Microkinetic modeling is a common tool to develop a quantitative mechanistic model of a catalytic reaction. Often, due to the trade-off between computational cost and capturing complexity, these models are developed with several approximations, including the choice of ab initio methods to ascertain kinetic and thermodynamic parameters, the mean-field approximation that treats the catalytic surface as homogeneous, etc. We argue, here, that many of these approximations can be relaxed and the accompanying computational cost can be significantly reduced by using machine learning methods.<br/><br/>In this talk, two such methods will be discussed using illustrative examples. First, we will show how the errors arising from the approximations of common density functional theory (DFT) exchange-correlation (XC) functionals can be corrected by (i) developing multifidelity machine-learned models using a combination of DFT data and higher-level estimates such as the random phase approximation (RPA) or experiments (e.g., calorimetry) and (ii) applying these models to compute the kinetics and thermodynamic parameters for the microkinetic models. Second, we will show how the errors arising from mean-field approximations can be overcome by (i) using lattice Monte Carlo (or diffusion-only kinetic Monte Carlo) to calculate ensemble-averaged rates of elementary steps for specific macroscopic surface coverages, (ii) developing coverage-dependent rate models using such simulation data, and (iii) formulating and solving a modified mean-field model that captures surface heterogeneity in the calculation of reaction rates. An underlying aspect of these two methods is the need to build machine-learned models with sparse, heterogeneous, multifidelity data with their own unique challenges. Finally, extensions of current microkinetic modeling techniques to use non-steady state data and their scale-up to larger reaction networks will be briefly discussed.

Machine learning for improved microkinetic modeling of catalytic reaction systems

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.