Cof world: Computational discovery of covalent organic frameworks for gas storage and separation applications by integrating molecular simulations and machine learning

Covalent organic frameworks (COFs) have emerged as a versatile class of porous crystalline materials for adsorption- and membrane-based gas separations owing to their high surface areas, structural diversities, and good thermal and chemical stabilities. Despite the rapidly growing number of synthesized COFs, only a limited fraction has been evaluated for industrially relevant separations, and the much larger space of computer-generated, hypothetical COFs (hypoCOFs) remains largely unexplored. This dissertation describes a high-throughput, multi-scale computational framework integrating grand canonical Monte Carlo (GCMC) simulations, molecular dynamics (MD) simulations, and machine learning (ML) to quantify separation performance of the COF spectrum and to establish transferable structure-property relationships for materials discovery. In the first part, we quantified adsorption-based and membrane-based H2/CO2 separation of ~300 synthesized COFs and ~5000 hypothetical COFs using GCMC and MD simulations. The results showed that many COFs surpass conventional adsorbents in adsorption selectivity and working capacity, while offering high regenerabilities. We also discovered that many COF membranes exceed the Robeson's upper bound thanks to intrinsically high H2 permeabilities. In the second part, we modelled a six-component gas mixture to mimic natural gas purification, to quantify adsorption-based separation potentials of ~600 synthesized COFs and ~3000 hypothetical COFs. Results showed that they can outperform several traditional adsorbents, zeolites, activated carbons, and carbon nanotubes. In the third and fourth parts, we introduced ML-integrated computational screening methodologies to screen ~70000 COF and hypoCOF materials for adsorption-based separations of equimolar CH4/H2 and CO2/CH4 gas mixtures under various cyclic adsorption conditions, such as pressure-swing adsorption (PSA), vacuum-swing adsorption (VSA), temperature-swing adsorption (TSA), and pressure-temperature swing adsorption (PTSA) conditions. Several hypoCOFs were discovered to achieve high adsorption selectivities and working capacities, outperforming synthesized COFs and MOFs. In the final two parts, we introduced the COF Space concept, aiming to fully explore the vast COF materials (~70 000 COFs and hypoCOFs). We established ML models that can rapidly and reliably predict the adsorption- and membrane-based separation performances of COFs and hypoCOFs, providing a computationally efficient alternative to conventional simulations. Using this ML-based screening strategy, we systematically mapped the entire COF material space in terms of (i) their adsorption properties for five different gases (CO2, CH4, H2, N2, and O2), (ii) their adsorption-based separation performance for six industrially relevant gas mixtures (CO2/CH4, CO2/N2, CO2/H2, CH4/H2, CH4/N2, and O2/N2), and (iii) their membrane-based separation performance for seven distinct gas pairs (CO2/CH4, CO2/N2, H2/CO2, H2/CH4, H2/N2, O2/N2, N2/CH4). The findings of this dissertation are therefore expected to directly inform rational COF design and the targeted development of new materials for diverse gas separation applications.