Online Tools for Uncertainty Quantification in nanoHUB
S Desai and M Hunt and A Strachan, JOM, 71, 2635-2645 (2019).
nanoHUB is a cyberinfrastructure for cloud computing, where simulation tool developers can make their products available to their community and users can run fully interactive simulations from a standard web-browser. We report recent cyberinfrastructure developments and workflows that enable calibration and automatic uncertainty propagation for the 400+ nanoHUB tools that use the rapid application infrastructure (Rappture). A Jupyter notebook enables users to upload training datasets, connect these data with the desired tool outputs and specify the calibration variables. The tool uses the Dakota software package that enables various calibration methods. We demonstrate its use via Bayesian calibration of an interatomic potential for molecular dynamics. We also extended Rappture to include automatic uncertainty propagation in deterministic tools. Real-valued tool inputs can be specified as a distribution, and nanoHUB uses collocation to perform multiple simulations spanning the input ranges. Simulation results are post- processed to create surrogate models, perform sensitivity analysis and uncertainty propagation.
Return to Publications page