To develop Machine Learning Interatomic Potential to study High Entropy Alloys

Abstract

High-Entropy Alloys (HEAs) have shown remarkable properties and found application in various fields such as aerospace, chemical industries, power generation, and high-temperature applications due to their superior mechanical and thermal properties. However, many HEAs remain to be explored. In recent decades, computational approaches have accelerated the exploration of diverse HEAs. However, the lack of force fields for different compositions has been a challenge in studying new alloys. First-principles methods, such as Density Functional Theory (DFT), provide high accuracy but are restricted to a few hundred atoms and a short time scale due to computationally demanding electronic structure calculations of complex chemical environments and diverse atomic interactions in HEAs. Molecular Dynamics (MD) relies heavily on accurate interatomic potential, which is difficult to develop for multi-component alloys, such as HEAs. Machine Learning Interatomic Potential (MLIP) overcomes these difficulties by learning from high fidelity DFT data while balancing computational cost, thereby achieving DFT-level accuracy. In this thesis, we have developed a Message Passing Atomic Cluster Expansion (MACE) for MoTaNbWHf0.5Ru0.5 High Entropy Alloy to investigate its mechanical properties at high temperatures. The training set covers an optimal configuration space that includes ground state, thermal interactions, and defect physics. Potential has been deployed to investigate the error metrics, equation of state, and stress-strain analysis.