(376e) A Combined High-Throughput Computing and Machine Learning Study Reveals Hydrogen Storage Performance Ceilings of Metal-Organic Frameworks

Ahmed, A., University of Michigan
Siegel, D. J., University of Michigan
MOFs with high hydrogen storage capacities are revealed via high-throughput screening and machine learning on nearly half a million compounds. Screening based on grand canonical Monte Carlo and other techniques identified over 2,000 MOFs that exceed the hydrogen capacity of the benchmark compound MOF-5 by more than 15%. Our data reveals the existence of a celling in usable volumetric capacity near 40 g H2/L, indicating that new design directions for improved volumetric performance are needed. Principal component analysis (PCA) and decision tree (DT) classification were performed on our database to inform design rules. PCA reveals the geometrical properties of MOFs (e.g., surface area, pore volume, pore diameter, density, and void fraction) that correlate with high hydrogen storage capacities. Furthermore, decision tree classification suggested design directions by delineating geometrical properties with acceptable ranges. These data provide guidelines for the design of MOFs with much improved volumetric performance.