Monday, November 18 2019
12:30pm - 1:30pm
Computational Mathematics Colloquium
Computational Mathematics Colloquium: Moving Beyond Forward Simulation to Enable Data-informed Physics-based Predictions

As a computational science community, we need to move beyond forward simulation and seek to enable modeling and simulation to guide optimal experimental design, to use experimental data to develop better computational models, and to combine both models and experiments to make credible predictions. These objectives are hindered by the fact that many problems in science and engineering are best described by multiphysics models that interact on a wide range of length and time scales. Developing fully resolved numerical models that incorporate all of the relevant scales will require more than just increases in computing power. We need to develop the mathematical and numerical tools to incorporate information from all relevant spatial and temporal scales in a simulation, to integrate data in a consistent manner, and to make credible predictions for quantities of interest with quantified error and uncertainty. All of these are required if we intend to use modeling and simulation to inform high-consequence decision-making.

In this presentation, we will discuss our recent progress towards the goal of making data-informed physics-based predictions. We will present the development of a modeling and simulation framework designed to enable the propagation of subgrid scale information into a “coarse scale” numerical discretization that leverages large-scale heterogeneous computational architectures. We also discuss the recent development of framework for solving stochastic inverse problems based on a combination of measure theory and Bayes’ rule. After presenting the main ideas, we will discuss a few extensions of this approach to incorporate the utilization of surrogate/approximate models, multi-fidelity Monte Carlo techniques and scalable approximations based on deterministic optimization. We conclude this presentation with a discussion of a recent formulation for optimal experimental design for prediction based on the stochastic inversion framework.

Tim Wildey is a Principal Member of Technical Staff in the Optimization and Uncertainty Quantification Department within the Center for Computing Research at Sandia National Laboratories. He earned his PhD in Mathematics from Colorado State University in 2007 working on adjoint-based a posteriori error analysis for coupled multiphysics problems. From 2007 to 2010 he was an ICES postdoctoral fellow at the University of Texas at Austin working in the Center for Subsurface Modeling on multiscale methods, scalable physics-based preconditioners, and advanced uncertainty quantification algorithms for flow and mechanics in porous media. He is a recent recipient of a DOE Office of Science Early Career award and is currently working on multiscale and multiresolution numerical methods, verification, uncertainty quantification, stochastic inverse problems, scientific machine learning and high performance computing for multiphysics and multiscale applications.
Speaker:Tim Wildey
Affiliation:Sandia National Laboratories
Location:SCB 4017

Download as iCalendar