| Abstract Scope |
Artificial intelligence (AI) and physics-based approaches, including density functional theory (DFT) and Integrated Computational Materials Engineering (ICME), are rapidly advancing materials discovery and design. However, their integration remains limited by the absence of a unified physical framework. Here, we propose a thermodynamics-governed paradigm based on free-energy landscapes and the hierarchical (recursive) structure of entropy. By decomposing entropy into configurational and intrinsic contributions, materials behavior can be described consistently across length and time scales within a single ensemble under fixed external constraints. In this framework, physics-based methods construct thermodynamically grounded free-energy landscapes, while AI efficiently learns, explores, and optimizes them. This energy-centered, hierarchy-aware perspective provides a common foundation for linking data-driven learning with fundamental thermodynamic principles, enabling more predictive, interpretable, and scalable materials discovery and design. Recent developments of this framework are detailed in Refs. https://doi.org/10.1007/s11669-026-01248-0 and https://doi.org/10.1088/3050-287X/AE7E9B |