Two-dimensional Group-III nitrides (h-BN, h-AlN, h-GaN, and h-InN) exhibit great promise for electronic and optoelectronic applications due to their hexagonal structures, thermal stability, and wide bandgaps. However, exploring their structure and performance on a large scale faces significant challenges: traditional density functional theory (DFT) is limited by prohibitive computational costs for large systems, while classical molecular dynamics (MD) relies on empirical potentials that often lack sufficient accuracy to describe complex interactions.
Herein, this study employs the Deep Potential (DP) method to construct machine learning potentials (MLPs) that combine DFT accuracy with the efficiency of MD simulations. The developed DP models achieved DFT-level precision in energy and force predictions, accurately reproducing phonon dispersion and thermodynamic functions (free energy, heat capacity, entropy) across a 0–1200 K temperature range. Large-scale MD simulations revealed distinct mechanical behaviors: h-BN exhibits high strength but brittleness, whereas h-AlN and h-GaN demonstrate a favorable balance of strength and ductility. Furthermore, non-equilibrium MD simulations uncovered significant length-dependent thermal conductivity in h-BN and h-AlN attributed to long phonon mean free paths, contrasting with the lower, scattering-dominated thermal conductivities of h-GaN and h-InN.
This work entitled “Machine Learning Potentials for Property Predictions of Two-Dimensional Group-III Nitrides” was published on Acta Physico-Chimica Sinica (published on November 24, 2025).

DOI: 10.1016/j.actphy.2025.100224