Predicting Local Environments of Entropy Stabilized Oxides Using DFT and Machine Learning

Entropy Stabilized Oxides (ESOs) are a novel class of materials which are enthalpically unfavorable, but entropically favorable due to high configurational disorder. The unique properties of these materials make them potentially useful as battery cathodes and in thermoelectric devices. Though not able to directly predict formation energies of ESOs, enthalpy based methods such as Density Functional Theory (DFT) remain useful for gathering bond length data, oxidation states, electronic band structure, and other statistics from the microstates representing the local environments of these materials. These statistics are useful for comparison to experimental methods such as XAS, provided that the microstates are representative of the real material. For representative systems, DFT can be quite computationally expensive, so instead we utilize Machine Learning (ML) algorithms to identify structural and energetic descriptors based on DFT. We aim to use these ML algorithms to scan through potential ESO candidates and predict which ones are the most promising for in-depth electronic structure calculations using the more accurate but computationally expensive DFT.