How much faster does the best polynomial approximation converge than Legendre projection? (Q1996234)

The author compares the convergence behaviour in the maximum norm of Legendre projections \(\mathcal{P}_n(f)\) with best polynomial approximations \(\mathcal{B}_n(f)\) and Chebyshev projections \(\mathcal{T}_n(f)\) for some classes of functions \(f\). The aim is to evaluate how much better is the accuracy of \(\mathcal{B}_n(f)\) than \(\mathcal{P}_n(f)\) and \(\mathcal{T}_n(f)\). As cited from the references of the paper, some results on \(\mathcal{T}_n(f)\) have been already obtained in the literature. For continuous functions the maximum error of \(\mathcal{T}_n(f)\) is inferior to that of \(\mathcal{B}_n(f)\) by at most a logarithmic factor, while for \(\mathcal{P}_n(f)\) is inferior to that of \(\mathcal{B}_n(f)\) by at most a factor of \(n^{1/2}\) and this fact also holds for more regular functions \(f\). Indeed for analytic functions the optimal rate of convergence of \(\mathcal{P}_n(f)\) is slower than that of \(\mathcal{B}_n(f)\) and \(\mathcal{T}_n(f)\) by a factor of \(n^{1/2}\). For differentiable functions, such as piecewise analytic functions, functions with derivatives of bounded variation and some functions of fractional smoothness, the optimal rate of convergence of \(\mathcal{P}_n(f)\) is the same as that of \(\mathcal{B}_n(f)\) and \(\mathcal{T}_n(f)\), i.e. the accuracy of \(\mathcal{P}_n(f)\) is inferior to that of \(\mathcal{B}_n(f)\) by only some constant factors. Finally the paper ends with some open problems for future research, such as the variation of pointwise convergence rate depending on the distance of the evaluation point from singularities of \(f\) and other types of projections, like Jacobi ones.