COLLOQUIUM Department of Computer Science and Engineering University of South Carolina Integrated Methodology for Building Confidence in the Predictive Capability of Computational Models Gabriel Terejanu Institute for Computational Engineering and Sciences The University of Texas at Austin Date: November 4, 2011 Time: 1430-1530 (2:30pm-3:30pm) Place: 300 Main B213 Abstract With the exponential growth of available computing power and the continued development of advanced numerical algorithms, computational science has undergone a revolution in which computer models are used to simulate increasingly complex phenomena. Additionally, such simulations are guiding critical decisions that affect our welfare and security, such as transportation, performance of energy and defense systems and air transport of harmful chemicals. Reliable predictions of such complex physical systems require sophisticated mathematical models of the physical phenomena involved. But also required is a systematic and integrated approach to build confidence in the predictive capability of these models in the presence of numerous errors. The errors inherent in any simulation are the result of many factors, including model structure inadequacies, uncertainties in model parameters, uncertain initial and boundary conditions, experimental uncertainties, as well as errors due to numerical discretization and sampling schemes. The integration of basic processes such as verification, calibration, validation, uncertainty quantification and data assimilation allows for the management of model prediction accuracy in the presence of all these potential errors. In this talk I will describe some of my work on model calibration, validation, uncertainty quantification, and data assimilation, with applications to space capsule reentry in the atmosphere and evolution of toxic clouds. These processes will serve as the building blocks for an integrated methodology that will allow decision makers to trust model predictions for guiding highly consequential actions. Gabriel Terejanu is currently a Postdoctoral Fellow in the Center for Predictive Engineering and Computational Sciences (PECOS) at the Institute for Computational Engineering and Sciences (ICES), The University of Texas at Austin. He received M.S. and Ph.D. degrees in Computer Science and Engineering from University at Buffalo, in 2007 and 2010, respectively. He also received an Engineering Diploma in Automation, specialization robots, from University of Craiova, Romania in 2004. While at University at Buffalo, his research was supported by DTRA, NSF and NGA. At present, his research is supported by DOE/NNSA, under the PSAAP program, and by KAUST. His research interests are in model validation and uncertainty quantification, information fusion, and decision making under uncertainty.