Reactive hydrogenation of interstellar carbon monoxide ices. The viewpoint of machine learning-driven simulations.

We investigated the hydrogenation of CO ices and the subsequent redistribution of the energy of reaction CO + H -> HCO by means of a reactive machine-learned potential trained on DFT data. Using that potential, we performed a set of molecular dynamics simulations on amorphous and crystalline CO ice models. From extensive sampling, we obtained binding energy distributions for the formyl radical (HCO) in both types of ices. During the simulations, we tracked the kinetic energy of the HCO radical and the ice matrix to characterize the pathways of energy redistribution, while validating the accuracy of the MLIP in real time. Our simulations showed that the reaction energy is unable to break the long-range ordering of crystalline CO ices but may produce local structural changes in the reaction sites that could develop into local amorphous regions within CO ices. Moreover, we observed the desorption of the nascent HCO radical from both crystalline and amorphous surfaces as well as long-range non-thermal diffusion in the amorphous one, supporting enhanced organic chemistry on these ices. The potential impact of these findings on current astronomical models is discussed.