Simulation of the first steps of β-pinene ozonolysis and attempt of geometry optimization with reinforcement learning

β-pinene is the second most abundant biogenic monoterpene released into the atmosphere. Its ozonolysis reaction produces non-photolytic hydroxyl radicals and secondary organic aerosols (SOA), both crucial in atmospheric chemistry. We constructed a potential energy surface (PES) for our stereochemically diverse open shell and closed shell pathways with Density Functional Theory (DFT) and then utilized Rice-Ramsperger-Kassel-Marcus/Master Equation (RRKM-ME) simulations in the MultiWell suite to predict the reaction outcomes of β-pinene ozonolysis under tropospheric conditions. We identified a novel source of hydroxyl radical (~9.5%) before forming Criegee Intermediates (CI) at 298.15 K and 15 torr. This research also presents a first attempt at implementing a deep reinforcement learning model to reduce the computational cost of geometry optimization calculations.