Automated construction of potential energy surfaces

Abstract

Methods to construct molecular potential energy surfaces through automated generation of ab initio electronic structure data are reviewed. Given a chosen method for fitting ab initio data (electronic energies represented at particular geometries) into an analytic surface, the questions of how best to select the data point locations and how to interface an electronic structure software package with fitting codes in parallel on a high-performance computing cluster are addressed. It is shown that methods based on interpolating moving least squares fitting are useful as they lend themselves to an algorithm which iteratively refines the fitted surface towards arbitrary accuracy. Several variants of the method are illustrated through examples including spectroscopic potentials for van der Waals systems, systems with high permutation symmetry, reactive systems, and systems with multiple coupled electronic states. An outlook identifying areas for future development is given.