Reliable, Accurate, Predictive Simulation of Contemporary Large-Scale Coupled Multiphsics-Multiresolution Problems
Graham F. Carey, CFDLab, ICES, UT Austin
November 18 2009 11:00 am
Johnston Hall Room 338
Accurate, reliable predictive simulation of complex nonlinear, coupled field problems that exhibit interactions at different spatial and temporal scales is a major computational challenge. A further complication of growing importance is the need to certify results and assess risk in predictive simulations. Part of this talk will describe theory and methodology underlying an integrated approach as well as algorithm and software needed to realize reliable predictive simulations with practical error bounds on model, discretization and other sources of error. Related aspects of “verification and validation” will also be discussed. The ideas will be illustrated for parallel mesh adaptive strategies using the open source software library ‘LIBMESH’ developed in our Computational Fluid Dynamics laboratory (CFDLab). Selected results on applications such as cellular patterns in Rayeigh-Benard-Marangoni heat and fluid flow problems, Ginzburg-Landau vortex models, biomedical/ biological models, high speed gas dynamics, and material separation with phase field interfacial models will be presented. Use of the Lonestar and Ranger parallel supercomputer systems at TACC has been made in the large-scale simulations. Work in progress with Libmesh coupled to Sandia software DAKOTA for UQ will be briefly discussed. Finally, I will comment on some aspects of predictive simulation and UQ for the NASA re-entry space vehicle.