Researchers have shown that scanning transmission electron microscopy (STEM) can be used to learn causal mechanisms – rather than mere correlations – of atomistic behavior in ferroelectric perovskite across ferroelectric–antiferroelectric phase transitions.1 Mechanistic knowledge of cause-and-effect relationships can disentangle complex emergent functionalities in ferroelectrics and is necessary for materials discovery, design, and optimization.
Functionalities of complex materials emerge from the interplay between multiple physical and chemical processes, such as polarization instabilities and cation ordering. Determining cause and effect is key. However, most machine learning methods are correlative, precluding identification of causal mechanisms. Here, researchers demonstrated that the combination of atomic-resolution electron microscopy and statistical learning yields an approach for learning the causal structure of atomic mechanisms of ferroelectric behavior in a perovskite across ferroelectric–antiferroelectric phase transition. First, atomically resolved STEM images were acquired for selected compositions of Sm-doped BiFeO3 from a combinatorial library. Extracted local compositional, structural, and polarization-field descriptors were then analyzed via a causal chain method, yielding cause-and-effect relationships between physical and chemical mechanisms and allowing disentanglement of complex emergent functionalities in ferroelectric materials.
