School of Mathematical and Statistical Sciences

Computational and Applied Math Proseminar

Friday, March 25, 12:00 p.m. PSA 113

Jordan Martel

School of Mathematical and Statistical Sciences

Radial Basis Function Methods for Time-Dependent Problems on the Sphere

Abstract Radial Basis Function Pseudo-Spectral (RBF-PS) methods have become a popular alternative to traditional Fourier and Chebyshev spectral methods in the numerical solution of PDEs, particularly in non-standard geometries. Although RBF-PS methods demonstrate many of the desirable accuracy properties of traditional spectral methods, they often suffer from severe instabilies when solving time-dependent problems. Here, we examine how the flatness of the radial basis functions and the choice of square or least-square collocation schemes affects the stability of these methods in the context of time-dependent problems on the sphere. The linear advection equation and non-linear barotropic vorticity equation are used to study performance. We conclude that a least-squares collocation scheme using moderately flat basis functions results in the most desirable stability properties.