Title: Informatics for Higher Order Finite Element Simulations Speaker: Philippe P. Pebay, Sandia National Laboratories Date/Time: Monday, August 8, 2005 Brief Abstract: Advanced finite element applications (FEAs) are required by Sandia in order to accomplish its role as stockpile steward without weapons testing. New FEAs use a combination of hierarchical (h) and polynomial (p) adaptivity to simulate a test with fewer compute resources than traditional simulations. However, there is currently a paucity of tools for dealing with the many additional complexities these codes present. In particular, choosing whether a region of the simulation domain should be h-, p-, or hp-adapted is currently accomplished with heuristics. When these heuristics fail, diagnosing their cause is difficult. Also, the way mesh adaptations are currently performed consumes a great deal of time, as estimates of the error can only be made after an adaptation has been performed. In practice, this results in an iterative process where multiple mesh adaptations must be made before the desired error bounds are reached. Both these problems reduce the gains that higher order FEAs would otherwise have over traditional FEAs. In this talk, we will discuss how Bayesian techniques can help analysts and developers devise an "a priori" error estimator based on both data obtained from previous runs, and expert knowledge. In particular, we will examine why optimal quadrature is critical for such approaches. CSRI POC: Richard Lehoucq, (505) 845-8929 |