The multilinear strong maximal function (Q2430511)

The authors establish a natural \(m\)-linear analogue of the classical strong maximal function from Euclidean harmonic analysis. The paper contains the following four main results. (1) For a Muckenhoupt basis \(\mathcal{B}\) and \(\vec{P}:=(p_1,\dots,p_m)\) with \(p_j\in(1,\infty)\), \(j\in\{1,\dots,m\}\), and \(\frac 1p=\sum_{j=1}^m \frac 1{p_j}\), the authors prove that the multi(sub)linear operator \[ \mathcal{M}_{\mathcal B}(f_1,\dots,f_m)(x) :=\sup_{B\in\mathcal B,\, B\ni x}\prod_{j=1}^m \left\{\frac{1}{|B|}\int_B|f_j(y)|\,dy\right\},\quad x\in{\mathbb R}^n, \] is bounded from the products of weighted Lebesgue spaces \(\prod_{j=1}^m L^{p_j}(\omega_j)\) to \(L^{p,\infty}(\nu)\) provided that the \((m+1)\)-tuple of weights \((\nu,\vec{\omega})\) with \(\vec{\omega}:=(\omega_1,\dots,\omega_m)\) satisfies the bump \(A_{\vec{p},\mathcal{B}}\) condition for some \(r\in(1,\infty)\), namely, \[ \sup_{B\in\mathcal{B}}\frac{1}{|B|}\int_B\nu(x)\,dx \prod_{j=1}^m\left(\frac{1}{|B|} \int_B\omega_j^{(1-p_j')^r}(x)\,dx\right) ^{\frac{p}{(p_j')^r}}<\infty. \] (2) As applications, for \(1<p_1,\dots, p_m<\infty\) and \(\frac 1p=\sum_{j=1}^m \frac 1{p_j}\), the authors obtain that when \(\mathcal B=\mathcal R\), the family of all open rectangles with sides parallel to the coordinate axes, and \(\nu\) is taken to be the weight \(\nu_{\vec{\omega}}:= \prod_{j=1}^m\omega_j^{p/p_j}\), the multi(sub)linear strong maximal function \[ M_\mathcal{R}(\vec{f})(x) :=\sup_{R\in\mathcal R,\, R\ni x}\prod_{j=1}^m \left(\frac{1}{|R|}\int_R|f_j(y)|\,dy\right),\quad x\in{\mathbb R}^n, \] is bounded from \(\prod_{j=1}^m L^{p_j}(\omega_j)\) to \(L^{p,\infty}(\nu_{\vec{\omega}})\). (3) The authors also prove a sharp distributional endpoint estimate for \(\mathcal M_{\mathcal R}\), which is analogous to the classical one of \textit{B. Jessen, J. Marcinkiewicz} and \textit{A. Zygmund} [Fundam. Math. 25, 217--234 (1935; JFM 61.0255.01 and Zbl 0012.05901)]. Precisely, there exists a positive constant \(C\), depending only on \(m\) and \(n\), such that for all functions \(\{f_j\}_{j=1}^m\) and \(\lambda\in(0,\infty)\), \[ |\{x\in\mathbb{R}^n:\;\mathcal{M}_\mathcal{R} (\vec{f})(x)>\lambda^m\}|\leq C\left\{\prod_{j=1}^m\int_{\mathbb{R}^n} \Phi_n^{(m)}\left(\frac{|f_j(x)|} {\lambda}\right)\,dx\right\}^{\frac 1m}, \] where \(\Phi^{(m)}:=\overbrace{\Phi_n\circ\Phi_n\circ\dots\circ\Phi_n}^{m\text{ times}}\) and \(\Phi_n(t)= t[1+(\log^+t)^{n-1}]\) for all \(t\in(0,\infty)\). This estimate is sharp in the sense that one cannot replace \(\Phi_{n}^{(m)}\) by \(\Phi_{n}^{(k)}\) for \(k\leq m-1\). (4) Finally, the authors present a more general version of the bilinear Marcinkiewicz interpolation theorem. Indeed, they prove strong type bounds for a certain range of exponents starting from weak type bounds and a distributional estimate like the one that the strong maximal function satisfies.