Explicitly Solvated Free Energy Barriers for OER on IrO2(110) using Machine Learned Interatomic Potentials

We developed a machine learning interatomic potential (MLIP) using the MACE architecture to model the IrO2(110)-water interface for oxygen evolution reaction (OER) studies. The training dataset was carefully constructed to include bulk IrO2, bulk water, the IrO2(110) surface with explicit water, and OER intermediates (*OH, *O, *OOH) with their possible interactions at neighboring Ir coordinatively unsaturated sites (CUS) and O bridge sites. Rigorous validation includes DFT benchmarking of energetics and geometries, alongside physical properties such as radial distribution functions, diffusion coefficients, and system stability over extended dynamics.

The MLIP enables calculation of the potential of mean force (PMF) for *OOH formation, a critical rate-limiting step in OER. Its computational efficiency permits simulations with hundreds of explicit water molecules over timescales prohibitive with DFT, capturing the full complexity of water reorganization and hydrogen bonding dynamics at the interface. This methodology provides a pathway toward accelerated discovery of novel OER catalysts by enabling rapid screening of material compositions and surface structures under realistic aqueous conditions.