.; Honzík, P. Rough bilinear singular integrals. Adv. Math. 2018, 326, 54–78. [Google Scholar] [CrossRef] [Scilit]
- He, D.; Park, B.J. Improved estimates for bilinear rough singular integrals. Math. Ann. 2023, 386, 1951–1978. [Google Scholar] [CrossRef] [Scilit]
- Grafakos, L.; He, D.; Honzík, P.; Park, B.J. Multilinear rough singular integral operators. J. Lond. Math. Soc. 2024, 109, e12867. [Google Scholar] [CrossRef] [Scilit]
- Dosidis, G.; Slavíková, L. Multilinear singular integrals with homogeneous kernels near L1. Math. Ann. 2024, 389, 2259–2271. [Google Scholar] [CrossRef] [Scilit]
- Hart, J. Bilinear square functions and vector-valued Calderón-Zygmund operators. J. Fourier Anal. Appl. 2012, 18, 1291–1313, Erratum in J. Fourier Anal. Appl. 2014, 20, 222–224. [Google Scholar] [CrossRef] [Scilit]
- Bu, R.; Fu, Z.; Zhang, Y. Weighted estimates for bilinear square functions with non-smooth kernels and commutators. Front. Math. China 2020, 15, 1–20. [Google Scholar] [CrossRef] [Scilit]
- Si, Z.; Xue, Q.; Yabuta, K. On the bilinear square Fourier multiplier operators and related multilinear square functions. Sci. China Math. 2017, 60, 1477–1502. [Google Scholar] [CrossRef] [Scilit]
- Grafakos, L.; He, D.; Slavíková, L. L2 × L2 → L1 boundedness criteria. Math. Ann. 2020, 376, 431–455. [Google Scholar] [CrossRef] [Scilit]
- Grafakos, L. Modern Fourier Analysis, 3rd ed.; Springer: New York, NY, USA, 2014. [Google Scholar]
- Duoandikoetxea, J.; Rubio de Francia, J.L. Maximal and singular integral operators via Fourier transform estimates. Invent. Math. 1986, 84, 541–561. [Google Scholar] [CrossRef] [Scilit]
- Daubechies, I. Orthonormal bases of compactly supported wavelets. Comm. Pure Appl. Math. 1988, 41, 909–996. [Google Scholar] [CrossRef] [Scilit]
- Meyer, Y. Wavelets and Operators. In Cambridge Studies in Advanced Mathematics; Cambridge University Press: Cambridge, UK, 1992; Volume 37. [Google Scholar]
- Triebel, H. Bases in Function Spaces, Sampling, Discrepancy, Numerical Integration; EMS Tracts in Mathematics; European Mathematical Society (EMS): Zürich, Switzerland, 2010; Volume 11. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.