High-Throughput Screening of Metal-Organic Frameworks for the Separation of Hydrogen Sulfide and Carbon Dioxide from Natural Gas

In this work, the adsorption performance of 6013 computation-ready, experimental metal-organic frameworks (CORE-MOFs) for the capture of H2S and CO2 from natural gas mixture (CH4, C2H6, C3H8, H2S and CO2) is calculated by high-throughput screening of grand canonical Monte Carlo (GCMC) simulation in 298 K and 10 bar. For the comprehensive consideration of both adsorption capacities and selectivities of H2S +CO2, first, we compare three different tradeoff methods (a tradeoff method (Tradeoff between SH2S+CO2/C1-C3 and NH2S + CO2, TSN), standard normal method (SNM), beta tradeoff method (Tradeoff between selectivity and capacity, TSC)). The effect of selectivity on the new tradeoff variables are appropriately reduced by these tradeoff methods, because some of selectivities are very high. Thus, the new tradeoff variables can comprehensively evaluate the adsorption performance of CoRE-MOFs. Moreover, the correlation of each MOF descriptor (including the largest cavity diameter (LCD), void fraction (phi), surface area (VSA) and isosteric heat (Q(st)(0))) with three tradeoff variables are analyzed by Pearson correlation coefficient, respectively. The LCDs are calculated by Zeo+ + software, but the phi and VSA are simulated by RASPA using probes of He and N-2, respectively. The Q(st)(0) of each adsorbate gas are calculated at infinite dilution condition using NVT-MC method. All GCMC simulations for the screening are carried out using RASPA software. The results show that TSC has the best correlation with four MOF descriptors and the linear model could sufficiently describe the relationship between TSC and four MOF descriptors. Pearson correlation coefficients of four descriptors were -0.613, -0.717, -0.673 and 0.536 on TSC, respectively. Multiple linear regression is applied to quantitatively determine the influencing degree of four descriptors on performance, respectively. Among the four descriptors, Q(st)(0), phi, and LCD have larger standardized regression coefficients compared with VSA. This indicates that Q(st)(0), phi, and LCD are more useful in describing the performances of the MOFs. Thus, these three descriptors are used in the decision tree modeling to define an effective path for screening high-performance MOFs. It is concluded that a maximum probability (77.6%) of fmding the good MOFs can be obtained from the three descriptors. Finally, the 20 best MOFs stand out from the whole database, and fmd that the alkali or alkaline earth metals in MOFs could effectively enhance the separation perfonnance of H2S and CO2. The microscopic insights and guidelines by this computational study can provide significant theoretical guidance for the development of adsorbent for the purification of natural gas.