(595c) Discovery of High-Performing MOFs Via Machine Learning

We describe a data-driven approach to developing metal-organic frameworks (MOFs) with high hydrogen storage capacities. MOFs are crystalline nanoporous materials containing a metal cluster bonded to organic linkers. High-throughput (HT) grand canonical Monte Carlo simulations and a Voronoi networks-based technique are used to compute the capacities and properties of nearly 500,000 compounds. The large properties-performance database resulting from these calculations presents a unique opportunity for machine learning (ML). Using ML, we predict the necessary structural properties of a MOF needed to achieve specific capacity targets. We refer to this approach as ‘reverse crystal engineering.’ This method illustrates the possibility of developing purpose-built materials with specified functionalities.