Connect via Zoom here!
Predictive Kinetic Modeling Combining Machine Learning, Automated Quantum Chemistry, and Automatic Mechanism Generation
Presented by Dr. Matt Johnson
Abstract:
Many important processes in the world are controlled by chemical kinetics: combustion of fuels in engines, catalytic syntheses of important feedstocks, polymerization, electrochemistry in batteries, and biological processes. However, most overall chemical processes do not occur in a single step from reactants to products and can involve hundreds of different elementary reactions and intermediates. How well we can resolve and parametrize these elementary reactions and intermediates controls our ability to predict the behavior of the associated process. These systems of species, reactions, and their associated parameters are usually referred to as detailed kinetic mechanisms. Efficient generation of predictive detailed mechanisms enables us to better understand the fundamental chemistry, optimize existing systems, and screen possible systems. Creating detailed kinetic mechanisms, however, requires us to determine both what reactions can happen in a given system and calculate or estimate how fast they occur. This can be incredibly tedious and challenging to do manually. I will present an efficient workflow utilizing automatic mechanism generation, automated quantum chemistry, and machine learning to generate and refine detailed kinetic mechanisms. First, I will discuss closing the loop on this workflow in gas-phase, and then developments enabling application to battery electrochemistry and heterogeneous catalysis.
