A machine learning based computational approach for prediction of cation distribution in spinel crystal

In this study, a machine learning based computational approach has been developed to investigate the cation distribution in spinel crystals. The computational approach integrates the construction of datasets consisting of the energies calculated from density functional theory, the training of machine learning models to derive the relationship between system energy and structural features, and atomistic Monte Carlo simulations to sample the thermodynamic equilibrium structures of spinel crystals. It is found that the support vector machine model yields excellent performance in energy predictions based on spinel crystal structures. Furthermore, the developed computational approach has been applied to predict the cation distribution in single spinel MgAl2O4 and MgFe2O4 and double spinel MgAl2-aFeaO4. Agreeing with the available experimental data, the computational approach correctly predicts that the equilibrium degree of inversion of MgAl2O4 increases with temperature, whereas the degree of inversion of MgFe2O4 decreases with temperature. Additionally, it is predicted that the equilibrium occupancy of Mg cations at the tetrahedral and octahedral sites in MgAl2-aFeaO4 could be tuned as a function of chemical composition. Therefore, this study presents a reliable computational approach that can be extended to study the variation of cation distribution with processing temperature and chemical composition in a wide range of complex multi-cation spinel oxides with numerous applications.