First-principles study of the properties of plutonium oxides and their interfaces based on machine learning

YL Fang and JT Wang and ZX Jiang and N Gao and R Zhou and WX Guo and T Wang and RS Li, MATERIALS TODAY COMMUNICATIONS, 45, 112372 (2025).

DOI: 10.1016/j.mtcomm.2025.112372

Two machine learning force fields (MLFFs) were specifically constructed for plutonium oxides using the on-thefly machine learning method. These MLFFs achieve quantitative accuracy in reproducing first-principles energies and forces, with root-mean-square errors (RMSE) of 0.0405 eV/& Aring; and 0.1807 eV/& Aring;, respectively, meeting stringent benchmarks for atomic-scale simulations. The predictive reliability of the force fields was validated through calculations of elastic constants and phonon dispersion, confirming their ability to capture both mechanical and dynamical properties with high fidelity. In a case study of the PuO2/Pu2O3 interface system, our MLFFs revealed that oxygen atom diffusion drives structural reconstruction toward a lower-energy stable configuration, resulting in a reduction of the interfacial energy to 5.72 J/m2. Phonon analysis further demonstrated suppressed vibrational frequencies post-reconstruction, consistent with enhanced structural stability. These findings underscore the precision of the MLFFs in modeling plutonium oxides, providing atomic-scale insights into corrosion mechanisms and interfacial energetics.

Return to Publications page