Monday, November 9, 2015: 4:35 PM
355B (Salt Palace Convention Center)
This work aims to augment the field of computational catalysis with uncertainty quantification (UQ). An efficient tool to describe the energetics and structure of atomistic systems is density functional theory (DFT). DFT proves useful for understanding how catalysts work if DFT results are combined with mean-field microkinetic modeling or kinetic Monte Carlo simulations. However, DFT is inexact in nature due to approximations necessary for computational tractability. These approximations in DFT cause uncertainty in microkinetic modeling results used for comparison with experiments. Therefore, reliable model results obtained from DFT should include a quantification of uncertainty. This work focuses on a forward problem in which uncertainties in DFT are propagated forward to model results such as turnover frequency (TOF), apparent activation barrier, and reaction orders. UQ is performed on a case study using DFT calculations of the water-gas shift reaction (WGS: CO + H2O -> CO2 + H2). The case study models the catalyst as a Pt(111) slab and a platinum cluster supported on titanium oxide. Uncertainty in model results is accounted and reduced as much as possible by introducing correlations in DFT energies using four separate functionals and a factor analysis to generate a covariance matrix with the converged Perdew-Burke-Ernzerhof result as the expected value. The additional functionals are the revised Perdew-Burke-Ernzerhof (RPBE), Heyd-Scuseria-Ernzerhof (HSE), and M06L. In this way, results from three different classes of functionals, GGA (generalized gradient approximation), meta-GGA, and hybrid functionals, are including in making predictions. Additionally, DFT gas molecule energies are constrained to experimental thermodynamics in the UQ simulation. Although uncertainty in model results spans orders of magnitude, the dominant catalytic cycle is identified which suggests DFT is accurate for producing relative results. Finally, we perform a comparative study to determine if a Pt(111) catalyst model or a Pt cluster on rutile titania can better explain experimental observations from Kalamaras et al. (J. Catal. 2009, 264, 117−129). Kullback-Leibler (KL) divergence calculations suggest that the Pt cluster model on rutile titania can explain experimental observations significantly better.