Center for High-Fidelity Boundary Plasma Simulations
This project will develop a high-fidelity boundary plasma simulation module for magnetic fusion plasma, integrating the global boundary region from the top of the pedestal inside the last closed flux surface to the material walls, including the magnetic separatrix and the scrape-off layer.
The need for high-fidelity boundary physics simulations run on the largest computers has been well documented in the recent DOE workshop reports. The HBPS center will provide verified and validated high-fidelity kinetic simulation codes running on extreme-scale computers to address the physics challenges arising in the boundary region of magnetically confined plasmas. The project’s exascale-aware codes will attain maximal concurrency on the largest leadership-class DOE computers. This work is essential for understanding existing tokamak experiments and confidently predicting ITER operation and performance. The critical questions to be considered are: will high-confinement-mode operation in ITER, both steady state and during transient edge-localized mode (ELM) events, yield an edge pedestal height consistent with the desired fusion power without damaging the divertor plates? Can ELMs be avoided in ITER?
FASTMath team members are providing finite element discretization methods; solver technologies; and unstructured mesh generation, adaptation and particle tracking methods. A high order finite element Landau collision operator with adaptive mesh refinement is being developed for use in plasma applications. Improved PETSc-based linear solver technologies for use by the XGC will be provided. High order accurate particle-mesh interpolation methods are being added to PETSc. The previously developed field following unstructured mesh generation tool designed to provide meshes needed for effective execution of the XGC PIC code will continue to be advanced and improved to meet new project demands. We are developing a new approach to support PIC calculations on unstructured meshes in which both the particle and mesh systems are distributed. An initial implementation of this new approach is being implemented into a version of XGC at this time.