CSE Symposium Keynote

Donald Estep, Colorado State University

TITLE: Adjoint-Fueled Advances in Error Estimation for Multiscale, Multiphysics Problems

DATE: Wednesday, April 22, 2009
TIME: 3:00 P.M.
PLACE: 2240 DCL
1304 W. Springfield Ave., Urbana, IL

Abstract

From the spread of disease in a national livestock distribution system to the storage of nuclear waste underground, from building fusion reactors to designing MEMs devices, from climate modeling to the performance of automobile tires, science and engineering are increasingly concerned with understanding and predicting the behavior of multiscale, multiphysics systems. The range of scales and complexity in such systems makes physical experimentation, modeling, and simulation difficult, sometimes impossible, and always expensive. As a consequence, there is an increasing - even strident - demand for quantification of error and uncertainty in model predictions.

Perhaps the most widely used technique for obtaining accurate numerical solutions of multiscale, multiphysics problems is operator decomposition. The general approach is to decompose a given model into components involving simpler physics over a relatively limited range of scales, and then to seek the solution of the entire system by using numerical solutions of the individual components. This approach has many appealing aspects, and in particular, provides a natural way to tackle problems encompassing multiple time and length scales. However, multiscale operator decomposition affects both accuracy and stability in both obvious and subtle ways that are difficult to quantify accurately.

In this talk, I will describe a powerful approach for quantifying sensitivity, error, and uncertainty based on duality and adjoint operators. This approach provides a way to quantify the effects of stability, which is the key to accurate error estimation. I will explain my view of stability, describe the connection to duality and adjoints, and apply these ideas to a variety of multiscale, multiphysics problems.