Improved estimates for condition numbers of radial basis function interpolation matrices (Q1711822): Difference between revisions

While radial basis functions are a well-known and most useful tool for multivariate approximation and interpolation (e.g. because of their availability independent of the spatial dimension and for arbitrary node distributions), it is an accepted problem that interpolation matrices have high condition numbers. This begins to be critical when the matrices are large due to large numbers of nodes. One way to improve this situation is to precondition the matrix. Be that as it may, it is relevant to know good estimates on the mentioned condition numbers, i.e., for Euclidean norm estimates, on the eigenvalues (lower and upper). This paper improves on both kinds of estimates (there have been known results before), especially on the lower bounds of the smallest eigenvalue (in modulus) of the collocation matrix for conditionally positive kernels. Among other results it also provides bounds on the largest eigenvalues of interpolation matrices for strictly positive definite kernels.