TL;DR
This paper details how the Chombo software framework was extended with new capabilities, specifically higher-order mapped-multiblock discretizations and mixed-dimensional programming, to support the COGENT code for simulating tokamak edge plasmas. These advancements enabled COGENT to tackle complex geometries and anisotropic transport, and the developed Chombo features have since proven valuable for other scientific applications, like space weather modeling.
What Problem Does This Paper Address?
Accurately simulating the edge region of tokamak plasmas is crucial for understanding fusion energy but presents significant numerical challenges. This region is characterized by complex geometry, particularly near the "X-point" where magnetic field lines are open, and highly anisotropic transport (much longer scales along magnetic field lines than perpendicular to them). Standard simulation methods struggle with these conditions, requiring specialized numerical techniques. The COGENT code, designed for this edge plasma modeling, needed advanced capabilities in meshing and discretization to handle these complexities efficiently and accurately.
Methods
The core of this work lies in the enhancements made to the Chombo framework to support COGENT. Key developments include:
- Higher-Order Mapped Multiblock (HOMMB) Discretizations: This approach decomposes complex geometries into simpler, logically rectangular blocks. Each block uses a "mapping function" to transform its local coordinates to the physical domain. High-order finite-volume methods (specifically fourth-order) are employed within each block to maintain accuracy. Crucially, a multiblock interpolation scheme was developed to seamlessly connect these blocks at their boundaries, even when these boundaries are not perfectly aligned. This allows for accurate simulations in challenging geometries like the X-point.
- Mixed-Dimensional Programming: For phase-space simulations (which involve both spatial and velocity dimensions), Chombo was extended to support "mixed-dimensional" algorithms. This allows different parts of the simulation to operate in their natural dimensionalities (e.g., 2D for velocity space, 3D for physical space) without requiring custom implementations for each. This is achieved through C++ namespaces and specialized interface functions that manage operations like slicing and reduction across different dimensional spaces.
Datasets
Not discussed in the paper. The paper focuses on the development of simulation capabilities rather than specific datasets for training or evaluation. The simulations described use model plasma parameters and realistic tokamak geometries.
Key Insights and Ideas
- Framework Extensibility is Key: The success of COGENT highlights how a well-designed, extensible software framework like Chombo can be adapted to meet the highly specific needs of cutting-edge scientific applications.
- Geometry-Aware Meshing is Crucial: For problems with complex geometries and anisotropic physics, meshing that aligns with physical structures (like magnetic field lines) is essential for efficiency and accuracy. The HOMMB approach provides a robust way to achieve this, even in challenging regions like the X-point.
- Mixed-Dimensionality Simplifies Complex Simulations: For simulations involving multiple physical dimensions (like phase-space), the ability to work with mixed-dimensional approaches significantly simplifies code development and improves usability by leveraging existing, dimensionally-specific implementations.
- Synergy Between Application and Framework Development: The close collaboration between the COGENT physics team and the Chombo developers was instrumental in identifying needs and implementing solutions that benefited both.
Key Results
- The developed HOMMB capabilities enabled COGENT to perform accurate and efficient simulations of tokamak edge plasmas in X-point geometries.
- Tests demonstrated fourth-order convergence in block interiors and third-order convergence near block boundaries for HOMMB discretizations.
- The mixed-dimensional programming features allowed COGENT to leverage Chombo's native 2D and 3D implementations for its 4D or 5D phase-space simulations, enhancing usability and correctness.
- These Chombo capabilities have been successfully applied to space weather modeling, enabling significant increases in stable timesteps (from 6 to 600 seconds) due to the use of cubed-sphere mappings and higher-order discretizations.
Limitations and Future Work
- Accuracy at Boundaries: While HOMMB achieves good accuracy, there's a noted loss of an order of accuracy at multiblock boundaries. This is an inherent challenge in such approaches.
- Complex Geometries and AMR: Representing complex geometrical features not aligned with coordinates (like divertor plates) still requires further development, with ongoing work on higher-order embedded-boundary discretizations.
- Curse of Dimensionality: For high-dimensional simulations like those in COGENT, the computational cost remains high. Future work includes exploring sparse grid methods to mitigate this "curse of dimensionality."
- GPU Portability: Efforts are underway to port key computational kernels to GPUs using the Proto DSL to improve performance.
- AMR for HOMMB: Extending adaptive mesh refinement (AMR) to HOMMB meshes is an active area of research to provide localized higher resolution where needed, particularly around the X-point.
Summary updated