Chemical Physics Letters, Vol.693, 152-158, 2018
An algorithm to use higher order invariants for modelling potential energy surface of nanoclusters
In order to fit potential energy surface (PES) of gold nanoclusters, we have integrated bispectrum features with artificial neural network (ANN) learning technique in this work. We have also devised an algorithm for selecting the frequencies that need to be coupled for extracting the phase information between different frequency bands. We have found that higher order invariant like bispectrum is highly efficient in exploring the PES as compared to other invariants. The sensitivity of bispectrum can also be exploited in acting as an order parameter for calculating many thermodynamic properties of nanoclusters. (C) 2018 Elsevier B.V. All rights reserved.