Non-uniform painless decompositions for anisotropic Besov and Triebel-Lizorkin spaces (Q1929191)

The main results of the paper, Theorems 5 and 7, provide equivalent sequence space norms for certain anisotropic spaces of Besov and Triebel-Lizorkin type. The authors use the symbols \({\mathbf{E}}^{\alpha,q}_p\) and \({\mathbf{e}}^{\alpha,q}_p\) to denote, respectively, the function and sequence spaces corresponding either to Besov norms or Triebel-Lizorkin norms. In the Besov space case, \({\mathbf{E}}^{\alpha,q}_p\) and \({\mathbf{e}}^{\alpha,q}_p\) are, respectively, the homogeneous Besov space \(\dot{B}^{\alpha,q}_p\) and the sequence space \(\dot{b}^{\alpha,q}_p\) defined in terms of the respective function and sequence norms \[ \begin{aligned}\|f\|_{\dot{B}^{\alpha,q}_p}^q&=\sum_{j\in\mathbb{Z}} |\det{A}|^{-j(\alpha+\frac{1}{2})q}\Bigl\| f\ast D_{A^j}\varphi\Bigr\|_{L^p}^q; \\ \|\{a\}_{j,k}\|_{\dot{b}^{\alpha,q}_p}^q&= \sum_{j\in\mathbb{Z}} |\det{A}|^{-j(\alpha+\frac{1}{2})q}\Bigl\||a_{j,k}|\chi_{A^j([0,1]^d+k)}\Bigr\|_{L^p}^q\, .\end{aligned} \] In the Triebel-Lizorkin space case, \({\mathbf{E}}^{\alpha,q}_p\) and \({\mathbf{e}}^{\alpha,q}_p\) are, respectively, the homogeneous Triebel-Lizorkin function and sequence spaces \(\dot{F}^{\alpha,q}_p\) and \(\dot{f}^{\alpha,q}_p\) defined in terms of the norms \[ \begin{aligned}\|f\|_{\dot{F}^{\alpha,q}_p}&=\Bigl\|\Bigl(\sum_{j\in\mathbb{Z}} |\det{A}|^{-j(\alpha+1/2)q}| f\ast D_{A^j}\varphi|^2\Bigr)^{1/q}\Bigr\|_{L^p} \\\text{and}\\ \|\{a\}_{j,k}\|_{\dot{f}^{\alpha,q}_p}&= \Bigl\|\Bigl(\sum_{j\in\mathbb{Z}} |\det{A}|^{-j(\alpha+1/2)q}\sum_{k\in\mathbb{Z}^d}|a_{j,k}|^q\chi_{A^j([0,1]^d+k)}\Big)^{1/q}\Bigr\|_{L^p}.\end{aligned} \] In all cases, standard modifications are made for \(q=\infty\) and less standard ones for \(p=\infty\). Here, \(A\) is a \(d\times d\) expansive matrix, i.e., it has \(d\) eigenvalues with magnitudes larger than one and \(D_A f(x)=|\det{A}|^{-1/2}f(A^{-1} x)\). One writes \(\psi\in \mathcal{C}(A,\Lambda)\) if \(\psi\) has a Fourier transform \(h\) supported in some annulus about the origin and bounded below in modulus on a set of the form \(Q=(A^t V)\setminus V\). In this notation, \(\Lambda \) is a relatively separated set, i.e., a finite union of sets \(\Lambda_\nu\) such that any two points in \(\Lambda_\nu\) have a distance of at least some fixed \(\delta>0\). The gap \(\rho(\Lambda)\) is defined as \(\rho(\Lambda)=\sup_{x\in\mathbb{R}^d} \,\inf_{\lambda\in\Lambda}|x-\lambda|\). One assumes that \(\rho(\Lambda)<2/\operatorname{diam}\operatorname{supp} h\). For \(\psi\in \mathcal{C}(A,\Lambda)\), set \(\Psi=\{\psi(\cdot;j,\lambda)=D_{A^j}T_\lambda\psi(\cdot):j\in\mathbb{Z},\, \lambda\in\Lambda\}\) with \(T_\lambda f(x)=f(x-\lambda)\). Theorem 5 states that, if \(\psi\in \mathcal{C}(A,\Lambda)\), then there exists a dual family \(\tilde{\Psi}\) of band-limited functions \(\tilde{\psi}_{j,\lambda}(\cdot)\), not typically dilates and translates of \(\psi\), such that, for each anisotropic Besov or Triebel-Lizorkin space \({\mathbf{E}}_p^{q,\alpha}\), \(0<p,q\leq \infty\) and \(\alpha\in\mathbb{R}\), the elements \(f\) of \({\mathbf{E}}_p^{q,\alpha}\) admit expansions \[ f=\sum_{j\in\mathbb{Z}}\sum_{\lambda\in\Lambda} \langle f,\, \psi(\cdot;j,\lambda)\rangle\, \tilde{\psi}_{j,\lambda} =\sum_{j\in\mathbb{Z}}\sum_{\lambda\in\Lambda} \langle f,\,\tilde{\psi}_{j,\lambda}\rangle \psi(\cdot;j,\lambda) \] with convergence in the sense of tempered distributions modulo polynomials, and with the norm equivalences \[ \|f\|_{{\mathbf{E}}_p^{q,\alpha}}\approx \bigl\| \{\langle f,\, \psi(\cdot;j,\lambda)\rangle\}_{j,\lambda}\bigr\|_{{\mathbf{e}}_p^{q,\alpha}} \approx \bigl\| \{\langle f,\, \tilde\psi_{j,\lambda}\rangle\}_{j,\lambda}\bigr\|_{{\mathbf{e}}_p^{q,\alpha}} . \] Theorem 7 has a similar conclusion, but instead of using band-limited \(\psi\) and associated \(\tilde{\psi}_{j,\lambda}\)s, one uses compactly supported \(\varphi\) and associated functions \(\widetilde{\Phi}=\{\widetilde{\varphi}_{j,\lambda}\}\subset {\mathbf{E}}_p^{q,\alpha}\) to give the desired expansions and norm equivalences. Here, \(\varphi\) is required to have vanishing moments up to some order depending on the indices of \({\mathbf{E}}_p^{q,\alpha}\) and a perturbation argument different from that used in the proof of Theorem 5 -- which uses balayage properties for Dirac measures -- is required.