This study employs Deep-Potential (DP) models with varying exchange-correlation (XC) functional accuracies, combined with large-scale two-phase molecular dynamics simulations, to determine the melting curves of key minerals ${\mathrm{SiO}}_2$ (coesite, stishovite, and $\beta$-stishovite), ${\mathrm{MgSiO}}_3$ (perovskite and postperovskite), and ${\mathrm{CaSiO}}_3$ perovskite under Earth's mantle conditions. This study quantitatively investigated the effect of XC functionals on the melting temperature of silicate phases, and we suggest the strongly constrained and appropriately normed (SCAN)-based DP models as the optimal choice. For ${\mathrm{MgSiO}}_3$, the melting curve predicted by the DP-SCAN model lies at the high-temperature end of the presently studied range, at approximately 6200 K at the core-mantle boundary. For ${\mathrm{CaSiO}}_3$, the DP-SCAN-based results clarify the discrepancy in melting temperatures up to 2000 K at the core-mantle boundary. And the results indicate that ${\mathrm{CaSiO}}_3$ perovskite is the most refractory silicate phase near the core-mantle boundary, with the melting temperature as high as 7000–8000 K. For ${\mathrm{SiO}}_2$, the previously reported steep increase in the Clapeyron slope of the melting curve due to the $\text{stishovite}\to \beta$-stishovite transition is not observed, and a new temperature-induced phase transition at high temperatures is a potential cause. Furthermore, this work provides insights into the crystallization sequence of midocean ridge basalts and underscores the importance of machine learning methods for understanding deep Earth processes.

Close

PrevNext

View figure in article

• Earth's interior
• High-pressure studies
• Liquid-solid phase transition
• Phase diagrams
• Machine learning
• Molecular dynamics