Neuroevolution Potential-Driven Accurate and Efficient Discovery of Mechanical Behavior and Nanocluster Dynamics in TiAl Alloys

Abstract

Titanium-aluminum (TiAl) alloys have excellent characteristics such as low density, high oxidation resistance, high-temperature durability, and unique mechanical properties. While atomic-scale simulations can optimize their structural design, conventional methods face limitations: ab-initio molecular dynamics (AIMD) is computationally expensive and restricted to small systems, whereas classical molecular dynamics (MD) simulations, though scalable, suffer from reduced accuracy. To overcome these challenges, we developed a high-accuracy neuroevolution potential (NEP) for Ti-Al alloys, which outperforms traditional embedded atom method (EAM) potentials. The NEP demonstrates strong agreement with density functional theory (DFT) calculations, achieving training errors of 5.8 meV/atom for energy, 126.2 meV/Å for force, and 34.5 meV/atom for virial. Validation through MD simulations confirmed the NEP's reliability in predicting elastic constants, melting point, and radial distribution functions of TiAl alloys. Additionally, the NEP accurately reproduced the tensile properties of single-crystal TiAl, aligning with experimental data. Further application of the NEP revealed insights into the rapid solidification of TiAl alloys from liquid to glassy states, including cluster evolution during cooling and the strain-rate dependence of metallic glass mechanics. These results demonstrate that the NEP offers an efficient and precise framework for investigating the microstructure, thermodynamic behavior, and kinetic properties of TiAl alloys.