Development of a Machine Learning Interatomic Potential for Zirconium and Its Verification in Molecular Dynamics

Area of Science:

• Materials Science
• Computational Materials Science
• Atomic-Scale Simulations

Background:

• Molecular dynamics (MD) simulations are crucial for understanding Zirconium (Zr) behavior under extreme conditions.
• Traditional empirical potentials struggle to accurately model complex Zr interactions, limiting simulation accuracy and generalizability.
• Limitations include fixed function forms and parameters, hindering the description of multi-body interactions and nonlinear deformation.

Purpose of the Study:

• To develop an accurate and generalizable Deep Potential (DP) model for Zirconium (Zr) using first-principles calculations and machine learning.
• To overcome the limitations of empirical potentials in simulating Zr's atomic-scale structural evolution and mechanical response.
• To enhance the reliability of MD simulations for Zr under diverse and complex service conditions.

Main Methods:

• Combined high-throughput density functional theory (DFT) calculations with machine learning to create the Zr Deep Potential (DP) model.
• Validated the developed DP model through extensive molecular dynamics (MD) simulations.
• Simulations covered key physical properties: lattice constants, surface energies, grain boundary energies, melting point, elastic constants, and tensile responses.

Main Results:

• The DP model demonstrated high consistency with DFT predictions for fundamental properties like lattice constants and melting point.
• Accurately captured atomic migration, structural evolution, and phase transformations under thermal excitation.
• Successfully reproduced plastic deformation, yielding, structural rearrangement, and stress-induced HCP to FCC phase transitions in Zr under large strains.

Conclusions:

• The developed Deep Potential model for Zirconium exhibits strong physical fidelity and numerical stability.
• It significantly improves the accuracy and generalization ability of MD simulations for Zr compared to traditional empirical potentials.
• The DP model provides a reliable tool for investigating Zr's atomic-scale behavior under complex conditions, including large deformations and phase transitions.