Two weight norm inequalities for the \(g\) function (Q477941)

In this paper the authors provide quantitative two weight estimates for the classical \(g\)-function. More precisely, given weights \(w\) and \(\sigma\), the authors show that there is a constant \( \mathcal{G}>0\) such that for all \(f\in L^2(\sigma )\) \[ \| g(f\sigma)\|_{L^2(w)}\leq \mathcal{G} \|f\|_{L^2(\sigma )}, \] if and only if there are positive constants \(\mathcal{A}_2\) and \(\mathcal{T}\) such that for all cubes \(I\subset\mathbb{R}^n\) with sides parallel to the axes (i) the two weights \(\sigma\), \(w\) on \(\mathbb{R}^n\) satisfy the joint Muckenhoupt \(A_2\) condition: \(\;\;\frac{\sigma (I)}{|I|}\frac{w (I)}{|I|} \leq \mathcal{A}_2,\) (ii) the following family of Sawyer's testing conditions hold: \[ \int\int_{Q(I)} \big (\nabla P_t(\sigma\mathbf{1}_I)(x,t)\big )^2dw\, tdt\leq \mathcal{T}^2\sigma(I), \quad Q(I):= I\times [0,\ell(I)], \] where \(\ell(I)\) is the sidelength of the cube \(I\). Moreover, if \(\mathcal{A}_2\) and \(\mathcal{T}\) are the best constants in (i) and (ii) then \(\mathcal{G}\simeq \mathcal{A}_2^{1/2}+\mathcal{T}\). Here \(P_tf\) is the Poisson extension of \(f\) to the upper half plane \(\mathbb{R}^{n+1}_+\) \[ P_tf(x) := \frac{\Gamma (\frac{n+1}{2})}{\pi^{(n+1)/2}}\int_{\mathbb R^n} f(x-y)\frac{t^{-n}}{(1+|y/t|^2)^{(n+1)/2}}dy, \] and the classical \(g\)-function is defined by \(g(f)(x) = \left ( \int_0^{\infty}|\nabla P_tf(x,t)|^2tdt\right )^{1/2}\), where \(|\nabla P_tf(x,t)|^2 := \left |\frac{\partial}{\partial t}P_tf(x)\right |^2 + |\nabla_xP_tf(x)|^2\). This theorem adds to an increasing collection of quantitative two weight estimates for singular operators, fractional integrals, maximal function, and now square functions. Chief among them is the first author's, and, in joint work with E. Sawyer, C.-Y. Shen, and I. Uriarte-Tuero [\textit{M. T. Lacey} et al., Duke Math. J. 163, No. 15, 2795--2820 (2014; Zbl 1312.42011)] and [\textit{M. T. Lacey}, Duke Math. J. 163, No. 15, 2821--2840 (2014; Zbl 1312.42010)], solution of the two weight problem for the Hilbert transform. The techniques used include random dyadic grids, averaging over good Whitney regions [\textit{H. Martikainen} and \textit{M. Mourgoglou}, Proc. Am. Math. Soc. 142, No. 11, 3923--3931 (2014; Zbl 1325.42015)], and a stopping-time construction. The authors spell out the corresponding result for Wilson's intrinsic square function [\textit{M. Wilson}, Weighted Littlewood-Paley theory and exponential-square integrability. Lecture Notes in Mathematics 1924. Berlin: Springer (2008; Zbl 1138.42011)]. In Definition 3.5 and thereafter the Poisson term \(P(K,\mathbf{1}_E\sigma)\) is introduced for \(K\) a dyadic cube and \(E\) a measurable set in \(\mathbb R^n\), but never defined in this paper. Here is the definition \[ P(K,\mathbf{1}_E\sigma) = \int_{\mathbb R^n}\frac{\ell{(K)}}{(\ell{(K)}+\, \text{dist}(x,K))^{n+1}} \mathbf{1}_E(x)\sigma(dx). \] The authors show the necessity of (i), the joint Muckenhoupt \(A_2\) condition, and in fact they show that \(\mathcal{A}_2\lesssim \mathcal{G}^2 \). Clearly (ii) is a localized testing condition on \(f=\mathbf{1}_I\), and \(\mathcal{T}\lesssim \mathcal{G} \). The authors actually show a more delicate necessary condition, the \textit{pivotal condition}, which they then show holds when (i) and (ii) hold: there is a finite constant \(\mathcal{P}\) such that for all dyadic cubes \(I^0\) \[ \sum_{\alpha\in\mathbb{N}}\sum_{K\in\mathcal{W}_{I_{\alpha}}} P(K,\mathbf{1}_{I^0}\sigma )^2w(K)\leq \mathcal{P}^2\sigma(I^0),\tag{pivotal} \] where \(\{I_{\alpha}:\alpha\in\mathbb{N}\}\) is any partition of \(I^0\) into dyadic cubes and \(\mathcal{W}_{I}\) denotes the maximal dyadic cubes \(K\subset I\) such that \(2^r\ell(K)\leq \ell(I)\) and dist\((K,\partial I)\geq \ell(K)^{\gamma}\ell(I)^{1-\gamma}\). It can be shown that these are the good cubes strongly contained in \(I\) which happen to be a Whitney collection in the sense that the cubes expanded by a fixed amount have bounded overlap. The authors show that under conditions (i) and (ii), the \textit{pivotal constant} \(\mathcal{P}\), defined to be the smallest constant such that ({pivotal}) holds, is bounded by \(\mathcal{P} \lesssim \mathcal{A}_2^{1/2}+\mathcal{T}\). The pivotal condition and constant are then the ones showing up in the stopping time used to prove the sufficiency of conditions (i) and (ii) for the boundedness of the classical \(g\)-function.