Rapid and accurate identification of effective metal organic frameworks for tetrafluoromethane/nitrogen separation by machine learning

Background

Effectively capturing tetrafluoromethane (CF₄), a notorious greenhouse gas having a greenhouse warming potential 6630 times higher than carbon dioxide, is important to mitigate climate change. Metal organic frameworks (MOFs) are promising adsorbents to entrap CF₄ with extreme high selectivity because they contain versatile functionalized ligands and tunable pores. However, the large population makes experimental methods unpractical to perform the large-scale screening and rational selection of efficient MOFs.

Methods

In this work, an intelligent method based on machine learning was developed to identify the important features of MOFs governing their CF₄/N₂ separation performances and establish the relationship between these features and performance metrics, including the CF₄ adsorption capacity, the adsorption selectivity of CF₄ over N₂, and their trade-off.

Significant findings

The random forest (RF) machine learning algorithm was found to exhibit the highest accuracy in performance prediction. The heat of adsorption, the relative molecular mass of MOFs, and the density of MOFs were three critical features that influenced the CF₄/N₂ separation performances. These dominant features indicate that the pore geometry, framework geometry, and the interaction law between CF₄ and MOFs significantly affected their CF₄/N₂ separation efficiency. Machine learning is thus a powerful tool to guide the design of CF₄-selective MOFs and extend the applicability of machine learning among chemical and environmental communities.