Restricted testing conditions for the multilinear maximal operator (Q2116386): Difference between revisions

This work investigates a testing condition bound for the multilinear (Hardy-Littlewood) maximal function \[\mathcal{M}(\vec{f})(x)=\mathcal{M}(f_1,\dots, f_m)(x)=\sup_{x\in Q}\prod_{i=1}^m \langle \vert f_i\vert \rangle_Q, \quad x\in\mathbb{R}^d, \] where \(\langle u\rangle_Q=\vert Q\vert ^{-1}\int_Q u\) is the average of \(u\) over \(Q\) and the supremum is taken over cubes in \(\mathbb{R}^d\) with sides parallel to the axes.\par One says that a weight pair \((\omega, \vec\sigma)=(\omega,\sigma_1,\dots,\sigma_m)\) satisfies the \(A_{\vec{p}}\) condition if \[[\omega,\vec\sigma]_{A_{\vec{p}}}=\sup_Q \langle \omega\rangle_Q \prod_{i=1}^m \langle \sigma_i\rangle_Q^{p/{p_i^\prime}}<\infty, \,\,\frac{1}{p}=\sum_{i=1}^m \frac{1}{p_i},\] and that \(\vec\sigma\in \mathrm{RH}_{\vec{p}}\) if, for some \(C>0\) and all cubes, \[\prod_{i=1}^m \langle \sigma_i\rangle_Q^{p/p_i} \leq C\langle \prod_{i=1}^m \sigma_i^{p/p_i}\rangle_Q\] (with \([\vec\sigma]_{\mathrm{RH}_{\vec{p}}}\) the infimum over such \(C\)).\par Let \(\Vert \mathcal{M}(\vec{\sigma}\cdot)\Vert_{\prod_{i=1}^m L^{p_i}(\sigma_i)\to L^p(\omega)}\) denote the smallest constant \(C\) such that for all \(f_i\in L^{p_i}(\sigma_i)\), \[\Vert \mathcal{M}(\vec{f\sigma})\Vert_{L^p(\omega)}\leq C\prod_{i=1}^m \Vert f_i\Vert_{L^{p_i}(\sigma_i)}^p\, .\] One says that \((\omega,\vec{\sigma})\) satisfies a \((\vec{p},\rho,D)\) doubling condition if there is a positive constant \(P\) (depending on \(\omega\), \(\vec{\sigma}\), \(d\), \(\vec{p}\), \(\rho,D)\) such that \[ \Vert \mathfrak{1}_Q\,\mathcal{M}(\vec{\sigma\mathfrak{1}_Q})\Vert_{L^p(\omega)}^p\leq P\prod_{i=1}^m \sigma_i(Q)^{p/p_i}\;\;\; (*)\] whenever there is a cube \(\tilde{Q}\supset Q\) and index \(i\) with sidelengths and \(\sigma_i\)-measures satisfying \(\ell (\tilde{Q})\geq \rho \ell (Q)\) and \(\sigma_i(\tilde{Q})\leq D\sigma_i(Q)\).\par With \(P\) as in the \((\vec{p},\rho,D)\) doubling condition, the main theorem states that if \(1<\rho<\infty\) and \(\vec\sigma\in \mathrm{RH}_{\vec{p}}\), for suitable \(D\) in the doubling condition, for weights \((\omega,\vec\sigma)\) one has \[[\omega,\vec\sigma]_{A_{\vec{P}}}+P\lesssim \Vert \mathcal{M}(\vec{\sigma}\cdot)\Vert_{\prod_{i=1}^m L^{p_i}(\sigma_i)\to L^p(\omega)} \lesssim ([\omega,\vec\sigma]_{A_{\vec{P}}}+P)[\vec\sigma]_{\mathrm{RH}_{\vec{p}}} \,.\] As in prior work cited, the proof is reduced to demonstrating the corresponding inequalities for cubes in a general dyadic grid in the following subcollections: (1) cubes that are of maximal sidelength relative to the test condition (*); (2) cubes that are close to the top of a sparse subfamily of cubes; and (3) cubes for which the left side of the inequality defining \([\vec\sigma]_{\mathrm{RH}_{\vec{p}}} \) is suitably small. One shows that these possibilities exhaust all cubes in the sparse subfamilies.