| Abstract Scope |
Ferroic materials, particularly ferroelectrics and multiferroics, exhibit rich behaviors driven by strong electron correlations, lattice instabilities, and coupled structural distortions. The complex interplay of competing interactions in oxides makes understanding the origin of functionalities a major challenge, directly impacting their suitability for advanced device applications. We combine causal modeling with conventional machine learning (ML) to uncover and quantify the fundamental drivers of ferroelectricity in perovskite oxides—enabling discovery of mechanisms that remain hidden in correlation-based approaches. Our analyses identify trilinear coupling between tilt, rotation, and A-site antiferroelectric displacements as the necessary condition for layered cation ordering, a precursor to hybrid improper ferroelectricity. Causal interventions, folded with physics-based constraints, reveal thermal pathways to control polarization switching, validated through first-principles simulations. Unlike standard ML methods, our approach provides interpretable, physics-consistent insights. While developed for ferroics, it generalizes across functional materials, integrating theory, experiment, and data-driven models for mechanistic discovery and materials design. |