(389b) Neural Networks Provide in-Depth Insights into Cation and Coverage Effects at (electro)Catalytic Interfaces

A realistic modeling of (electro)catalytic interfaces poses considerable challenges due to the inherent complexity of the catalyst surface and the dynamic reaction environment under operating conditions. In this talk, I will present our work on using neural networks to tackle these complexities in order to develop realistic models of (electro)catalytic interfaces. In particular, I will discuss two examples involving the use of message-passing neural networks based on the MACE architecture to study cation and adsorbate coverage effects at these interfaces. In the first example, we develop MACE potentials trained on DFT simulations to study the structure of electrified interfaces and the effects of cations during electrochemical CO₂ conversion. In the second example, we use MACE models to predict coverage-dependent adsorption energies and achieve excellent performance on several in- and out-of-domain tasks using examples of CO adsorption on low and high index facets of Cu and co-adsorption of CO and CHOH adsorbates on Rh surfaces. Our findings highlight the promise of developing neural network accelerated simulations towards realistic modeling of (electro)catalytic interfaces.