Machine learning demonstrates the impact of proton transfer and solvent dynamics on CO capture in liquid ammonia.

Direct air capture of CO using supported amines provides a promising means to achieve the net-zero greenhouse gas emissions goal; however, many mechanistic details regarding the CO adsorption process in condensed phase amines remain poorly understood. This work combines machine learning potentials, enhanced sampling and grand-canonical Monte Carlo simulations to directly compute experimentally relevant quantities to elucidate the mechanism of CO chemisorption in liquid ammonia as a model system. Our simulations suggest that CO capture in the liquid occurs in a sequential fashion, with the formation of a metastable zwitterion intermediate. Furthermore, we identified the importance of solvent-mediated proton transfer and solvent dynamics, not only in the reaction pathway but also in the efficiency of CO chemisorption. Beyond liquid ammonia, the methodology presented here can be readily extended to simulate amines with more complex chemical structures under experimental conditions, paving the way to elucidate the structure-performance of amines for CO capture.