Understanding CO adsorption in MOFs combining atomic simulations and machine learning

Abstract

Abstract This study introduces a computational method integrating molecular simulations and machine learning (ML) to assess the CO adsorption capacities of synthesized and hypothetical metal–organic frameworks (MOFs) at various pressures. After extracting structural, chemical, and energy-based features of the synthesized and hypothetical MOFs (hMOFs), we conducted molecular simulations to compute CO adsorption in synthesized MOFs and used these simulation results to train ML models for predicting CO adsorption in hMOFs. Results showed that CO uptakes of synthesized MOFs and hMOFs are between 0.02–2.28 mol/kg and 0.45–3.06 mol/kg, respectively, at 1 bar, 298 K. At low pressures (0.1 and 1 bar), Henry’s constant of CO is the most dominant feature, whereas structural properties such as surface area and porosity are more influential for determining the CO uptakes of MOFs at high pressure (10 bar). Structural and chemical analyses revealed that MOFs with narrow pores (4.4–7.3 Å), aromatic ring-containing linkers and carboxylic acid groups, along with metal nodes such as Co, Zn, Ni achieve high CO uptakes at 1 bar. Our approach evaluated the CO uptakes of ~ 100,000 MOFs, the most extensive and diverse set studied for CO capture thus far, as a robust alternative to computationally demanding molecular simulations and iterative experiments.

Keywords

• Metal–organic framework (MOF)
• Carbon monoxide (CO)
• Adsorption
• Molecular simulation
• Machine learning