Matrix $A_p$ weights via $S$-functions (Q4336044)

Let \(W\) be an \(d \times d\) matrix weight, i.e., a \(L^1\) function whose values are self-adjoint, nonnegative \(d \times d\) matrices. The functions \(f\) considered below take values in \(C^d\). The author considers the inequality \[ \int_T (W(t)^{2/p} Hf(t), Hf(t))^{p/2} dm(t) \leq C \int_T (W(t)^{2/p} f(t), f(t))^{p/2} dm(t) \] and finds conditions on the matrix weight \(W\) under which it holds. The second term defines the norm on the space \(L^p(W)\). When \(d=1\), the necessary and sufficient condition is the Hunt-Muckenhoupt-Wheeden condition: \[ \sup_I \left( \frac1{|I|} \int_I W \right)^{1/p} \cdot \left( \frac1{|I|} \int_I W^{-q/p} \right)^{1/q} < \infty \] where \(q = p/(p-1)\) and the supremum is taken over all intervals. In \textit{S. Treil} and \textit{A. Volberg} [J. Funct. Anal. 143, No. 2, 269-308 (1997)] is given a necessary and sufficient condition in the vector valued case for \(p = 2\), namely \[ \sup_I \left|\Biggl( \frac1{|I|} \int_I W \Biggr)^{1/2} \cdot \Biggl( \frac1{|I|} \int_I W^{-1} \Biggr)^{1/2} \right|< \infty, \] where \(W^{1/2}\) stands for the operator power of nonnegative operators. The matrix-valued \(A_p\) condition (for all \(p\)) is defined in terms of metrics associated with \(W^{1/p}(t)\) as follows (there is also an equivalent form involving the matrices that define the metric \(\rho\) on \(C^d)\) -- \(\rho_t\) is a family of norms on \(C^d\), which define \[ L^p(\rho) = \left\{ f: (\int \rho_t(f)^p dt)^{1/p} < +\infty \right\}. \] The dual metric of a metric \(\rho\) is \(\rho^*\) and it is defined by \[ \rho^*(x) = \sup_{y \neq 0} \frac{|(x,y)|}{\rho(y)}, \] and for the \(\rho_t\), \[ \rho_{p,I}(x) = \left( \frac1{|I|} \int_I \rho_t^p(x) dt \right)^{1/p}, \] while \(\rho^*\) denotes the corresponding integral for the family \(\rho^*_t\). One always has \(|(x,y)|\leq \rho_{p,I}(x) \rho^*_{q, I}(y)\). When the reverse inequality holds, \[ \rho^*_{q, I}(y) \leq C (\rho_{p,I}(x))^* , \] \(\rho\) is called an \(A_p\) metric. When \(\rho^*_{q, I}(y)\) is replaced by \[ \rho^*_{0,I} = e^{\frac1{|I|} \int_I \log \rho^*_t(x) dt}, \] the definition above gives the \(A_{p, \infty}\) condition, i.e., if \(\rho\) satisfies \[ \rho^*_{0, I}(y) \leq C (\rho_{p,I}(x))^*, \;\forall \;I, \] then \(\rho\) is called an \(A_{p, \infty}\)-metric. By Jensen's inequality, \(A_p \Rightarrow A_{p, \infty}\). For the problem the author considers \(\rho_t(f) = |W^{1/p}(t)f|\), and \(\rho^*_t(f) = |U^{-1/q}(t)f|\), with \(W^{1/p}(t) = U^{-1/q}(t)\) for a.e. \(t\). He also considers the discrete Triebel-Lizorkin spaces \(f_p^{02}\), which consist of all sequences enumerated by the set of dyadic intervals \(\mathcal D\) such that \[ |s |^p_{f_p^{02}} = \int \left( \sum_{I \ni x} |s_I|^2 \frac1{|I|} \right)^{p/2} dx < \infty. \] He proves that \(g \mapsto \sum_{J \in \mathcal D} A_J^{-1} \left( \int gh_J \right) \cdot h_J\), where \(h_J\) is the Haar function on \(J\), is bounded from \(L^q(\rho^*)\) to \(f_q^{02}\) for \(\rho \in A_{p, \infty}\), and that \[ T_{\pm} \phi = \sum_{J \in \mathcal D} \pm h_J \int \phi h_J \] are uniformly bounded in \(L^p(\rho) \Leftrightarrow \rho \in A_p\), where \(\left( \int_I \rho_t(x)^p dx\right)^{1/p} \asymp |A_I(x)|\) defines the matrices \(A_I\). The result for \(T_{\pm}\) implies that any Calderón-Zygmund operator with \(T(1) = T^*(1) = 0\) is bounded on \(L^p(\rho)\) if \(\rho \in A_p\). He works out the reduction of the Hilbert transform to \(T_{\pm}\).