(662e) Predicting the Adsorption Energies of Cyclic Hydrocarbons Adsorbed on Bimetallic Nanoclusters Using Gaussian Process Regression

Existing catalyst design approaches predict adsorption energies of C_1-3 adsorbates on bimetallic catalysts with accuracies of 0.2 eV. However, similarly robust methods for larger cyclic hydrocarbons are lacking. Cyclic hydrocarbons are key intermediates during the dehydrogenation of liquid organic hydrogen carriers like methyl cyclohexane [1]. When adsorbed on bimetallic nanoparticles, cyclic hydrocarbons present a unique combinatorial challenge because of their diverse configurations across a wide-range of active site compositions. We present a Gaussian Process Regression (GPR) model that predicts the adsorption energy of toluene on bimetallic Pt cuboctahedra nanoclusters containing 55 atoms. This probe system was selected because the adsorption energy of toluene serves as a descriptor for the dehydrogenation of methyl cyclohexane. The composition space included random alloys of Pt with d-block elements. Using density functional theory (DFT), we calculated the adsorption energies of toluene on 125 active sites spanning a range of coordination numbers and compositions. GPR was selected due to its ability to develop robust models with small datasets and its inherent uncertainty quantification capabilities. Our feature set included the composition, electronegativity, and cohesive energy of metals atoms in the active site and the first-nearest neighbors. An 80-20 train-test split was employed. Figure (a) indicates that GPR predicts the adsorption energy of toluene across various compositions of nanoalloys with errors of 0.12 eV (test set). Figure (b) shows that trends in adsorption energies with increasing hetero-metal composition are consistent with DFT calculations. By including features related to the adsorbate structure, our model is extended to other intermediates within the methyl cyclohexane dehydrogenation network. This study highlights the power of GPR in predicting adsorption energies of cyclic hydrocarbons on bimetallic nanoclusters, thereby opening avenues for designing tailored nanoparticle catalysts.

[1] Okada Y., Extended abstracts of the 9th Tokyo Conference on Advanced Catalytic Science and Technology, Fukuoka, KL14, (2022).