Restricted testing for positive operators (Q2238545)

The authors consider the problem of characterizing classical two-weight inequalities for positive operators (like the Hardy-Littlewood maximal function \(M\) or the fractional integral \(I_\alpha\)) in terms of testing conditions, restricted to cubes satisfying a doubling condition. In particular, given a pair \((\sigma, \omega)\) of weights (i.e., positive Borel measures) in \(\mathbb R^n\) and \(\Gamma >1\), \((\sigma, \omega)\) satisfies the \(\Gamma\)-testing condition for the maximal function \(M\) if there is a constant \(\mathfrak{T}_M(\Gamma)(\sigma, \omega)\) such that \[ \int_{Q}\left|M\left(\mathfrak{1}_{Q} \sigma\right)\right|^{2} \mathrm{~d} \omega \leq \mathfrak{T}_{M}(\Gamma)(\sigma, \omega)^{2}|\Gamma Q|_{\sigma}, \quad \text { for all } Q \in \mathcal{P}^{n} \] where \(\mathcal{P}^{n}\) is the collection of cubes in \(\mathbb R^n\) with sides parallel to the coordinate axes, and side lengths \(2^\ell\), for some \(\ell\in\mathbb Z\), and \(\Gamma Q\) denote the cube concentric with \(Q\) with \(\ell(\Gamma Q)= \Gamma \ell(Q)\). \(\mathfrak{T}_{M}(\Gamma)(\sigma, \omega)\) is the least constant in the above inequality and \(\mathfrak{T}_{I_\alpha}(\Gamma)(\sigma, \omega)\) is defined analogously. Another important definition is the following weaker testing condition, in which one need only test the inequality over cubes that are ``doubling'': Given a pair \((\sigma, \omega)\) of weights in \(\mathbb R^n\) and \(D, \Gamma> 1\), \((\sigma, \omega)\) satisfies the \(D\)-\(\Gamma\)-testing condition for the maximal function \(M\) if there is a constant \(\mathfrak{T}_{M}^D(\Gamma)(\sigma, \omega)\) such that \[ \int_{Q}\left|M\left(\mathfrak{1}_{Q} \sigma\right)\right|^{2} \mathrm{~d} \omega \leq \mathfrak{T}_{M}^D(\Gamma)(\sigma, \omega)^{2}| Q|_{\sigma}, \quad \text { for all } Q \in \mathcal{P}^{n}\text { with } |\Gamma Q|_{\sigma}\le D|Q|_\sigma. \] Unlike the cases of the classical two-weight theorem for the maximal function and fractional integrals in [\textit {E. Sawyer}, Stud. Math. 75, 1--11 (1982; Zbl 0508.42023)], where the testing condition is already sufficient for the boundedness of the maximal function and fractional integrals, these restricted testing conditions are not by themselves sufficient for the norm inequality: \[ \int_{\mathbb{R}^{n}}|M(f \sigma)|^{2} \mathrm{~d} \omega \leq \mathfrak{N}_M(\sigma, \omega)^{2} \int_{\mathbb{R}^{n}}|f|^{2} \mathrm{~d} \sigma, \quad \text { for all } f \in L^{2}(\sigma). \] The main result proved for \(M\) is the following: If \[ A_{2}(\sigma, \omega):=\sup _{Q \in \mathcal{P}^{n}} \frac{\sigma(Q)}{|Q|} \frac{\omega(Q)}{|Q|}<\infty \] and \(\Gamma>1\), then there is \(D>1\) depending only on \(\Gamma\) and the dimension \(n\) such that \[ \mathfrak{N}_{M}(\sigma, \omega) \approx \mathfrak{T}_{M}^{D}(\Gamma)(\sigma, \omega)+\sqrt{A_{2}(\sigma, \omega)} \] for all locally finite positive Borel measures \(\sigma\) and \(\omega\) on \(\mathbb{R}^{n}\). A similar result holds for the fractional integral operator. The proof for \(M\) requires the use of careful application of a probabilistic argument of the type pioneered in [\textit {F. Nazarov} et al., Acta Math. 190, No. 2, 151--239 (2003; Zbl 1065.42014)] and, for the fractional integral, involves Whitney decompositions at each threshold that deals with the tails of the kernels of the operator.