Littlewood-Paley characterization and duality of weighted anisotropic product Hardy spaces (Q2253116)

The article concerns different characterizations for weighted anisotropic product Hardy spaces and their dual spaces on \(\mathbb{R}^n \times \mathbb{R}^m\). It is a continuation of the earlier work [Math. Nachr. 283, No. 3, 392--442 (2010; Zbl 1205.42021)] by \textit{M. Bownik} et al. On \(\mathbb{R}^n\) the anisotropic setting arises from replacing the standard isotropic length scaling by an \(n\times n\) dilation matrix \(A\) with eigenvalues of modulus greater than 1. In this transition the Euclidean balls \(B(x,2^k)\) of the isotropic setting are replaced by balls \(B_\rho (x,b^k)\) with respect to a quasi-norm \(\rho\) and a scaling factor \(b=|\det (A)|\) associated with \(A\). The Lebesgue measure of \(B_\rho (x,b^k)\) is then comparable with \(b^k\). The class of Muckenhoupt weights \(w\) is defined in the anisotropic case by requiring that for some \(q\in (1,\infty)\), \[ \sup_{x\in \mathbb R^n} \sup_{k\in \mathbb Z} \Big( b^{-k} \int_{B_\rho (x,b^k)} w(y) \, dy \Big) \Big( b^{-k} \int_{B_\rho (x,b^k)} w(y)^{-\frac{1}{q-1}} \, dy \Big)^{q-1} < \infty . \] The product domain setting, on the other hand, adds to \(\mathbb{R}^n\) an another `independent' variable from \(\mathbb{R}^m\). The related anisotropic structure is then determined by a pair \(\vec{A} = (A_1,A_2)\) of dilations acting on \(\mathbb{R}^n\) and \(\mathbb{R}^m\), respectively. The class of product weights on \(\mathbb{R}^n\times\mathbb{R}^m\) is defined by requiring that, in both variables separately, the weights satisfy a uniform anisotropic Muckenhoupt condition. The associated quasi-norms and scaling factors are denoted by \(\rho_i\) and \(b_i\). Given a tensor product \(\Phi = \Phi^{(1)} \otimes \Phi^{(2)}\) of two Schwartz functions with vanishing integrals on \(\mathbb{R}^n\) and \(\mathbb{R}^m\), the authors consider the anisotropic product Lusin-area function of a tempered distribution \(f\) on \(\mathbb{R}^n \times \mathbb{R}^m\), which they define as the `averaged' \(g\)-function \[ \vec{S}_\Phi (f)(x) = \Big( \sum_{k_1,k_2 \in \mathbb Z} b_1^{k_1}b_2^{k_2} \int_{B_{\rho_1}(x_1,b_1^{-k_1}) \times B_{\rho_2}(x_2,b_2^{-k_2})} |f\ast \Phi_{k_1,k_2}(y) |^2 \, dy \Big)^{1/2} , \] where \(\Phi_{k_1,k_2}(x) = b_1^{k_1}b_2^{k_2}\Phi (A_1^{k_1}x_1 , A_2^{k_2}x_2)\) and \(x=(x_1,x_2)\in\mathbb{R}^n\times\mathbb{R}^m\). The weighted anisotropic product Hardy space \(\widetilde{H}_w^p(\vec{A})\) is then defined for \(p\in (0,1]\) by means of the quasi-norm \(\| f \|_{\widetilde{H}_w^p(\vec{A})} = \| \vec{S}_\Phi (f) \|_{L^p_w(\mathbb R^n \times \mathbb R^m)}\) with certain technical restrictions on \(\Phi\). This space was introduced in [loc. cit.] with a different choice of ambient space, and the first result of the article establishes the equivalence between that and a new, more suitable choice for an ambient space. The authors go on to introduce four auxiliary spaces for \(p\in (0,\infty)\). Firstly, they define the Hardy spaces \(\ddot{H}^p_w(\vec{A})\) by means of a Littlewood--Paley \(g\)-function so that \[ \| f \|_{\ddot{H}_w^p(\vec{A})} = \Big\| \Big( \sum_{j_1,j_2\in \mathbb Z} |\varphi_{j_1,j_2} \ast f|^2 \Big)^{1/2} \Big\|_{L^p_w(\mathbb R^n\times \mathbb R^m)}, \] where \(\varphi\) is a Schwartz function on \(\mathbb{R}^n\times\mathbb{R}^m\) with vanishing moments. Corresponding discrete versions \(\ddot{h}^p_w(\vec{A})\) consisting of sequences \(s = \{ s_R \}\) are defined by setting \[ \| s \|_{\ddot{h}^p_w(\vec{A})} = \Big\| \Big( \sum_{R\in\mathcal{R}} |s_R|^2|R|^{-1}\chi_{R} \Big)^{1/2} \Big\|_{L^p_w(\mathbb R^n\times \mathbb R^m)} , \] where \(\mathcal{R}\) is a collection of dilated rectangles. Secondly, they define the weighted anisotropic product Campanato spaces \(\mathcal{L}_{p,w}(\vec{A})\) by means of \[ \begin{multlined} \| f \|_{\mathcal{L}_{p,w}(\vec{A})} =\\ \Big( \sup_\Omega \frac{1}{w(\Omega)^{\frac{2}{p} - 1}} \int_\Omega \sum_{j_1,j_2\in \mathbb Z} \sum_{R\in\mathcal{R},\, R\subset\Omega,\, R \sim (j_1,j_2)} |\varphi_{j_1,j_2} \ast f(x)|^2 \frac{|R|^2}{w(R)^2} \chi_R(x)w(x) \, dx \Big)^{1/2} ,\end{multlined} \] where the inner summation is restricted to rectangles \(R\) of scale \((-j_1,-j_2)\), and their discrete versions \(\ell_{p,w}(\vec{A})\) by \[ \| s \|_{\ell_{p,w}(\vec{A})} = \Big( \sup_\Omega \frac{1}{w(\Omega)^{\frac{2}{p} - 1}} \sum_{R\in\mathcal{R}, R\subset\Omega} |s_R|^2 \frac{|R|^2}{w(R)^2} \Big)^{1/2} . \] In both expressions the supremum runs over all open sets \(\Omega\subset\mathbb{R}^n\times\mathbb{R}^m\) with \(w(\Omega) < \infty\). The second result of the article shows the boundedness of a \(\varphi\)-transform \(S_\varphi\) (and its inverse) from \(\ddot{H}_w^p(\vec{A})\) to \(\ddot{h}_w^p(\vec{A})\) and from \(\mathcal{L}_{p,w}(\vec{A})\) to \(\ell_{p,w}(\vec{A})\). As a corollary, the spaces \(\ddot{H}_w^p(\vec{A})\) and \(\mathcal{L}_{p,w}(\vec{A})\) are independent from the choice of \(\varphi\). The main result concerning the duality \(\ddot{H}_w^p(\vec{A})^* = \mathcal{L}_{p,w}(\vec{A})\) for \(p\in (0,1]\) is established by first proving the discrete version \(\ddot{h}_w^p(\vec{A})^* = \ell_{p,w}(\vec{A})\). Finally, the equivalence between the Lusin-area function and the Littlewood--Paley \(g\)-function guarantees that \(\widetilde{H}_w^p(\vec{A})\) coincides with \(\ddot{H}_w^p(\vec{A})\) for \(p\in (0,1]\). The proofs make use of Christ's dyadic cubes with respect to the quasi-metric arising from \(\rho\), and an anisotropic version of Journe's covering lemma both of which played an important role already in [loc. cit.].