The Overlapped Radial Basis Function-Finite Difference (RBF-FD) Method: A Generalization of RBF-FD

Abstract: We present a generalization of the RBF-FD method that computes RBF-FD weights in finite-sized neighborhoods around the centers of RBF-FD stencils by introducing an overlap parameter $\delta \in [0,1]$ such that $\delta=1$ recovers the standard RBF-FD method and $\delta=0$ results in a full decoupling of stencils. We provide experimental evidence to support this generalization, and develop an automatic stabilization procedure based on local Lebesgue functions for the stable selection of stencil weights over a wide range of $\delta$ values. We provide an a priori estimate for the speedup of our method over RBF-FD that serves as a good predictor for the true speedup. We apply our method to parabolic partial differential equations with time-dependent inhomogeneous boundary conditions-- Neumann in 2D, and Dirichlet in 3D. Our results show that our method can achieve as high as a 60x speedup in 3D over existing RBF-FD methods in the task of forming differentiation matrices.