Revealing Hidden Patterns through Chemical Intuition and Interpretable Machine Learning: A Case Study of Binary Rare-Earth Intermetallics RX

Machine learning algorithms have been applied successfully in many areas of materials chemistry but often suffer from an inability to extract chemical insight. To demonstrate that an approach combining machine learning and chemical intuition can be effective in generating interpretable models, the structures of binary equiatomic rare-earth intermetallics RX, whose relationships have long defied understanding, were investigated as a case study. A structure map was developed based on only two parameters, which are derived from simple elemental descriptors (atomic number, period and group numbers, radii, and electronegativity) of the R and X components. This map reveals the previously hidden patterns of structural regularities of RX intermetallics. It explains the preference for CsCl-, TlI-, or FeB-type structures among these compounds, predicts the structures of missing members, rationalizes the occurrence of metastable phases and polymorphic transformations, and offers strategies for structure stabilization through the addition of a third component.