Artificial Neural Networks Fitting of Potential Energy Curves and Surfaces: The 1/R Conundrum

Abstract

Within the Born-Oppenheimer approximation, the potential energy of a molecular system is written as a sum of electronic energy and nuclear-nuclear repulsion energy terms. The potential energy surface (PES), computed ab initio, as a function of bond distances and bond angles, has traditionally been represented using analytic functions and/or interpolation methods. We show here that the ab initio computed electronic energy values of a molecular system can be fitted more accurately than the corresponding potential energy values using the artificial neural network methodology. The exact Coulombic internuclear repulsion energy can be added subsequently to the fitted electronic energy to obtain an accurate PES.