Artificial neural network potential energy surface for silver nanoparticles

A potential energy surface (PES) for describing the interactions among the atoms in Ag nanoparticles was derived using the feedforward artificial neural network (ANN) method. Based on the preliminary success of constructing ANN PESs using a small number of data sets for Pt, Au, and Ag clusters/nanoparticles, we studied here the accuracy of the ANN method to build the PES for Ag nanoparticles to be employed in molecular dynamics (MD) simulations by including more data sets obtained from density functional theory (DFT) calculations. In this work, more neurons were used to improve the fitting accuracy. The results demonstrated that the new fitting provides a more balanced result in terms of accuracy in training and testing with respect to the previously fitting, however, more asymptotic DFT data sets are required to construct a global ANN PES suitable for MD simulations on the formation of Ag nanoparticles.