(2hn) Narrow the Gap between Simulated Adsorption Properties and Experimental Results in MOFs
AIChE Annual Meeting
Sunday, November 13, 2022 - 1:00pm to 3:00pm
Teaching Interests: Numerical method, Data Analysis, Machine Learning for Chemical Engineering
Title: Narrow the Gap between Simulated Adsorption Properties and Experimental Results in MOFs
Computationally predicting the adsorption properties of Metal-organic Frameworks (MOFs) in a high throughput way has become a useful means of identifying promising materials for chemical separations. In almost all cases, the defect-free and rigid framework assumption for MOF is used in high throughput simulations for computational convenience. However, these simplifications sometime introduce gaps between the simulated results and experimental observations. In my research work, I systematically examined the influence of MOF defects and flexibility and built out-of-box python packages to assist the high throughput screening. The data obtained is further used to train machine learning models, which significantly lower the computational cost.
Structure defects are unavoidable during synthesis processes or might be introduced during adsorption processes by acid gas and water. Thorough defect studies have been conducted for some well-known MOFs, for example, UiO-66 and ZIF-8. These studies illustrated that defect has a strong influence on MOF adsorption properties, physical properties, and mechanical properties, but their conclusions are constrained by the specific MOF. Thus, a systematic defect study based on a larger database screening is needed, while a package that can easily generate defect structure is required. We developed a python package that can generate missing linker defect structures and dangling linker defect structures at user-specified concentrations. Capping agents are added based on the local chemistry environment change. At the DFT level of accuracy, a subset of the CoRE MOF database was fully optimized to identify the maximum defect concentration without triggering unit cell change. Structures with lower defect concentrations were optimized using DFT only for metal clusters and capping agents. At the force field level of accuracy, all the defect structures are optimized using UFF4MOF and based on FF relaxed structure mechanical properties are computed. We systematically studied the adsorption properties of MOFs at different missing linker defect concentrations. The surface area and accessible volume increase as the defect concentration increases. However, this trend ends when the defect concentration reaches a certain value, and the MOF starts to shrink due to the decrease of the bulk modulus. We used ethene/ethane, CO2, and H2S as examples to test the defect influence on adsorption properties. Loadings and selectivity in MOFs with different defect concentrations are computed using GCMC. Most MOFs have lower uptakes with the increase of defect concentrations at 1 bar due to the decrease of Van der Waals force. However, different adsorbates change at different magnitudes, which complex the trend for selectivity change.
Framework flexibility triggered by thermal vibration or external stimuli exists in nearly all MOFs. Thorough flexibility studies have been conducted for some well-known MOFs, for example, MIL-53. These studies illustrated that flexibility has a strong influence on MOF adsorption properties, but their conclusions are constrained by the specific MOF. Thus, a systematic flexibility study based on a larger database screening is needed, while a package that can easily incorporate the flexibility effect is required. Monte Carlo (MC)/Molecular Dynamics (MD) hybrid method is used to predict the flexibility influence on MOF adsorption properties, which has already been successfully implemented in polymer systems. This method consists of loops of MC and MD, where adsorption properties are calculated in MC steps, and framework flexibility is described in MD steps. Based on this method, we developed a package named TAXI, which only takes MOF structure files as input and can conveniently incorporate flexibility into MOF adsorption simulation. UFF4MOF force field is used to describe MOF flexibility in this package, whose reliability is proven by the consistency between UFF4MOF relaxed structure and DFT optimized structure. We validated how TAXI can be used to reliably predict adsorption properties by comparing simulated isotherms with experimental consensus isotherms in a variety of MOFs. The results show that flexible isotherms obtained from TAXI generally are closer to experimental isotherms compared to the isotherms using rigid structures. Then we applied the TAXI package to the CoRE MOF database to predict saturation loadings of multiple adsorbates. Some MOFs have a large saturation uptake difference between rigid results and flexible results. We investigated those MOFs by comparing their physical properties and visualizing their structures. We found that there are mainly two ways that flexibility can have an impact on adsorption loading, including swelling or compression, and changes in pore environment caused by ligands. Besides MOFs themselves, the influence of adsorbates on MOF flexibility is also discussed in this work.
In conclusion, the influence of MOF defect and flexibility are systematically discussed to narrow the gap between simulated adsorption properties and experimental results in MOFs.