Machine learning tools have tremendous potential to accelerate computationally complex physics-based simulations. One example is the need to accelerate materials discovery through first-principles Density Functional Theory (DFT) calculations.

This seminar will encompass how machine learning interatomic potentials accelerate the discovery of crystalline nanoporous solids such as zeolites and metal-organic frameworks (MOFs) that are employed in a wide range of catalytic processes due in part to their tunable micro-environments. Kinetics at intracrystalline sites can be modified by changing pore size, pore architecture, or polarity. These environments impart shape-selectivity that preferentially stabilizes transition states, but the large design space including pore architecture, polarity, and catalytic active site identity preclude comprehensive kinetic studies. DFT describes the electronic states of reactive intermediates and transition states but cannot access the longer length scales necessary to quantify the fluxionality of coadsorbed species or solvents. Classical simulations can accurately simulate these conformational changes but require parameterized values. Machine learning interatomic potentials have emerged as a technique to derive parameterized models from DFT data, and our aim is to adapt these models to predict the entropy and diffusion of reactive intermediates in nanoporous catalysts.