(285d) Gaussian Process-Based Model Order Reduction for the Prediction of Gaseous Storage in Metal-Organic Frameworks.

Metal-organic frameworks (MOFs) are a unique class of crystalline porous materials with potential applications in a multitude of industrial settings including gas storage. The nature of MOFs means that their structures are highly tailorable allowing a large number to be synthesized. This poses the inherent problem of identifying promising MOFs for a given application without the need for complex and time-consuming calculations.

This study aims to further the understanding of MOFs by deriving a novel method that uses synthesised MOFs to develop a prediction technique that results in faster and more efficient material design and discovery for gas storage applications. The ability to accurately predict gas adsorption properties and identifying the most dominant geometrical features of MOFs affecting adsorption phenomena, allows scientists to quickly and efficiently design and discover new MOFs. To do this, we adopt a Gaussian Process (GP) surrogate modelling approach, applied to a large database of MOFs that have been experimentally synthesized. This GP surrogate is then used to perform a Global Sensitivity Analysis of the data, to quantify both the most important geometric characteristics of MOFs and the interaction between them, which is also then used as part of a dimensionality reduction approach to identify the parameter subspace required to accurately predict gas adsorption. A comparison is then made between these subspaces for the different adsorbed gases to assess the ability to predict across different gas species, here oxygen and methane.

Results, shown in Figure 1, demonstrate that the surface area and void fraction dominate the deliverable capacity of both oxygen and methane in MOFs. The dimensionality reduction shows that the pore properties can be confidently reduced to just three primary dimensions while also improving the prediction accuracy when the storage gas is varied. Overall, the use of a GP surrogate model for a Global Sensitivity Analysis and a reduced order model has revealed the accurate determination of important pore properties and has helped pave the way for accurate prediction of MOF performance regardless of the gas. This will aid experimental researchers in designing MOFs more efficiently concentrating on optimising the important pore properties.