Application of the training of density functional theory potentials within machine learning to adsorptions and reaction paths on Platinum surfaces

This study is conducted by making density functional theory (DFT) calculations within gpaw code and then trained the DFT potentials within a machine learning approaches of amp and genet. Adsorbate molecules are chosen based on occurrences in fuel cells. The outcomes from the machine learned training of the potentials show differences in adsorption energies of only up to the order of 10(-2) eV when trained sufficiently well. However, such small differences in energies do not seem to carry into a differences in electronic structures. It shows that (110) surface is better in facilitating a dissociation process of hydrogen and oxygen molecules into the respective ions with a barrier of dissociation of up to 0.75 eV and 1.91 eV, respectively. The relatively small adsorption energies of hydrogen and oxygen molecules, 0.82 eV and -0.10 eV, respectively, seem to be indicative of a favorable mobility of the respective ions within the electrolyte. Carbon monoxide adsorbs relatively stronger with adsorption energy of up to 1.80 eV. As a result, the formation of carbon dioxide and water molecules as a byproducts is strongly activated, but is exothermic, and seem to preferably take place on (100) and (111) surfaces. A weak adsorption energies of these byproduct molecules could also be indicative of a good mobility of the molecules within fuel cells, once formed. Training of the DFT potentials also appear to show differences in energies of reaction paths only to within the order of 10(-1) eV.