Artificial neural networks and their utility in fitting potential energy curves and surfaces and related problems

Abstract

Artificial intelligence (AI) and machine learning (ML) methods have touched practically all aspects of our life. Their utility ranges from separating different quality agricultural produce to facial recognition to guiding us through most steps in our day-to-day life. In this perspective article, we demonstrate the utility of artificial neural network (ANN) method in fitting potential energy curves and surfaces and point out the potential applications to predicting and analyzing dynamical observables. Although the regression methods seem to be successful in fitting potential energy surfaces using limited ab initio data, the ANN method yields accurate fits of surfaces when enough number of ab initio points on the potential energy surface become available. The possibility of utilizing the ANN method for fitting excitation function data is pointed out and the implications are discussed.

Graphical abstract

This perspective article illustrates how the artificial neural network can be used to interpolate accurately potential energy curves and surfaces for molecular systems and how the method can be extended to systems with avoided crossing of potential energy curves and to multidimensional excitation function data.