| Abstract Scope |
Modern high-throughput databases and machine-learning (ML) models describe only ordered crystals, whereas alloys, electrodes, oxides, catalysts, and thermoelectrics exhibit substitutional or positional disorder, vacancies, and temperature-dependent ordering. This mismatch produces incomplete phase-stability landscapes, biased training data, and misleading synthesizability predictions. We present the Beyond-Order Workflow (BOW), a disorder-aware framework extending OQMD from ordered compounds to configurational ensembles and finite-temperature stability. BOW curates experimental disordered structures and maps partial occupancies using special quasi-random structures, finite ensembles, cluster-expansion sampling, or machine-learning-interatomic-potential-assisted simulations, selected by chemistry, structural complexity, and target property. Resulting first-principles data and ML models quantify mixing energetics, local chemical preferences, and site-specific substitutions, integrated with competing ordered phases and yield entropy-inclusive, temperature-dependent convex hulls. Balancing accuracy, cost, and uncertainty while documenting assumptions and convergence, BOW advances an “OQMD 2.0” where disordered phases become first-class thermodynamic objects rather than arbitrary ordered approximants, improving stability prediction, synthesis planning, and physically grounded AI. |