New Method Significantly Speeds Up Melting Temperature Predictions for Materials

Recent advancements in computational materials science have been made with the development of the SLUSCHI (Solid and Liquid in Ultra Small Coexistence with Hovering Interfaces) package. This automated tool interfaces with the VASP (Vienna Ab initio Simulation Package) to determine the melting temperatures of various materials. However, traditional methods often require extensive computational resources, taking weeks and consuming hundreds of thousands of CPU hours to complete molecular dynamics simulations necessary for accurate predictions.

In a new paper, researchers have successfully integrated the SLUSCHI package with the LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) code, which significantly accelerates the process of determining melting temperatures. The new approach reportedly outperforms the original VASP-based method by at least an order of magnitude in speed. The LAMMPS simulations utilize machine learning potentials that are pre-built using first-principles data, allowing for faster calculations without compromising accuracy.

The findings indicate that approximately 60% of the melting temperature predictions fall within 200 K of experimental values, with a root mean square error (RMSE) of 187 K. This performance is slightly less precise than the first-principles density functional theory (DFT) RMSE value of 151 K. The ability to rapidly screen materials for their melting temperatures can facilitate the rational design of new materials, aligning with the materials genome initiative, which aims to accelerate the discovery and deployment of advanced materials.

The authors of the paper, titled "Accelerating material melting temperature predictions by implementing machine learning potentials in the SLUSCHI package," include Audrey CampBell, Ligen Wang, and Qi-Jun Hong. The full paper can be accessed on arXiv at arXiv:2408.13416.