(216f) Efficiently Determining Solvent Effects at Metal-Water Interfaces

Accurately modeling interfaces between liquid and solid phase is one important frontier in computational catalysis. Explicitly modeling these interfaces can be challenging due to the large number of molecules needed, as well as the large number of possible configurations. Implicit or continuum solvent models can be much quicker, but have only recently been included in plane wave density functional theory (DFT) codes. In contrast, implicit solvation methods are much more mature in molecular DFT codes. How implicit solvation methods perform in modeling periodic catalytic systems is still unclear. In this work we systematically evaluated how interfaces between water can metals can be treated using implicit solvation methods.

In total we considered 41 different adsorbates and used three different solvation methods implemented in the NWChem, VASP, and JDFTx codes. Agreement between the three programs was strong. We classified the adsorbates into five different groups, and found that trends within each group were consistent. For instance, adsorbed atomic or weakly polar species showed almost negligible solvation effects, while adsorbed aromatic compounds were strongly stabilized by the presence of water solvent (up to 0.44 eV). We also modeled four prototypical reactions in vacuum and water: oxygen reduction reaction, formic acid oxidation, C-C cleavage, and water-gas shift. Reaction energies changed up to 0.23 eV when solvent was present. We analyzed the contributing energy terms to the solvation energy of adsorbed species, and found that relevant desciptors (e.g. Bader charge of adsorbates, dipole moments) are correlated to these solvation energy terms. We furthermore created several artificial neural networks, where we evaluated potential descriptors for predicting solvation energies of adsorbates. Our best neural network indicated that Bader charge of adsorbed species, solvation energies of gas species, gas dipole moments, and molecular surface area of adsorbates are all simple descriptors that can be used to predict solvation energies of adsorbates. Our work provides guidelines on when solvation effects may be important for surface chemistry, and also provides valuable insights on modeling such effects in an efficient manner using implicit solvation.