Multiscale Simulation Method for Plasma Flows

In high-energy density flows, a number of dynamical and structural data is needed to be collected at the microscale using first principles calculations. Experimental results and their analysis, on the other hand, are determined by measurements at the macro-scale in space over long scales in time. Thus, one major disparity that is currently inhibiting progress in this area is the extrapolation of microscale information into macroscopically relevant scales. Inertial confinement fusion experiments, for example, are fundamentally multi-physics in nature; understanding the connection between experimental observables, and the underlying microphysics, is needed to fully assess capsule performance. In this work, we use uncertainty quantification driven learning algorithms coupled with directed sampling to automate the learning the learning of a multiscale model that is valid for both atomistic processes and continuum phenomena. The multiscale framework couples a molecular dynamics with a Boltzmann kinetic model to describe mesoscale phenomena, and which is tuned by minimizing the model error. We will discuss the code written to build a valid multiscale model relevant to high-energy density experiments, and how that tool may be leveraged to facilitate the learning of plasma models.