Constraint-Conditioned Closure Filtering for Reaction Pathway Prediction

Chemical reactions emerge from the interplay between quantum-mechanically admissible molecular pathways and contextual constraint regimes. This work reframes reaction prediction by separating substrate-defined possibility from context-conditioned realization. A two-stage filtering architecture—accessibility and closure compatibility—is introduced to efficiently prune combinatorial reaction spaces prior to computationally expensive electronic-structure calculations. Accessibility filtering evaluates environmental admissibility, while closure compatibility quantifies dynamic viability through four interacting factors: intermediate persistence, sequential continuity, resistance to competing dissipation, and basin stabilization. A toy branching model demonstrates the separation between accessibility and realizability, and a case study on CO hydrogenation over Ni versus TiO₂₋ₓ/Ni catalysts shows how constraint regimes reorganize closure-compatible pathways, shifting selectivity toward C₂₊ hydrocarbons in agreement with experimental observations. The framework provides a scalable, context-aware pre-pruning layer compatible with quantum chemistry, microkinetic modeling, and machine-learning-based reaction prediction, advancing toward efficient and realistic computational chemistry workflows.