Polynomial interpolation via mapped bases without resampling (Q2279893)

Polynomial interpolation and many other approximation methods (such as kernel-based approximation or radial basis functions) often suffer from their dependence on the given data points (``knots'') which includes deterioration of stability and error properties like Runge's or Gibbs phenomena if the knots are chosen badly. A trivial example for such usually bad choices are equally spaced points. There are several types of remedies for such difficulties: particular selection of data points from a larger amount given or choice of Chebyshev points etc. Unfortunately, one is not always free to use Chebyshev points, so the authors of the current paper suggest to map the given points by auxiliary functions into suitable data points. In other words, there is a composition of the bases of the desired approximation method (e.g., powers for polynomial interpolation) and artifical functions for each basis function that are composed with it. This can be achieved in two ways: the authors consider first the use of these compositions directly as basis functions for a new approximation algorithm. Alternatively, one can map the data first and then insert them into the classical methods (e.g., summing them up with the usual Lagrange functions for interpolation). The new maps have, of course, to satisfy some restrictions in order to avoid useless trivial results, and the authors require certain density properties that allow the use of Weierstraß-type theorems to show that convergence theorems of the new methods are still available.