TitleOral: Accelerating high-throughput screening of high thermal conductivity metal-organic frameworks with active learning

Metal-organic frameworks (MOFs) have garnered interest for their potential in gas storage, separation, and catalysis applications. Nonetheless, effectively managing the heat generated during exothermic adsorption processes in MOFs is an important challenge that constrains their practical utility.

Previous experimental and modeling research on thermal transport in MOFs has largely identified structures with low thermal conductivity, typically less than 2 W/m-K. This is because the high porosity and low density of MOFs impede the efficient transport of phonons, which are responsible for thermal transport. Moreover, the wide range of chemical compositions in MOFs can lead to differences in atomic mass and bond stiffness within their structures, further contributing to increased phonon scattering. Here, we elucidate the intricate connections between MOF structure and thermal conductivity, to enable regulation and customization of their thermal transport characteristics.

To achieve this objective, we performed a high-throughput screening using a combination of classical molecular dynamics (MD) simulations with Green-Kubo calculations and surrogate models based on geometric message-passing graph neural networks. This screening encompasses over 80,000 MOFs generated through the ToBaCCo-3.0 code, considering a range of structural and compositional features such as pore size, density, node-linker combinations, topology, and metal-node connectivity. To efficiently navigate this search space and identify prospective candidates for guiding expensive all-atom MD computations, we employed an active learning approach driven by uncertainty analysis.