We’re developing techniques to quantify numerical error in multiphysics simulations. Understanding approximation error is an important component of a broader UQ strategy, and we’re investigating both adjoint and forward propagation methods. We’re also developing mathematical and statistical techniques to quantify different types of uncertainties (aleatory, epistemic, model form) that are present in multiphysics simulation models. These non-intrusive techniques include those for parameter screening, global sensitivity analysis, response surface analysis, and Bayesian inferences. In addition, we’re investigating hybrid UQ methodologies that enable the blending of the more rigorous and efficient intrusive UQ methods with non-intrusive and semi-intrusive methods at physics module level. Many of these methods have been incorporated into an open source software package called PSUADE. In addition, we’re investigating hybrid UQ methodologies that enable the blending of the more rigorous and efficient intrusive UQ methods with non-intrusive and semi-intrusive methods at physics module level. This flexible methodology facilitates a plug-and-play concept for in-situ UQ and sensitivity analysis that will be useful for high-fidelity stochastic multiphysics simulations. We’re also researching and developing stochastic data assimilation methods to quantify uncertainties associated with high-dimensional stochastic source inversion. These methods are useful in applications such as seismic and power grid analysis. We’re exploring efficient nonlinear and non-Gaussian methods such as kernel principal component analysis and adjoint-based Bayesian inference. View content related to Uncertainty Quantification.
Tammy Dahlgren has worked primarily in software development and research, as well as on efforts ranging from systems and middleware to applications development and software quality assurance. “I like challenges, trying different things, and the opportunity to make a positive impact,” she says.
The flourishing of simulation-based scientific discovery has also resulted in the emergence of the verification and validation (V&V) and uncertainty quantification (UQ) disciplines. The goal of these emerging disciplines is to enable scientists to make precise statements about the degree of confidence they have in their simulation-based predictions. Here we focus on the UQ discipline, which is essential for validating and verifying computer models.
Livermore computer scientists took on the challenge of streamlining uncertainty quantification calculations on Sequoia.
LLNL and IBM researchers have found innovative ways to increase the number of simultaneous jobs the Sequoia supercomputer can run.