Building and Exploring Databases of Porous Materials for Adsorption Applications
In recent years, the field of porous materials has witnessed increasing interest towards customisable structures. Among them are metal-organic frameworks (MOFs), covalent organic frameworks (COFs), metal-organic cages (MOCs) and organic cages (OCs). MOFs are coordination networks assembled from organic ligands and metal clusters, while COFs are their fully organic equivalents. MOCs are crystalline structures obtained from the packing of discrete cage-like organometallic molecules; OCs being in turn their organic equivalents. While all these materials are attracting increasing attention, MOFs remain the star. The building block approach to their relatively straightforward synthesis, combined with an easy crystallographic characterisation, has enabled scientists to synthesise an increasing number of structures and deposit their data into the Cambridge Structural Database (CSD). It is from the same database that I previously derived, in 2017, the world’s first automatically updated MOF subset (the so-called CSD MOF subset). Three years after its creation, the number of structures increased from 70,000 to almost 100,000. Building on my previous work, I developed a set of tools for experimental and computational scientists alike to explore the CSD MOF subset. I devised a series of methods for the targeted classification of MOFs into different groups: MOFs from different chosen families, with specific surface functionalisation, chirality, as well as channel and framework dimensionalities. The obtained information, along with their geometric characterisation was made accessible via an online interactive data visualisation platform, thereby mapping out the properties landscape of the CSD MOF subset. I then carried out a high-throughput screening (HTS) of a selected number of MOFs for their hydrogen storage performance at 298 K and 200, 500 and 900 bar. In addition to confirming one of the top-performers for this task, I uncovered interesting structure-property relationships by matching the adsorption data to the structural information obtained from this classification. I found that the best performing structures tend to be part of one of these families: CPO-27, Cu-Cu paddlewheel, IRMOFs or zirconium-oxide MOFs. Structures with three-dimensional porous channels and/or with halogen surface functionalisation also seemed to have greater performance. I extended the developed methods to the identification of COFs in the CSD. However, the experimental difficulty in obtaining their crystallographic data means only a few structures were deposited in the CSD, therefore only a small portion of COFs reported in the literature were found. Due to their inherent structural difference, MOCs and OCs could not be identified using the methods used for MOFs and COFs. I, therefore, designed a separate approach for their identification, using a combination of topological data analysis, supervised and unsupervised classification. After successfully obtaining two datasets – the largest OC dataset to the best of my knowledge and the only existing MOC dataset, I carried out a HTS on them for their separation performance on a 20/80 mixture of xenon/krypton at 298 K and 10 bar. I identified the top MOC and OC for this application, and confirmed the high performance of the CC3 family. The mapping of the classification of these two datasets to the adsorption data unveiled interesting insights into the range of performance from these CC3-type structures. In summary, in addition to the previously built MOF subset, I successfully developed subsets of COFs, MOCs and OCs. I also built programmable and customisable tools for the exploration and visualisation of these subsets. The obtained structural landscapes proved useful in uncovering insightful structure-property relationships when mapped to adsorption data.