Direct approximation on spheres using generalized moving least squares (Q1689314)

This paper presents a moving square method on the sphere of \(\mathbb R^d\) that differs from the classical one considering a different approach -- the so-called direct approach. Let \(X\) be the set of points in the sphere. The classical method is based on the approximation of a certain function \(u\) by a set of functions and requires the action of the operator on such functions. Here, the action of the operator on \(u\) is approximated considering the nodal values of \(u\) in the points of \(X.\) The existence and uniqueness of the generalized moving square approximation is established in Theorem 4.1 for a particular weight function without proof because, as the author mentioned, it is the same of the corresponding result in \(\mathbb R^d\). The application of the new method in the context of partial differential equations on the sphere is given in Section 6 and some numerical experiments are included in Section 7. The convergence properties of the method are analysed in Theorem 5.1 for the identity, gradient and Laplace-Beltrami operators, assuming that \(X\) is quasi-uniform. The proof of this result follows standard arguments and it is based on the fact that the coefficient functions of the approximation of the operators on \(u\) define a local polynomial reproduction in the sense of Theorem 3.4. It should be remarked that Theorem 3.4 has a central role in this paper. It establishes the existence of a stable local reproduction based on a set \(X\) of points on the sphere with specified properties. Its proof is based on Theorem 3.2 that establishes conditions that allow to conclude that \(X\) induces a norming set of functionals, Theorem 3.1 that provides a useful representation of any functional in terms of a norming set and Theorem 3.3 that establishes useful upper bounds for the gradient and Laplace-Beltrami operators.