AI-accelerated search for global minimum configurations of Pt-Ni nanoclusters using embedded-atom method interatomic potential

Abstract

Platinum (Pt) and nickel (Ni) alloy nanoclusters are promising high-performance materials for use in modern industry because of their structural durability and thermal stability. Determining the most stable structure of Pt-Ni nanoclusters is crucial to understanding their structural and physicochemical characteristics. Investigating complex potential energy surfaces is hindered by the computing expense and sluggish convergence of classical structure-search methods. We overcame these restrictions by effectively exploring the configurational space and predicting low-energy structures of Pt-Ni nanoclusters using a deep reinforcement learning (DRL) framework based on the proximal policy optimization (PPO) algorithm with an embedded atom method (EAM) interatomic potential. In addition, machine learning interatomic potential (MACE) was used to undertake structural reoptimization and ensure accuracy and reliability. Density functional theory (DFT) energies were used to assess the real global minimum structures. The thermal stability of the anticipated global minimum structures was investigated using molecular dynamics (MD) simulations at 500 K. This comprehensive AI-based workflow with thermodynamic stability studies shows that the predicted Pt-Ni nanocluster structures possess robust stability and good high-temperature performance.