Molecular Simulation for Atmospheric Reaction Exploration and Discovery: Non-Equilibrium Dynamics, Roaming and Glycolaldehyde Formation Following Photo-Induced Decomposition of syn-Acetaldehyde Oxide

TL;DR

This study uses neural network-based molecular dynamics simulations to explore the non-equilibrium atmospheric reactions of syn-CH3CHOO, revealing insights into reaction dynamics, product formation, and bond strength sensitivities.

Contribution

It introduces a machine learning approach with reactive potentials to accurately simulate atmospheric reaction dynamics and product formation, including glycolaldehyde.

Findings

Quantitative agreement with experimental energy distributions.

Identification of O--O bond strength range D_e in [22,25] kcal/mol.

Observation of roaming dynamics leading to glycolaldehyde formation.

Abstract

The decomposition and chemical dynamics for vibrationally excited syn-CH$_3$CHOO is followed based on statistically significant numbers of molecular dynamics simulations. Using a neural network-based reactive potential energy surface, transfer learned to the CASPT2 level of theory, the final total kinetic energy release and rotational state distributions of the OH fragment are in quantitative agreement with experiment. In particular the widths of these distributions are sensitive to the experimentally unknown strength of the O--O bond strength, for which values $D_e \in [22,25]$ kcal/mol are found. Due to the non-equilibrium nature of the process considered, the energy-dependent rates do not depend appreciably on the O--O scission energy. Roaming dynamics of the OH-photoproduct leads to formation of glycolaldehyde on the picosecond time scale with subsequent decomposition into CH$_2$OH+HCO. Atomistic simulations with global reactive machine-learned energy functions provide a viable route to quantitatively explore the chemistry and reaction dynamics for atmospheric reactions.