Efficient Automated Workflow for Radical Reaction Networks in Environmental Disinfectants: A Peroxyacetic Acid Case Study with Machine Learning Interatomic Potentials

Peroxy-type disinfectants initiate chain reactions upon activation, but the underlying mechanisms of organic radical formation remain difficult to fully elucidate. In this study, we combine Density Functional Theory (DFT) with machine learning-based interatomic potentials to automate the construction of the reaction network for peroxyacetic acid (PAA). Using double-hybrid functionals, wavefunction analysis, we reveal that peroxide bond cleavage is primarily driven by an electronic excitation from the HOMO to the LUMO+1 orbital, resulting in bond dissociation. A pretrained machine learning model, refined through active learning, efficiently captures reaction pathways in molecular dynamics simulations. Conventional gas-phase calculations often neglect solvent effects and environmental factors, such as explicit solvation and dissolved oxygen, both of which are essential for accurately predicting chemical reactivity. Notably, O₂ facilitates the formation of CH₃OO·, which in turn produces CH₂O and ·OH radicals, further propagating the radical network. N₂/O₂ aeration experiments further highlight the crucial role of oxygen in driving reactivity, consistent with our computational predictions. The integrated approach of this study can readily be extended to issues of reaction mechanisms under microscopic media in environmental systems. Accordingly, we provide a research framework and have developed accompanying software for use.