Interpolatory estimates, Riesz transforms and wavelet projections (Q515342)

The authors study orthogonal wavelet projections of the form \[ W^{(\epsilon)}(u)= \sum_{Q\in \mathcal S} \left\langle u, \varphi^{(\epsilon)}_Q \right\rangle \varphi^{(\epsilon)}_Q|Q|^{-1}, \quad u\in L^2(\mathbb R^n). \] Here the direction \(\epsilon\) is an element in the set \(\mathcal A=\{0,1\}^n\setminus \{(0,\ldots,0)\}\), \(\mathcal S\) is the set of all dyadic cubes, and \(\{\varphi^{(\epsilon)}_Q/\sqrt{|Q|} \,:\, Q\in \mathcal S, \epsilon \in \mathcal A\}\) is an orthonormal basis in \(L^2(\mathbb R^n)\) that satisfies decay and Hölder estimates as well as sectional estimates for \(\mathbb E_i(\varphi^{(\epsilon)}_Q)\) for \(i\in \{1\leq j \leq n\,:\, \epsilon_j=1\}\) where \(\mathbb E_i(f)\) denotes the integral of \(f\) with respect to the variable \(x_i\). These conditions guarantee that this basis is equivalent to the Haar system. The authors extend estimates on the directional projection \(W^{(\epsilon)}\) from the \(L^2\)-setting to \(L^p(\mathbb R^n)\) and an upper bound for \(\| W^{(\epsilon)}(u)\|_p\) is given in terms of \(\|u\|_p\), \(\|R_i\|_p\), and \(\|R_i(u)\|_p\) where \(R_i\) is the Riesz transform and \(\epsilon_i=1\). The highly nontrivial proof involves a number of intermediate results concerning e.g.\ rearrangements and the authors discuss the motivation for this work as well.