Machine learning for functional materials: insights from compositional, electronic, and dynamic processes

Abstract

Functional materials design frequently involves phenomena that span multiple length and time scales, where exhaustive exploration by first-principles calculations or experiments is impractical. Given the current capabilities of artificial intelligence, machine learning in this area typically serves less as a stand-alone predictor but more as a practical tool for guiding exploration within complex design spaces. After briefly reviewing key developments, this perspective adopts a case-driven approach and organizes machine learning applications in this field into three recurring levels: macroscopic compositional design, where the main challenge is combinatorial complexity; microscopic electronic and structural responses, where properties depend sensitively on subtle local variations; and dynamic or time-dependent processes, where behavior is governed by pathways or sequential signals. Representative examples show how each level benefits from different methodological choices and practical tools, ranging from data-driven screening and optimization to physics-aware surrogates and pathway-based or sequential learning. Examining these cases in parallel allows us to extract transferable principles, including physically informed feature design, data efficiency, and uncertainty awareness, while also clarifying the practical limitations of current approaches.