Past Event: Oden Institute Seminar

What Entitles Us to Make Predictions?

Robert Moser, Gabriel Terejanu & Todd Oliver, Center for Predictive Engineering and Computational Sciences, ICES

Abstract

Niels Bohr once said "Prediction is very difficult, especially about the future." His comment is relevant to computational science and the use of computational models in decisions about physical systems. In computational science and engineering, the predictions we need to make are generally for quantities for which there are no data under the conditions of the prediction, since otherwise no predictions would be needed. Fundamentally, then, the predictions we are interested in are extrapolations from available information. At the Center for Predictive Engineering and Computational Sciences (PECOS), we are developing a conceptual context in which to assess such predictions. It involves four distinct processes: uncertainty modeling, model calibration, validation and predictive assessment. As we apply them, these processes rely on several ingredients, including: model inadequacy modeling, Bayesian inference, Bayesian model selection, sensitivity analysis and prior knowledge regarding the reliability and domain of applicability of the physics models involved. This approach will be discussed, and as examples, the ideas will be applied to several physics modeling problems being pursued at PECOS.