Efficient parametrization and deployment of constant potential models based on automatic differentiation: application for simulating heterogeneous M–N–C catalytic electrodes

Abstract

In this study, we present a differentiable charge equilibration (Q_eq) approach integrated into the JAX-based Differentiable Molecular Force Field (DMFF) package. This innovation enables highly efficient automatic differentiation of various properties with respect to atomic Q_eq parameters, facilitating the construction of a versatile force field optimization and simulation framework for electrochemical catalytic interfaces. Using M–C–N catalysts as a model system, we investigate CO₂ adsorption in aqueous electrolytes. Our results reveal that the surface charge of M–C–N materials governs the orientation and density distributions of interfacial water molecules, giving rise to distinct hydration layers. Notably, a negatively charged electrode is optimal for CO₂ adsorption at the doping center as it balances electrostatic attraction and competitive adsorption of H₂O. These observations provide critical insights for the rational design of CO₂ reduction catalysts and demonstrate the importance of simulation with explicit solvent for electrochemical interfaces.