Two-level-split decomposition of anisotropic Besov spaces (Q2267394): Difference between revisions

Let \(\theta\) be an ellipsoid in \(\mathbb R^n\). Let \(\Delta^k_{h,\theta} f(x)\) with \(h \in \mathbb R^n\), \(k \in \mathbb N\), be the usual differences if \([x, x+kh] \subset \theta\) for the connecting line segment of \(x\) and \(x+kh\). Otherwise, \(\Delta^k_{h,\theta} f(x) =0\). Let \[ \omega_k (f,\theta)_p = \sup_{h \in \mathbb R^n} \| \Delta^k_{h, \theta} f \, | L_p (\mathbb R^n) \|, \quad 0<p \leq \infty, \] be the \(L_p\)-moduli of smoothness on \(\theta\). Let \(\Theta = \bigcup_{m \in Z} \Theta_m\) be a discrete multilevel ellipsoid covering of \(\mathbb R^n\) such that \(\Theta_m\) consists of ellipsoids \(\theta\) with \(|\theta| \sim 2^{-am}\) for some \(a>0\) and \(\mathbb R^n = \bigcup_{\theta \in \Theta_m} \theta\). Compatibility conditions both for ellipsoids on the same level \(\Theta_m\) but also for neighbouring levels are required. Classical examples are canonical sets of dyadic balls (isotropic case) or ellipsoids respecting the axes of coordinates (classical anisotropic case). Let \(\alpha >0\) and \(0<p,q \leq \infty\). The authors introduce homogeneous and inhomogeneous Besov spaces related to \(\Theta\) as subspaces of \(L^{loc}_p (\mathbb R^n)\), quasi-normed by \[ \| f \, | \dot{B}^\alpha_{pq} (\Theta) \| = \Big( \sum_{j \in Z} \Big( \sum_{\theta \in \Theta_j} |\theta|^{- \frac{\alpha p}{n}} \omega_k (f, \theta)^p_p \Big)^{q/p} \Big)^{1/q} \] and \(\| f \, | B^\alpha_{pq} (\Theta) \|\) with \(j \in \mathbb N\) in place of \(j \in \mathbb Z\). Here, \(k \in \mathbb N\) with \(k > c \alpha\) for some \(c>0\) (depending on \(\Theta\)). This generalizes the classical isotropic and the classical anisotropic spaces. It is the main aim of this paper to develop the theory of these spaces. This includes comparisons with distributionally defined spaces and homogeneous spaces on quasi-metric sets, nonlinear \(m\)-term approximations, Jackson estimates, as well as discussions of improved smoothness of characteristic functions.