Extending Shannon's ionic radii database using machine learning
In computational material design, the ionic radius is one of the most important physical parameters to predict material properties. Motivated by the progress in computational materials science and material informatics, we extend the renowned Shannon's table from 475 ions to 987 ions. Accordingly, a rigorous machine learning (ML) approach is employed to extend the ionic radii table using all possible combinations of oxidation states (OS) and coordination numbers (CN) available in crystallographic repositories [1]. Then, to improve the accuracy of the predictions, we removed 24 detected outliers, which is 4.92 % of the data, and re-trained the ML model to provide better accuracy [2]. After that, we used the interatomic distances of binary compounds to estimate the negative ionic radii using a self-consistent calibration approach [3].
For positive ions, please cite:
[1] Baloch, Ahmer AB, et al. "Extending Shannon's ionic radii database using machine learning." Physical Review Materials 5.4 (2021): 043804.
[2] Mohammad Alsalman et. al.,“Outliers in Shannon’s Effective Ionic Radii Table and the the Table Extension by Machine Learning,” Computational Materials Science, vol. 228, p. 112350, 2023.
For negative ions, please cite:
[3] Mohammad Alsalman et. al., “Anions’ Radii—New data points calibrated to match Shannon’s table,” Computational Materials Science, vol. 244, p. 113153, 2024.