Lei Huan, Pacific Northwest National Laboratory
520 Pao Yue-Kong Library
Computational modeling of multiscale multiphysics systems essentially involves quantifying the uncertainty of quasi-equilibrium properties around individual metastable states as well as prediction of non-equilibrium dynamics over the entire phase space, which is centered around modeling of the nonlocal spatio/temporal correlations and scale-dependent fluctuations in the target system. Traditional computational models based on those canonical governing principles (e.g., Fick’s, Darcy’s law) often show limitation. We propose a data-driven approach to model such system based on efficient parameterization of the generalized Langevin Equation (GLE), where the effects of the smaller scale interactions on the scale of interest (i.e., the scale of the field variable) are properly accounted as the memory kernel of GLE. The approximated kernel formulation satisfies the second fluctuation-dissipation conditions with consistent invariant measure. The proposed method enables us to accurately characterize the challenging non-equilibrium properties such as transition rate where traditional hypothesis-driven modeling equations show limitation.