Stabilization of spline bases by extension (Q2135620)

Polynomial splines in one and more dimensions form good and useful bases for piecewise polynomial approximations of functions and data, but it is necessary to have stable vector space basis functions in order to provide reliable approximations inpractice. In this article piecewise polynomial basis functions and their stability properties are improved by so-called extension methods. They are not just standard pre-conditioning schemes (although those are useful too) because they are tailor-made improvements that are adapted to particular sorts of splines. In all cases, the basis functions are required to have compact support. In particular, they normally are not just tensor-product B-splines (although one can start the reformulation process from those) but they can end up as box-splines or so-called T-splines for example. They are formed by removing the unstable parts of the bases after beginning with the mentioned functions of local support. Of course they may only be removed or altered by taking linear combinations of other generating functions. Hierarchical splines are included as well. Many applications are mentioned with a particular emphasis on solving PDEs with splines in two or three dimensions. The approximation order of the employed function spaces must be retained in the process, so there are condition to preserve certain monomials given by the authors. Moreover, local support of the resulting function is always required as well. Trigonometric and exponential splines are also included in the novel algorithms for stabilisation that is studied by the authors. The condition numbers of the Gram matrices then using these new bases and their approximation order are central parameters in the authors' results and show the improved performance by the new approximations.