Localized Bases for Kernel Spaces on the Unit Sphere

Abstract

Approximation/interpolation from spaces of positive definite or conditionally positive definite kernels is an increasingly popular tool for the analysis and synthesis of scattered data and is central to many meshless methods. For a set of N scattered sites, the standard basis for such a space utilizes N globally supported kernels; computing with it is prohibitively expensive for large N . Easily computable, well-localized bases with "small-footprint" basis elements-i.e., elements using only a small number of kernels-have been unavailable. Working on S 2 , with focus on the restricted surface spline kernels (e.g., the thin-plate splines restricted to the sphere), we construct easily computable, spatially well-localized, small-footprint, robust bases for the associated kernel spaces. Our theory predicts that each element of the local basis is constructed by using a combination of only O((log N ) 2 ) kernels, which makes the construction computationally cheap. We prove that the new basis is Lp stable and satisfies polynomial decay estimates that are stationary with respect to the density of the data sites, and we present a quasi-interpolation scheme that provides optimal Lp approximation orders. Although our focus is on S 2 , much of the theory applies to other manifolds-S d , the rotation group, and so on. Finally, we construct algorithms to implement these schemes and use them to conduct numerical experiments, which validate our theory for interpolation problems on S 2 involving over 150,000 data sites.

The overlapped radial basis function-finite difference (RBF-FD) method: A generalization of RBF-FD

“…The derivatives of the local Lagrange functions give the RBF-FD weights on this stencil [40], and their integrals can be used to generate quadrature rules [25]. Unlike in [25], our goal is not to use the local Lagrange functions as approximants, but rather to use them to develop a stabilization procedure.…”

A Galerkin Radial Basis Function Method for Nonlocal Diffusion

“…There is evidence that the Lagrange functions for the thin plate splines can be replaced with local Lagrange functions, which are cheaper to construct than the Lagrange functions. The local property has been recently proven for spheres, and current work investigates these methods for domains in R n [7,8,10]. The local property is due to results on both the decay of the Lagrange functions away from their centers and the decay of the coefficients of the Lagrange functions.…”

“…RBF Galerkin methods have been investigated, but previously had difficulty with quadrature. In the setting of certain Riemannian manifolds such as the n-sphere S n , RBF methods can be extended and yield interpolation methods as well as collocation and Galerkin methods for solving partial differential equations on spheres [3,8,9,12,13,16].…”

Preconditioning for Radial Basis Function Partition of Unity Methods

“…In such circumstances, using preconditioners based on approximate cardinal basis functions computed on a reduced node set has been shown to be successful (see, e.g., [4,7,20,27,34]). In our case, we are starting from a local approximation, so the coefficient matrices are already sparse.…”