Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 8: Solid-Liquid Interfaces 1: Reactions and Electrochemistry
O 8.8: Talk
Monday, September 5, 2022, 12:30–12:45, S054
Neural network surrogates for kinetic Monte Carlo models of electrocatalytic surfaces — •Younes Hassani Abdollahi1,2, Jürgen Fuhrmann3, and Sebastian Matera1,2 — 1Institut f. Mathematik, Freie Universität Berlin, Arnimallee 6, 14195 Berlin, Germany — 2Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 14195 Berlin, Germany — 3Weierstraß-Institut f. Angewandte Analysis u. Stochastik, Mohrenstr. 39 10117 Berlin, Germany
The kinetic Monte Carlo method (kMC) is the physically most sound approach for addressing the kinetic interplay of elementary processes at electrocatalytic surfaces but also comes at high computational costs. Therefore, computationally efficient surrogate models are highly desirable which allow the utilization of kMC simulation results in coarser scale simulations.
Using the oxygen reduction reaction on Pt(111) as a prototypical example, we investigate regression neural networks as surrogates to reproduce the stationary TOF as a function of all reaction conditions, i.e. electrostatic potential, concentrations, and temperature. We found that a relatively shallow perceptron with 2 layers of 32 and 32 neurons, respectively, and SiLU activation functions serve as an appropriate choice. We demonstrate the performance of this model with a varying number of kMC data points. Finally, we discuss how this model can be incorporated into a multiscale modeling approach, which addresses the interaction of transport and kinetics.