Multi-Scale Computational Design of Metal–Organic Frameworks for Carbon Capture Using Machine Learning and Multi-Objective Optimization

In this article, we computationally design a series of metal–organic frameworks (MOFs) optimized for postcombustion carbon capture. Our workflow includes assembling building blocks and topologies into an initial set of hypothetical MOFs, using genetic algorithms to optimize this initial set for high CO₂/N₂ selectivity, and further evaluating the top materials through process-level modeling of their performance in a modified Skarstrom cycle. We identify two groups of MOFs that exhibit excellent process performance: one with relatively small pores in the range of 3–5 Å and another with larger pores of 6–30 Å. The performance of the first group is driven effectively by the exclusion of N₂ from adsorption, with binding sites able to accommodate only CO₂ molecules. The second group, with larger pores, features binding sites where CO₂ molecules form multiple interactions with oxygen and functional groups of several building blocks, leading to a high CO₂/N₂ selectivity. Within the employed process model and its assumptions, the materials generated in this study substantially outperform 13X reference zeolites, in silico optimized ion-exchanged LTA zeolites, and CALF-20. While this study does not address the synthesizability, stability, or water interactions of the proposed materials, it marks a significant step forward in developing practical MOFs for carbon capture in three key areas. First, it introduces a generative workflow based on the process-level performance of new materials. Second, it identifies structural features of optimal MOFs for carbon capture, which can serve as design guidelines for future development. Finally, the potential existence of numerous promising materials offers hope that some may progress to laboratory testing and eventual scale-up.