Study on Cu 50 Zr 50 Metallic Glass Based on Machine Learning Force Field: Correlation between Virial and Equipartition Theorem Violations and Atomic-Scale Dynamic Behaviors

Abstract The thermodynamic properties of amorphous materials are crucial. However, their lack of long-range order makes both vibrational theory and experimental characterization methods ineffective in studying the related microscopic behaviors. This fundamental limitation hinders a mechanistic understanding of how relaxation behaviors govern macroscopic properties. Traditional simulation methods such as empirical potential-based molecular dynamics and density functional theory can link relaxation behaviors and macroscopic statistical properties, but are often limited by trade-offs between scale and accuracy. In this work, we use a machine learning potential with Density Functional Theory-level accuracy to investigate the microscopic relaxation behavior and their influence on kinetic energy and Virial in large-scale Cu 50 Zr 50 metallic glass. Our simulations reveal two distinct types of atomic relaxation behaviors with different temperature dependencies, which influence the dynamics of the glass above and below the glass transition temperature. Furthermore, we observe a breaking of equipartition theorem and the Virial theorem during the evolution of Cu 50 Zr 50 metallic glass. The deviation from this relation gradually decays to zero over time as microscopic atomic relaxation processes are activated. These findings can provide a more accurate method for understanding the atomic-level mechanisms underlying the formation and evolution of glassy materials.