0n the best approximation by ridge functions in the uniform norm (Q5956459): Difference between revisions

The authors prove that the distance from \(W^{r,d}_p\) to \(M_n\) in the \(L_q\)-norm behaves like \(n^{-r/(d-1)}\). Here \(W^{r,d}_p\) and \(M_n\) are the following classes of functions of \(d\) variables (\(d\geq 2\)): \(W^{r,d}_p\) is an \(L_p\)-Sobolev unit ball of \(r\) times differentiable functions, \(M_n\) is the collection of all linear combinations of \(n\) ridge functions, where \(f\) is a ridge function if there is an \(a\in {\mathbb R}^d\) and a \(g: {\mathbb R} \to {\mathbb R}\) such that \(f(x) = g(ax)\). The special case for \(p=q=2\) was proved previously by the second author. Not only does this work extend that result to all \(2\leq q\leq p\), but also the asymptotic estimate applies to the uniform norm, \(p=q= \infty\).