Elucidating Mn Promoter Structures and Stability on Co Nanoparticles through a Machine Learning Potential-Powered Genetic Algorithm

Abstract Mn promotion has emerged as a prominent route to deliver the next-generation Co Fischer–Tropsch (FT) catalysts, urgently needed to scale up the production of aviation fuels. Nonetheless, to date, the underlying promotional mechanism has not been fully understood. While theoretical calculations have been previously carried out to address this task, the choice of the models has so far been limited, leaving possibly important factors unexplored. In this work, we have employed a genetic algorithm (GA) powered by machine learning potential (MLP) to optimize Mn structures on realistic models of 6–8 nm fcc and hcp Co nanoparticles (NPs) obtained via our recently published DFT–Monte Carlo (MC) approach. The resulting phase diagrams indicate that Mn structures are hydroxylated during FT, while their stoichiometry after activation is MnO–MnO1.5. Mn was found to strongly bind to the Co NPs, although increasing loading considerably weakened the metal–promoter interactions. The structural characterization of the low-energy candidates produced during the GA runs revealed that Mn forms monolayer-like, mostly amorphous patches on the Co surface, which largely influences the promoter structures. Furthermore, we observed that Mn preferentially binds to the Co low-coordinate sites (LCS), which become almost fully covered already at the lowest investigated promoter loading. Last, we employed a magnetic moment–oxidation state correlation to predict the charge of Mn in our models, which was found to always be roughly 2+ regardless of the NP model, Mn phase, and loading. This was ascribed to a charge compensation mechanism involving the Co surface, which keeps Mn in the particularly favorable 2+ electronic configuration despite the actual promoter phase stoichiometry.