[SY-D4] Toward a machine learning aided interatomic potential for multi-element alloys: Application to binary compounds
As the chemical complexity of novel material systems continues to increase, the need for the rapid development of predictive, scalable interatomic potentials has grown as well. Machine learning, neural networks, and other data-driven techniques has shown promise in condensing large amounts of data from first principles and density functional theory calculation into classical dynamical equations with linear scaling, making them ideal for molecular dynamics simulation. A formalism for the development of a multi-species interatomic potential is presented and applied to binary metals. The resulting potential is then analyzed in terms of predictive power, validity, performance, and development time.