High Throughput Discovery of Inorganic Compounds*

With the development of machine learning (ML) tools, we can now automate the prediction of new inorganic compounds. Promising examples of ML tools include the Data Mining Structure Predictor (DMSP), [1] ionic substitution predictor, [2] and improved crystal graph convolutional neural network (iCGCNN). [3] Furthermore, we can now confirm the stability of these compounds using density functional theory (DFT) databases, such as the Open Quantum Materials Database (OQMD). [4] Leveraging these tools, we performed high throughput predictions of thousands of new inorganic compounds. We discuss each ML model and describe the new compounds. For example, we found that the DMSP and iCGCNN were well suited to discover metallic compounds, while the ion substitution successfully predicted ionic compounds. We also discuss specific areas of interest such as mixed anion compounds and double perovskites.
[1] Fischer, C. C., et al. Nat. Mater. 2006, 5, 641-646.
[2] Hautier, G., et al. Inorg. Chem. 2011, 50, 656-663.
[3] Park, C. W.; Wolverton, C. Phys. Rev. Mater. 2020, 4, 063801.
[4] Kirklin, S., et al. npj Comput. Mater. 2015, 1, 15010.