Weighted martingale multipliers in the non-homogeneous setting and outer measure spaces (Q887330)

The classical Haar system, which is an orthonormal basis in~\(L^2(\mathbb{R})\), is an unconditional basis in the weighted space \(L^2(\mathbb{R}, w\,dx)\) if and only if the weight satisfies the so called dyadic Muckenhoupt \(A_2\)~condition that reads as follows: \[ \sup_I\left( \frac{1}{|I|}\int_I w\,dx\right)\cdot\left(\frac{1}{|I|}\int_I w^{-1}\,dx\right):= [w]_{A_2}<+\infty, \] where \(I\) runs over the set of dyadic intervals. This result can be generalized to bases of Haar subspaces in \(L^2(\mathbb{R}^d,w\,dx)\). In the present paper, the authors investigate what happens in the non-homogeneous situation, a typical example being the standard dyadic lattice in \(\mathbb{R}^d\) but provided as underlying measure with an arbitrary Radon measure \(\nu\) in \(\mathbb{R}^d\). Then they consider Haar subspaces orthogonal to constants in~\(L^2(\nu)\) and want to describe weights~\(w\) for which the Haar subspaces form an unconditional basis in~\(L^2(w\,d\nu)\). They prove that in this general case the corresponding martingale \(A_2\) condition is necessary and sufficient for the system of a Haar subspace to be an unconditional basis in the weighted space \(L^2(w\,d\nu)\). In fact, they prove the stronger result that the unconditional basis constant admits an estimate which is linear in the \(A_2\) characteristic \([w]_{A_2}\), a fact that is new even for the case of the Lebesgue measure. The authors reduce the proof of the main result to weighted estimates for so-called martingale multipliers. It is worth noting that there is a difference between doubling \(\nu\) and non-doubling~\(\nu\). In the doubling situation the dyadic \(A_2\) condition is sufficient to conclude weighted estimates for more general martingale transforms than martingale multipliers. However, the authors provide an example of a martingale transform which is not bounded in \(L^2(w\,d\nu)\) with a non-doubling \(\nu\) in spite of \(w\) satisfying the \(A_2\) condition. On the other hand, the main result states that for the general non-homogeneous martingale multipliers \(T\) the estimate \[ \|T\|_{L^2(w)} \leq C\cdot [w]_{A_2 }\cdot \|T\|_{L^2} \] holds with an absolute constant \(C\). In order to state main results in a more precise way, consider a \(\sigma\)-finite measure space \((X,\mathfrak{S},\nu)\) with a filtration \(\mathfrak{S}_n\), \(n\in\mathbb{Z}\), \(\mathfrak{S}_n\subset \mathfrak{S}_{n+1}\), where each \(\mathfrak{S}_n\) is an atomic \(\sigma\)-algebra. This means that there exists a countable disjoint collection \(\mathcal{D}_n\) of sets of positive measure (atoms) such that every \(A\in \mathfrak{S}_n\) is a union of sets \(I\in\mathscr{D}_n\). Denote by \(\mathscr{D}=\bigcup\limits_n\mathscr{D}_n\) the collection of all atoms. For an atom \(I\in\mathscr{D}\) the children of \(I\) is the set \(\text{ch}(I)\) of atoms \(I'\in\mathscr{D}_{n+1}\), where \(n=\sup \{m: I\in \mathscr{D}_m\}\). Writing \(\langle f\rangle_I=\nu(I)^{-1}\int_If\,dw\) for the average on \(I\) of a measurable function \(f\) and \(E_If=\langle f\rangle_I\cdot\mathbf{1}_I\) one can define the martingale difference operator \[ \Delta_If=\sum_{I'\in\text{ch}(I)} E_{I'}f-E_If,\quad I\in\mathscr{D}. \] Let \(\mathscr{D}_{-\infty}\) denote the collection of atoms \(I\) of the form \(I=\bigcup\limits_k I_k\), where \(I_k\in \mathscr{D}_k\) and \(I_k\subset I_{k-1}\), and write \(\overline{\mathscr{D}}=\mathscr{D}\cup \mathscr{D}_{-\infty}^{\text{fin}}\), where \(\mathscr{D}_{-\infty}^{\text{fin}}\) are the atoms~\(I\) in \(\mathscr{D}_{-\infty}\) with finite measure (\(|\nu(I)|<+\infty\)). Also introduce the subspaces \(\overset{\circ}{D}_I:=\Delta_I L^{\infty} (\nu)\) and then the question is when the subspaces \(D_I^w=\text{clos}_{L^2(w)}\overset{\circ}{D}_I\) form an unconditional basis in \(L^2(w\,d\nu)\). From the general theory of bases the next result follows easily: \(\bigstar\) The following statements are equivalent: {\parindent=6mm \begin{itemize} \item[(i)] The system of subspaces \(\{D_F^w:=\text{clos}_{L^2(w)}\overset{\circ}{D}_I: I\in\overline{\mathscr{D}},\, \mathscr{D}_I^w\neq 0\}\) is an unconditional basis in \(L^2(w\,d\nu)\). \item [(ii)] The martingale multipliers \(T_{\sigma}\), \(\sigma=\{ \sigma_I\}_{I\in \mathscr{D}}\), \(|\sigma_I|\leq 1\), where \[ T_{\sigma}f=\sum_{I\in \mathscr{D}} \sigma_I\Delta_If, \] are uniformly bounded in \(L^2(w\,d\nu)\). \end{itemize}} Now the main result of the paper states the conditions under which statement ii) holds. \(\bigstar\) Let \(w\) be a weight such that \(w\not\equiv 0\) on any \(I\in \mathscr{D}_{-\infty}\). Let \(w\) satisfy the martingale~\(A_2\) condition \[ \sup_{I\in \mathscr{D}} \langle w\rangle_I\cdot \langle w^{-1}\rangle_I:=[w]_{A_2}<+\infty. \] Then all martingale multipliers \(T_{\sigma}\), \(\sigma=(\sigma_I)_{I\in \mathscr{D}}\), \(|\sigma_I|\leq -1\), are uniformly bounded: \[ \|T_{\sigma}\|_{L^2(w\,d\nu)} \leq C[w]_{A_2}, \] where \(C\) is an absolute constant. Even for the case of the Lebesgue measure as reference measure \(\nu\) the first power in the \(A_2\) constant is sharp. The authors prove also the remarkable fact that it is sufficient to check the weighted estimates for martingale multipliers on characteristic functions of atoms.