INTRODUCTION TO NEURAL NETWORKS FOR NON-LINEAR REGRESSIONS: POTENTIAL ENERGY SURFACE FITTING

The present work demonstrates how neural networks are used to do non-linear regressions. The technique is presented in a simple and didactic manner and applied to fit potential energy surfaces for the FeC molecule and for the reaction H + H2. It shows how to do the fitting for single- and multi-variable system providing examples and code that can be easily extended to many problems in chemistry. All the code used to perform the fitting and generate the results is available as a Jupyter Notebook, which can be used without neither installation nor configuration

Keywords:
neural networks; potential energy surface; Python; non-linear regression