Professor Igor Mezic currently leads an ARO MURI initiative to apply the Koopman operator to analysis and prediction of combustion phenomena.
He is joined by research groups from Cal Tech, Princeton, MIT, University of Washington, and University of Michigan.
Dynamic Modes of Ignition Phenomena: Learning Chemistry from Data
There are no model-independent methods for the identification of the causal chemical mechanisms hidden within the emergent dynamics of ignition phenomena. Additionally, Molecular dynamics simulations of atomistic models of hydrogen oxidation do not require prior knowledge of possible mechanisms or intermediates and can be validated against electronic structure calculations; along with the relative simplicity of hydrogen oxidation, this makes these simulations a natural starting point for the development of data-driven methods of analysis for both numerical simulations and experimental measurements. Here, we demonstrate a machine learning methodology for dynamical processes, based on Koopman mode analysis, to extract dynamic modes for the kinetic stability and instability of reacting mixtures out of chemical and thermal equilibrium and apply it to extensive atomistic simulations of hydrogen oxidation. By defining persistent dynamic modes, we have developed an automated means to extract persistent local (in time) effective reactions along with a fuel-agnostic measure of ignition-delay time.
Cory Brown (top) applies Koopman operator methods to develop a definition and predictive model of combustion.
Allan Avila (bottom) analyzes atomic scale interactions by using Electron Structure Theory to inform pairwise interactions in molecular dynamics simulations (https://lammps.sandia.gov/).