Adaptive Hermite Spectral Methods in Unbounded Domains

Abstract

Recently, new adaptive techniques were developed that greatly improved the efficiency of solving PDEs using spectral methods. These adaptive spectral techniques are especially suited for accurately solving problems in unbounded domains and require the monitoring and dynamic adjustment of three key tunable parameters: the scaling factor, the displacement of the basis functions, and the spectral expansion order. There have been few analyses of numerical methods for unbounded domain problems. Specifically, there is no analysis of adaptive spectral methods to provide insight into how to increase efficiency and accuracy through dynamical adjustment of parameters. In this paper, we perform the first numerical analysis of the adaptive spectral method using generalized Hermite functions in both one- and multi-dimensional problems. Our analysis reveals why adaptive spectral methods work well when a "frequency indicator" of the numerical solution is controlled. We then investigate how the implementation of the adaptive spectral methods affects numerical results, thereby providing guidelines for the proper tuning of parameters. Finally, we further improve performance by extending the adaptive methods to allow bidirectional basis function translation, and the prospect of carrying out similar numerical analysis to solving PDEs arising from realistic difficult-to-solve unbounded models with adaptive spectral methods is also briefly discussed.