Berlin 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 28: Networks: From Topology to Dynamics I (joint session SOE/DY)
DY 28.6: Talk
Wednesday, March 20, 2024, 16:15–16:30, TC 006
Automated large-scale chemical breakdown simulations with machine learned reaction rates — •Joe Gilkes1, Mark Storr2, Reinhard J. Maurer1, and Scott Habershon1 — 1University of Warwick, United Kingdom — 2AWE plc, United Kingdom
Degradation of organic materials occurs over many years and involves rare reaction events over expansive networks of elementary processes.
Building such networks and propagating them in time to evaluate the mechanisms by which materials break down requires tackling combinatorially large chemical spaces and accurately calculating the rates at which thousands of reactions proceed. Such a process comes with a high computational cost, and kinetic simulation of such networks is made prohibitively difficult when variable experimental conditions must be considered, as reaction rates must also vary over the duration of a simulation.
We demonstrate a workflow for the automated construction and solution of chemical breakdown networks using machine-learned reaction rates and a discrete kinetic update approximation when propagating networks in time. This allows for rapid iterative exploration of chemical reaction space with arbitrary variable experimental conditions. We demonstrate this by constructing and simulating a detailed reaction network for the pyrolysis of ethane.
Keywords: kinetics; reaction; network; ML; rates