High Performance Computing Projects
Currently, our most visible external project in the area of HPC is the ROCUQ uncertainty Quantification project.
Efficient Quantification of Uncertainties in Complex Computer Code Results
In order to estimate the uncertainties in the System Response Quantities (SRQs) of interest from a large, high-fidelity computer simulation, common brute-force methods such as Monte Carlo sampling of input parameter distributions, coupled with thousands of executions of the high-fidelity model may be employed. For models which run in seconds or minutes, and for applications where massively parallel computers are available and can be used to run multiple sample members simultaneously, this can be a simple, robust, and effective technique. However, where many thousands of high-fidelity model realizations may be needed with simulations whose runtimes are measured in days, weeks, or even months, this is not a feasible method. The innovative ROCUQ methodology uses facets of stratified Monte Carlo, coupled with innovative methods for using reduced order models and “informing” their results with a few runs of the high-fidelity model to produce estimates of the uncertainty in the SRQs from the high-fidelity simulations with as few as five to ten high-fidelity runs. It is the goal of this project to encapsulate the ROCUQ methodology in a computer application designed to facilitate day-to-day use by engineers and scientists. We are leveraging facilities that already exist in the Sandia National Laboratories DAKOTA infrastructure.
