|About this Abstract
||2016 TMS Annual Meeting & Exhibition
||Computational Methods for Uncertainty Quantification, Model Validation, and Stochastic Predictions
||Searching Transition States under Model-Form Uncertainty in Density Functional Theory Simulation
||Lijuan He, Yan Wang
|On-Site Speaker (Planned)
Model-form uncertainty and numerical errors are inherent in density functional theory (DFT) calculation of material systems. They need to be considered in DFT-based transition state search in order to improve the robustness of prediction. In this work, a new search algorithm based on a composite Kriging mechanism is developed to search local minima and saddle points on a potential energy surface (PES) with the consideration of uncertainty. Different from existing searching methods, the algorithm keeps a memory of searching history by constructing a surrogate model. The search result on the surrogate model provides the guidance for new search direction on the PES. This approach significantly improves computational efficiency. The model-form uncertainty in DFT is quantified by the statistical variance associated with model inputs and embodied in the construction of surrogate models. Confidence intervals of the locations of saddle points and activation energies are provided to assess the uncertainty effect.
||Planned: A print-only volume