The relation between \(n\)th minimal errors of two classes of information for approximation in Hölder space (Q2574307)

There are studied Hölder spaces \(H=C^{r,\alpha}([0,1]^d)\) and \(G=L_2([0,1]^d)\) with the inner product \[ \langle g,h\rangle_G=\int_{[0,1]^d}g(x)h(x)dx\;\forall g,h \in G. \] Let be \(S:H\rightarrow G, S(f)=f, f\in H\) the embedding operator. It is proved that \[ e_n(S,\Lambda^{std})\leq min_{k=0,1,\dots} \left(e_k(S,\Lambda^{all})^2+\frac{C.k}{n.n^{\frac{2(r+\alpha)}d}}\right)^{1/2}, \] where \(e_n(S,\Lambda^{std})\) and \((S,\Lambda^{all})\) are the minimal error of standard and linear information in the worst case, average case and randomized settings. \(C\) is a constant.