Exploring Sampling and Size Effects in DFT Investigation of High Entropy Oxides

High-entropy oxides (HEOs) have the potential to improve lithium-ion batteries and thermoelectric materials, but experimental discovery and synthesis of novel HEO compositions can be time-consuming and expensive. HEOs are characterized by long-range order in the lattice structure, with minimal movement of the metal atoms away from their ideal positions, but short-range chemical disorder leading to distortion of the metal-oxygen bonds from one local environment to another. Using a combination of density-functional theory (DFT) and statistical mechanics, we can predict the distribution of metal-oxygen bond lengths, oxidation states, and enthalpies of formation over an ensemble of HEO samples. By increasing the supercell size used for our ensemble of DFT calculations we can increase the number of metal atoms per unit cell and therefore the length scale for sampling configurational disorder. We explore the effects of supercell size and number of samples on predicted formation temperatures, bond lengths, and electronic properties of several compositions of HEOs which form in a rocksalt structure. Our results will be used to train a machine learning model to identify promising compositions of novel HEOs, potentially allowing for more efficient, cost-effective synthesis of HEOs.