Construction and application of deep learning potential for CaO under high pressure.

Dynamic evolution of CaO solidification at 2720 K (a)-(c), solid–liquid coexistence at 2730 K (d)-(f), and liquidation at 2740 K (g)-(i) using DPMD simulations. The region represented by the green shadow indicates the presence of the molten phase. The initial configuration consists of a liquid phase at 3300 K and a solid phase at 3000 K. [Display omitted] Although the physical properties of CaO have been extensively discussed as a crucial component of the Earth's mantle, there remains a significant discrepancy between the theoretically predicted melting temperature (T m) and experimental measurements. Therefore, further investigation of the melting properties of CaO is required to clarify these discrepancies. In this study, a deep learning potential (DP) model is used to perform large-scale molecular dynamics simulations with greater accuracy. The DP model accurately reproduces the lattice constants, elastic parameters and equations of state of CaO, in contrast to empirical potentials. The atomic structures of the solid and liquid states are also discussed using radial distribution functions. Using the DP model, the T m of CaO were obtained to be 3140, 2915, and 2730 K by single-phase method, void method, and solid–liquid coexistence method. This study discusses possible reasons for the discrepancy in the T m of CaO. Despite agreement with first-principles calculations, inherent defects in the pseudopotential lead to discrepancies between the simulation results and the experimental measurements in the DP model. In addition, the relative strength of atomic bonding can be assessed by comparing the Vickers hardness and the cohesive energy of MgO and CaO. It is evident that MgO has the lowest energy and is the most stable, suggesting a higher theoretical T m than CaO. [ABSTRACT FROM AUTHOR]