Monte Carlo multiphysics simulation on adaptive unstructured mesh geometry

Abstract

Monte Carlo simulation based on Constructive Solid Geometry (CSG) brings unique challenges for multiphysics simulation, including establishing field transfers with mesh-based physics codes, the combination of stochastic and deterministic solvers, and high computational expense. In this work, an adaptive, on-the-fly mesh-based Monte Carlo geometry algorithm is implemented in Cardinal to reduce the barrier-to-entry for high-fidelity multiphysics by (i) eliminating ambiguity in defining CSG cells for temperature and density feedback, (ii) enabling simple mesh convergence studies, and (iii) more closely integrating Computer Aided Design (CAD) workflows with Monte Carlo methods. During Picard iterations, an OpenMC mesh geometry is adaptively refined or coarsened by contouring temperature and/or density fields from a thermal-fluid solver. This algorithm is applied to a full-core Molten Salt Fast Reactor (MSFR) geometry with NekRS Large Eddy Simulation (LES) coupled to OpenMC neutron transport. A performance study indicates a net speedup of 2.3$\times$ in the OpenMC solver when using an adaptive geometry for cell sizes chosen intermediate to the as-built CAD geometry versus 1:1 element tracking, which points to future algorithmic research in accelerated Monte Carlo mesh tracking.