(709g) Elucidating the Dependence of Nanoparticle Stability & Reactivity Metrics on the Choice of Exchange Correlation Functionals Using a Data Driven Approach

In silico design of nanostructures involves computing stability metrics like their cohesive and surface energies, and reactivity metrics like the coordination environment of constituting atoms. These metrics can be calculated using Density Functional Theory (DFT) or estimated using surrogate models like the Alloy Stability Model or Generalized Coordination Numbers. While DFT is prone to uncertainties arising from the exchange correlation functional, our understanding of how uncertainties in total energies influence metrics of nanoparticles is emerging, yet incomplete. To address this question, we train the Alloy Stability Model with ensembles generated by the Bayesian Error Estimation Functional (BEEF) and determine standard deviations in surface energies, cohesive energies, and the surface densities of active sites. We observe a standard deviation of ~10-15% in the surface energy ratios obtained from both the ensemble and from other considered functionals. For the cohesive energies, the entire ensemble shows a linear relationship with respect to reciprocal of the nanoparticle diameters, thus following the Gibbs-Thomson correlation. Using the surface energy ratios, we build a library of Wulff nanoparticles having diameters of 2 to 7 nm. We quantitatively represent different morphologies of nanoparticles using the density of active sites per surface atom (e.g., edges and terraces). Densities of active sites vary by ~10-15%, consistent with variations seen in the surface energy ratios. To estimate the errors associated with reactivity metrics, we determine the current densities per mass of Pt for the oxygen reduction reaction. We use correlations in the literature based on generalized coordination numbers of the surface atoms in the ensemble of Wulff particles. Trends in the mass averaged current density with respect to particle size are similar for all functionals in the ensemble. By quantifying errors in the stability and reactivity metrics of nanoparticles, our approach increases the fidelity of nanoparticle catalysts designed from first principles.