Melting Behavior and Phase Stability of CaO from Neural Network Potentials: a Molecular Dynamics Study

Abstract

We investigate the melting behavior of calcium oxide (CaO) under extreme conditions, a problem that remains poorly constrained due to experimental limitations despite its relevance for geophysical and technological applications. We develop a Machine Learning Interatomic Potential (MLIP) for CaO with PANNA 2.0 and the LATTE descriptor, training it on a dataset of $\sim$12,000 configurations including solid, liquid, interfacial, and void-containing structures, extracted from ab-initio molecular dynamics data employing PBEsol exchange-correlation functional. We perform large-scale molecular dynamics simulations to compute the melting temperature at ambient pressure using both the void-nucleated melting (VNM) and two-phase coexistence (TPC) methods, obtaining $T_m=3055\pm11$ K and $T_m=2847\pm15$ K, respectively.\\ We calculate an enthalpy of fusion of $ΔH_f\sim73$ kJ/mol, in agreement with thermodynamic assessments and ab initio calculations. We also reproduce the thermal expansion and obtain a volume increase of $\sim$29% at Tm, consistent with the corresponding decrease in density extracted from spatially resolved number density profiles. Finally, we calculate the high-pressure melting curve of CaO up to 20 GPa, providing one of the very few computational determinations of this quantity to date. The results confirm that the overheating ratio $η$ is not constant under pressure, increasing from 17% at ambient pressure to 24% at 20 GPa, confirming previous findings and ruling out the assumption of a fixed overheating ratio. Our results establish MLIP-based simulations as a robust and efficient framework for investigating phase stability in ionic oxides and provide new insight into the melting behavior of CaO under extreme conditions.