Credit: Image courtesy of Ulissi et al. Nature Communications 8, 14621 (2017)
Seemingly simple reactions, such as the reaction of syngas (carbon monoxide (CO) and hydrogen) on a rhodium surface to convert fossil and biomass resources into chemicals and fuels, can have hundreds of species and reactions within their reaction network. This creates thousands of potential reaction pathways, which would be very time consuming to individually investigate with high-performance computer simulations. Machine-learning techniques predict the most important reactions with a small number of simulations.