A molecular-continuum multiscale solver for liquid-vapor flow : modeling and numerical simulation

Abstract

In this dissertation we focus on compressible liquid-vapor flow with sharply resolved phase boundaries. The flow dynamics are determined by the microscale behavior at the phase boundary. In order to describe liquid-vapor flow accurately on the continuum scale, without imposing ad-hoc closure relations, we propose a (universal) multiscale model. It combines continuum-scale flow models with molecular-scale particle simulations that define the interface dynamics. The complete multiscale model is comprised of several parts. In addition to continuum-scale hyperbolic conservation law models, we review particle models such as molecular dynamics simulations. The particle simulations are used to build microscale Riemann solvers and define the flow at the interface. The discretization of the continuum-scale sharp-interface flow is performed by an interface-preserving moving mesh finite volume scheme. In order to keep the multiscale model computationally feasible, while conserving physical key quantities (e.g. mass), surrogate solvers based on constraint-aware neural networks are applied. Finally, combinations of micro- and macroscale models with increasing complexity are explored and simulation results are presented.