A new ionic radii scale derived from perovskite oxides

Abstract

The objective of this work was to develop a new radii scale to better geometrically describe perovskite oxides and aid in high-throughput discovery efforts. We hypothesized that by creating a scale dependent on anion and structure we can increase the performance of geometric descriptors for the perovskite oxide family. We partitioned the bond lengths into radii using radius ratio rules and extended the scale using a ridge regression model to increase its coverage to include more elements, coordination environments and oxidation states. As a proof of concept, we compared our scale to the more universal Shannon radii scale and its use in the Goldschmidt and Tau tolerance factors to classify ABO₃ compositions as perovskite or not. Using the Goldschmidt tolerance factor with our scale we were able to classify (0.814 Matthews correlation coefficient, 0.914 wt F-Score, 0.910 Balanced Accuracy) an experimental dataset of 163 ABO₃ compositions as perovskite or nonperovskites. Scores of 0.715, 0.867, 0.838 and 0.664, 0.838, 0.790 for Goldschmidt and Tau tolerance factor using the Shannon radii scale were obtained, respectively. The observed marked improvement encouraged us to use our scale to predict a list of feasible ABO₃ perovskites. Our simple methodology is not constrained to perovskites only and can easily be used to create new radii scales for other ionic systems and provide better geometric descriptions.