Approximation of differentiable and analytic functions by splines on the torus (Q2208264)

Spline functions are piecewise polynomial approximations that can be used in real spaces or on manifolds. The authors of this article use them for approximations on a torus in any dimension. There are several issue that need to be addressed in this context: (1) The distribution of the knots for the splines on the \(d\)-dimensional torus, (2) comparisons with other kinds of approximations which would be typical for approximating functions on the torus, and (3) establishing convergence results for different types of approximants. In this paper, all three questions are discussed. The first is addressed by considering optimal knot distributions on the multivariate torus with respect to the \(n\)-width of the approximations (i.e., minimising the \(n\)-width). For the second one, comparisons with the obvious choices of trigonometric functions as approximations are made. For the third issue, approximations using kernel functions \(K(\cdot)\) are used and convergence for approximants from Sobolev spaces or analytic functions are proved. Concretely, \(L^q\) convergence orders of approximations to Sobolev functions of the form \(K*\varphi\) with \(\varphi\) from \(L^p\) are found, \(K\) is the above kernel and \(*\) denoting standard convolution. A Dirichlet-type condition \(1/p-1/q\geq\frac12\) is imposed for the convergence estimates. For the so-called \(s_k\) splines which are similar in shape to interpolants using the ubiquitous multiquadric radial basis function, namely \[ s_k = c+\sum c_k K(\cdot-x_k),\] existence and uniqueness results are found.