Adsorption of benzene on graphene studied by ML-accelerated <i>ab initio</i> molecular dynamics simulations

Abstract

Periodic ab initio molecular dynamics simulations accelerated by machine learning were conducted to investigate the adsorption of benzene on graphene at a coverage of 0.11 ML and a temperature of 150 K. For this purpose, seven density functional approximations (DFAs) were used, differing in the exchange–correlation functional (PBE, SCAN, HSE06, and vdW-optB86b) and the long-range dispersion correction (D2, D3, D4, and MBD). Thermal effects on the structure, enthalpy (ΔadsH), and Gibbs free energy (ΔadsG) of adsorption were analyzed. The relatively fast surface diffusion of the adsorbate was identified as the dominant anharmonic effect causing an increase in ΔadsG of up to 50% compared to static harmonic calculations. In contrast, the anharmonic effect on ΔadsH was shown to be nearly negligible. The computed thermodynamic data were used to predict the desorption kinetics parameters, i.e., activation energy (Ea), Arrhenius pre-factor (ν), and position of desorption maximum (Tmax). Among the DFAs tested, the PBE + D4 method was found to provide the best overall agreement with the experimental data. The SCAN + MBD, PBE + D3, and PBE + D2 methods predict Ea values that agree with the experimentally determined value within the reported uncertainty but significantly underestimate Tmax.