Computational screening of metal–organic frameworks for separation of CO2 and N2 from wet flue gas

Abstract

In response to the challenging task of effecting CO₂ and N₂ adsorption separation under humid gas conditions, this study employs a methodology that integrates molecular simulation with high-throughput screening. The focus is on investigating the adsorption separation of a ternary mixture (CO₂/N₂/H₂O) utilizing the most recent experimental synthesis of metal–organic frameworks (MOFs) from the CoRE-MOF-2019 database. To circumvent the competitive adsorption of water vapor, materials with excessive hydrophilicity are systematically excluded. Subsequently, a univariate analysis is conducted on the remaining 1343 MOFs, exploring the intricate relationships between key structural parameters such as pore limiting diameter, maximum pore cavity diameter of free channels (LCD), pore volume (V_pore), volume surface area (VSA), weight surface area, density (ρ), porosity (φ), Henry coefficient (K), adsorption heat (\(Q_{{{\text{st}}}}^{0}\)), and metal types. The investigation reveals positive correlations between ρ, K, and \(Q_{{{\text{st}}}}^{0}\) with selectivity, while other descriptors exhibit negative correlations. Notably, MOFs enriched with Cd and Cu demonstrate superior performance. Subsequent analysis employs Pearson coefficients and a decision tree model to rank descriptors and identify the top three descriptors (LCD, VSA, and \(Q_{{{\text{st}}}}^{0}\)) influencing performance. Utilizing these descriptors, the decision tree model delineates optimal design criteria: \(Q_{{{\text{st}}}}^{0}\) > 28.296 kJ mol⁻¹, LCD < 5.893 Å, and VSA > 727.596 m² cm⁻³. To predict the performance of MOFs that have not yet been synthesized or experimentally validated, we employed the nine descriptors for model training and out-of-sample validation. The decision tree classifier exhibits high prediction accuracy and shows excellent transferability, effectively delineating the boundaries between different performance classes by capturing structural–selectivity correlations. This process culminates in the screening of 15 optimal MOFs, offering theoretical insights for the adsorption separation of CO₂ in humid flue gas.