Artificial Neural Network Method to Construct Potential Energy Surfaces for Transition Metal Nanoparticles: Pt, Au, and Ag

Abstract

Potential energy surfaces (PESs) for transition metal nanoparticles of Pt, Au, and Ag were derived using the artificial neural network (ANN) method. Three feedforward neural networks were constructed to fit the nonlinear relationship between the binding energy and the nanoparticle information, i.e. size and atomic coordinates, based on the data obtained from density functional theory calculations. The test results demonstrated that the newly derived ANN PESs can successfully predict the binding energy at the local minima of the global potential energy surfaces. More promisingly, the ANN PESs may be used in the molecular dynamics simulations for studying transition metal nanoparticles that are larger in size than those being studied here.