Machine Learning Potential Development and High-temperature Property Calculation for High-entropy Boride Ceramics

Molecular dynamics simulations of high-entropy boride ceramics (HEBCs) in extreme high-temperature environments are constrained by limited accuracy and temperature stability of empirical force fields. In this work, a high-accuracy deep-learning potential (DP) was proposed and developed for (Hf0.2Zr0.2Ta0.2Ti0.2Nb0.2)B2 systems via first-principles calculations and deep learning method. It is shown that, through expanding datasets via the active learning strategy, the DP model stability under high-temperature conditions (i.e., ~3000 K) could be significantly enhanced. The developed DP achieves high accuracy while maintaining computational efficiency. Validation results from the developed DP manifest that predictions of the volumetric equation of state align well with first-principles calculations, demonstrating the model’s good scalability. The lattice constants and mechanical properties predicted by DP-enabled molecular dynamics simulations show excellent agreements with experimental observations, with relative errors within 2%. Furthermore, the simulations successfully reveal the anisotropic thermal expansion behavior of HEBCs and rectify the anomalous trends reported in previous research. Therefore, this developed DP model provides a reliable tool for atomic-scale simulations of high-entropy boride ceramics under extreme conditions, and holds significant scientific value for advancing the in-depth understanding of their high-temperature service behavior.