Theoretical Methodology to predict Melting-Like Transition Modeling in Clusters using Artificial Neural Network.

Clusters with 100 or fewer atoms exhibit melting-like transitions fundamentally different from bulk melting. A better understanding of how clusters melt is essential for applications of cluster materials in electronics, especially catalysis. We are developing a methodology for the simulation and analysis of cluster melting. We did classical Parallel Tempering Monte Carlo simulations of argon clusters modelled with a 6-12 Lennard-Jones potential. The potential energy distribution, heat capacity, and a sample of cluster configurations were obtained at different temperatures. The configurations were represented by features based on the pair distribution function and arrangement of nearest neighbours. We determined the freezing temperature T_f and melting temperature T_m>T_f using an Artificial Neural Network (ANN) classifier. The ANN Classifier also helps characterize the solid-liquid coexistence and yields the solid fraction at every temperature. The ANN classifier is computationally efficient; 30 PTMC replica with 64000 steps gives nearly converged results.