Optimal recovery of isotropic classes of twice-differentiable functions defined on \(d\)-dimensional Euclidean space (Q2447630)

This paper generalizes the results of [\textit{V. Babenko} et al., J. Complexity 26, No. 6, 591--607 (2010; Zbl 1215.41001)] with respect to functions from compact domains to Euclidean space \(\mathbb{R}^d\). Namely, the authors consider a class of twice-differentiable multivariate functions under certain smoothness and boundedness conditions. They evaluate function values and gradients from members of this class at a discrete set of points and ask which method approximating the original functions given these evaluations has the best worst-case error. They then ask which set of sampling points which have a bounded ``average cardinality'' (similar to Beurling density) yield the optimal approximation error. They prove what the optimal errors are in both situations, with the former depending on the sampling set and the latter on the bound on average cardinality. In order to prove these theorems, they use the machinery from covering problems.