Leveraging molecular simulations and machine learning to assess gas adsorption and separation performances of MOF, COFs, IL/MOF, and IL/COF composites

Metal-organic frameworks (MOFs) and covalent organic frameworks (COFs), known for their high surface areas, high thermal and chemical stabilities and tunable properties, have emerged as promising candidates for adsorption- and membrane-based gas separations. Given the vast number of synthesized MOFs (>128000) and over a million computer-generated, hypothetical MOFs (hMOFs), computational methods are essential for evaluating the gas separation performance of these materials. This dissertation explores the gas adsorption and separation performances of MOFs, COFs, and their composites with polymers and ionic liquids (ILs) for applications, including CO2 capture, natural gas purification, and air separation. A multi-scale computational approach integrating molecular simulations, COSMO-RS calculations, density functional theory (DFT) calculations, and machine learning (ML) is employed to efficiently examine large material databases and uncover structure-property relationships. In the first part, we focused on the CH4/N2 separation performances of a total of 5034 MOFs and COFs, and several IL/MOF, MOF/polymer, and COF/polymer composites by performing grand canonical Monte Carlo (GCMC) and molecular dynamics (MD) simulations. Our results showed that IL incorporation significantly enhances CH4/N2 selectivity, and both MOF and COF membranes outperform conventional polymers. In the second part, we extended this approach to IL/COF composites for CO2/N2 separation, revealing significantly improved selectivities, and CO2 permeabilities surpassing those of polymer and zeolite membranes. In the third part, we developed ML models trained on simulated CH4 and N2 adsorption data of 4612 synthesized MOFs and tested the transferability of these models on 98601 hMOFs. Our results revealed that many hMOFs exhibited high CH4 selectivities and working capacities while several outperforming synthesized MOFs. The fourth part involves an ML-integrated workflow to investigate 1322 different types of IL/ZIF-8 composites, covering the largest variety of ILs studied to date (8 cations and 35 anions) at various IL loadings. We performed GCMC simulations to compute CO2, CH4, and N2 adsorption properties of these composites, and the resulting simulated data were used to train ML models capable of predicting gas uptake in any IL/ZIF-8 composite based on the chemical and structural features of the ILs. Our results showed the high accuracy of the models through the comparison of ML predictions with experimental and simulation data. The last part focused on the development of ML models for CO2, O2, and N2 adsorption data of synthesized MOFs and transferring the information gained from these models to four different hMOF databases. As a result, CO2, O2, N2 adsorption and CO2/N2 and O2/N2 separation performances of ~130000 structures were predicted, which offers valuable insights for materials discovery. The results of this thesis will provide molecular-level understanding of gas adsorption and diffusion in MOFs and enable the rational design of novel MOFs and MOF-based composites for gas separation applications.