Speaker
Description
Hermes-3 is an open-source fluid edge plasma code written using the BOUT++ framework. Building on the features of SD1D, Hermes-2 and STORM, it enables self-consistent simulation of scrape-off layer plasmas in 1D, 2D and 3D for both steady-state and unsteady/turbulent regimes. In this work, we present an update on several activities contributing towards improved neutral model capability in the code.
Hermes-3 is a fully implicit code. In 2D mean-field, it relies on the SNES solver in PETSc with backward Euler time integration and an LU-type preconditioner. The implementation of separate flux limiters for cross-field neutral advection, conduction and viscosity led to severe performance degradation. Upon investigation, the limiter implementation was found to lead to several numerical problems due to instances of non-differentiability and insufficient regularisation. Resolving these has led to a marked speedup and improvement in robustness across a range of levels of limiter saturation.
Following this, implementing a fully differentiable slope limiter has led to reduced solver failures, steadier timesteps and improved performance. A further significant speed enhancement was found by selectively lagging terms in the neutral flux limiter formulation, revealing their large impact on the stiffness of the system.
We also present an update on coupling Hermes-3 to VANTAGE (Versatile, Accurate Neutral Transport and Analysis for GPU Execution), a new performance-portable kinetic neutral model. Mass conservation is demonstrated in a simple slab simulation with an ionization-recombination balance between the fluid plasma and kinetic neutrals.
This work has been funded by the Fusion Futures Programme. As announced by the UK Government in October 2023, Fusion Futures aims to provide holistic support for the development of the fusion sector.
Performed in part under the auspices of the U.S. DOE by LLNL under contract DE-AC52-07NA27344.