The infinite-dimensional widths and optimal recovery of generalized Besov classes (Q1872635)

The average dimension of a linear space \(L\) of functions defined on \({\mathbb R}^d\) is the dimension ``per unit of volume'' defined as the limit \[ \lim_{\alpha \to \infty} \inf {\dim(P_\alpha L)\over (2\alpha)^d}, \] where \(P_\alpha f(x)= \chi_\alpha(x)f(x)\) and \(\chi_\alpha(x)\) is the characteristic function of the cube \([-\alpha, \alpha]^d\). For example, the space of polynomial splines with equidistant knots defined on \({\mathbb R}\) is infinite dimensional but its average dimension is finite. Let \(X\) be a linear normed space of functions defined on \({\mathbb R}^d\). For \(\sigma >0\) and a centrally symmetric subset \(C \subset X\), the Kolmogorov \(\sigma\)-width is the quantity \[ {\widetilde {d}}_\sigma(C, X)=\inf_L \sup_{f \in C} \inf_{g \in L} \|f-g\|_X, \] where the infimum is taken over all subspaces \(L\) of the average dimension \(\leq \sigma\). The Gel'fand \(\sigma\)-widths are defined similarly. In the paper under review the authors find the order of the \(\sigma\)-widths of certain generalized Besov classes. In the particular case of the usual Besov classes \(S^r_{p\theta}\) the order is \(\sigma^{-r/d}\).