Information-theoretic approaches for multiscale simulations and data assimilation

Tuesday, October 23, 2018 - 11:10am - 12:00pm
Keller 3-180
Francesca Boso (Stanford University)
We employ an information-theoretic approach that allows for seamless integration of multi-resolution data into multi-scale simulations to upscale/downscale hydraulic conductivity of heterogeneous porous formations. Available data (at either the fine- or the coarse-scale) are used to inform models at the opposite scale by setting a probabilistic equivalence between the fine and the coarse scale, with closures (parameters and/or constitutive laws) that are determined via minimization of observables error and mutual information across scales. We investigate how this can guide us to formulate scaling laws and we explore means to accelerate scaling of dynamic processes and to reduce data requirements.