MLCS: Detecting Stable Surface Adsorbates with Bayesian Optimization
Detecting Stable Surface Adsorbates with Bayesian Optimization
Doctoral candidate of applied physics, Aalto University
Reliable identification of stable surface adsorbates requires thorough exploration of the potential energy surface (PES) of adsorption. Simulating organic molecules on inorganic surfaces, however, is prohibitively expensive with conventional quantum mechanical methods, such as density-functional theory (DFT). Instead, traditional structure search methods have relied on chemical intuition, focusing only on the likely minimum-energy structures. At the organic-inorganic interface, this intuition is difficult to apply and can lead to biased and incorrect results.
We combine DFT simulations and artificial intelligence (AI) to resolve stable surface adsorbate structures. We apply the Bayesian Optimization Structure Search (BOSS) method  to study the adsorption of camphor (C10H16O) on a Cu(111) surface. BOSS is a new AI tool, which accelerates the structure search via an intelligent and unbiased sampling of the PES. BOSS minimizes the number of required energy computations with DFT and determines the complete PES for a clear identification of the most stable structure.
In this study, we first analyze camphor conformers with a 3-dimensional search of methyl group rotations. We then employ the identified minimum-energy structure of the molecule to investigate its adsorption on a Cu(111) surface as a function of molecular orientation and translations (see Fig. 1a). By combining QM simulations and AI, we identify the stable adsorbate structures of camphor on Cu(111) and the associated energy barriers for molecular rotation and translation.