As prototypical strongly correlated materials, manganese oxides serve as key platforms for studying electronic correlations. Under high pressure, drastic changes in local coordination, bond lengths, and electronic screening modify the electron correlation strength. However, due to the difficulty in determining the Hubbard $U$ value to correctly describe the localized $d$-electron state for various structures in high-pressure simulations, previous works generally used fixed $U$ values to investigate the pressure-driven phase transition of manganese oxides, resulting in significant controversies. Here, a machine learning approach is developed to fit the Hubbard $U$ parameters for exploring the high-pressure properties of ${\mathrm{Mn}}_{\mathrm{y}}{\mathrm{O}}_{\mathrm{x}}$ and applied to construct convex hulls at pressures of 0–200 GPa. Several new high-pressure phases are found and long-standing controversies about the phase-transition sequence of MnO and ${\mathrm{Mn}}_2{\mathrm{O}}_3$ are resolved by high-throughput structure search and comparing x-ray diffraction patterns. This work provides new insights into the high-pressure behavior of the Mn-O system and contributes to advancing studies of transition metal oxides characterized by localized $d$ electrons.

Close

PrevNext

View figure in article

• High-pressure studies
• Strongly correlated systems
• DFT+U
• Machine learning