Transformative discovery of rare-earth functional materials for energy and electronics via machine learning

Rare-earth elements, with their unique 4 f electronic configurations, impart exceptional optical, magnetic, catalytic, and hydrogen storage properties to functional materials, making them indispensable for next-generation energy and electronic technologies. However, the development of these materials is hindered by complex, non-linear relationships between composition, structure, and performance. Machine learning (ML) has emerged as a transformative paradigm to accelerate their discovery and optimization. This review systematically surveys the cutting-edge applications of ML to overcome longstanding challenges, such as navigating high-dimensional design spaces. We start by outlining the foundational ML workflow tailored for rare-earth systems, then extend to advanced methodologies, including graph neural networks for structural representation, multimodal learning for integrating heterogeneous data, and generative models for inverse design. Through representative case studies in catalysis, hydrogen storage, luminescence, and magnetism, we illustrate how these techniques decipher complex performance descriptors, predict key properties, and enable the data-driven creation of novel materials. Finally, we critically address prevailing challenges like data scarcity and model interpretability, proposing actionable strategies such as integrating physics-informed algorithms. This review provides a comprehensive roadmap for leveraging ML to unlock the full potential of rare-earth functional materials.