Machine learning accelerated interfacial fluxionality in Ni-supported metal nitride ammonia synthesis catalysts

The development of energy-efficient catalysts for ammonia synthesis under mild conditions is crucial for reducing the energy demands and carbon footprint of the industrial Haber-Bosch process. In this study, we investigated ammonia synthesis via the associative Mars-van Krevelen (MvK) mechanism using B1-structured metal nitrides, focusing on manganese nitride (MnN) due to its low vacancy formation energy and potential as a metal-support interface. Density functional theory (DFT) calculations identified the MnN (100) facet as the most stable, with a nickel (Ni) nanowire implemented on the surface to facilitate H₂ dissociation while surface nitrogen vacancies activate N₂. A free energy diagram for Ni-MnN (100) at 400 °C was and a dual-site microkinetic model was developed to determine reaction orders, apparent activation energies and the rate limiting step (RDS). To capture temperature-induced catalyst restructuring, ab initio molecular dynamics (AIMD) simulations and machine learning interatomic potentials (MLPs) were employed to improve the sampling of interfacial active sites over longer timescales. We found significant active site fluxionality leading to active site structural rearrangements that reduced vacancy formation energies. A hydrogen coverage analysis at reaction temperatures revealed coverage-dependent dynamic restructuring of Ni active sites, with lowered free energy change of the RDS that correlates with Ni p-Band center. MLPs were observed to predict coverage-dependent fluxionality with training data exclusive to high coverage regimes. By integrating DFT, AIMD, and MLP-based molecular dynamics, we established a computational framework for understanding dynamic metal-support interactions in transition metal nitride catalysts, demonstrating its applicability not only to ammonia synthesis under mild conditions but also to broader classes of supported catalysts and reactions.