Artificial Intelligence and QM/MM with a Polarizable Reactive Force Field for Next-Generation Electrocatalysts

Abstract

To develop new generations of electrocatalysts, we need the accuracy of full explicit solvent quantum mechanics (QM) for practical-sized nanoparticles and catalysts. To do this, we start with the RexPoN reactive force field that provides higher accuracy than density functional theory (DFT) for water and combine it with QM to accurately include long-range interactions and polarization effects to enable reactive simulations with QM accuracy in the presence of explicit solvent. We apply this RexPoN-embedded QM (ReQM) to reactive simulations of electrocatalysis, demonstrating that ReQM accurately replaces DFT water for computing the Raman frequencies of reaction intermediates during CO₂ reduction to ethylene. Then, we illustrate the power of this approach by combining with machine learning to predict the performance of about 10,000 surface sites and identify the active sites of solvated gold (Au) nanoparticles and dealloyed Au surfaces. This provides an accurate but practical way to design high-performance electrocatalysts.