(600bl) Bayesian Methods in Multiscale Modeling

Multiscale modeling is characterized by problems that are statistical in nature.  For example: when bridging between scales, how should the error inherent in the approximations made through reduced order models be handled?  If error is present in calculations made at one scale, how do those affect the predictions of the model made at larger scales?  What is the role of experimental data?  This poster will present a Bayesian framework for answering these questions, involving the development of prior probability distributions through quantum chemical calculations, model form discrepancies, calibration with experiment and propagation of uncertainty.  This framework has been developed as a part of the U.S. Department of Energy's Carbon Capture Simulation Initiative (CCSI), the goal of which is to accelerate the adoption of new carbon capture technologies through science-based simulation.  The statistical methodology will be presented in detail, and an application of the methodology will be presented in the context of CCSI: the use of quantum chemical calculations in CO₂ sorbent models and process simulations.