Implicit high-order gas-kinetic schemes for compressible flows on three-dimensional unstructured meshes I: Steady flows

Yaqing Yang, Liang Pan, Kun XuJournal of Computational Physics 2024 · DOI

TL;DR

This paper introduces implicit high-order gas-kinetic schemes (HGKS) for simulating compressible flows on complex 3D geometries. By developing both non-compact and compact implicit HGKS, the authors significantly accelerate convergence for steady-state simulations, achieving robust and accurate results across a range of flow regimes, including supersonic and viscous flows.

What Problem Does This Paper Address?

Simulating compressible flows, especially those with complex geometries, is crucial in many engineering applications. Traditional explicit numerical methods often struggle with slow convergence rates for steady-state problems, requiring an excessive number of time steps. While implicit methods offer faster convergence, they typically involve solving large systems of equations, which can be computationally expensive and memory-intensive, particularly for high-order schemes on unstructured meshes. This paper aims to bridge this gap by developing efficient and accurate implicit high-order gas-kinetic schemes that overcome the limitations of existing methods for steady-state compressible flow simulations on complex 3D geometries.

Methods

The core of this work lies in developing implicit versions of high-order gas-kinetic schemes (HGKS). Two main approaches are presented:

  • Implicit Non-Compact HGKS: This scheme utilizes a third-order weighted essentially non-oscillatory (WENO) reconstruction. The stencils for reconstruction span two levels of neighboring cells. The implicit formulation is coupled with the nonlinear generalized minimal residual (GMRES) method for solving the resulting linear systems.
  • Implicit Compact HGKS: To improve resolution and parallelism, a compact HGKS is developed using Hermite WENO (HWENO) reconstruction. This scheme's stencils are limited to one level of neighboring cells. The cell-averaged conservative variables and their gradients are updated simultaneously, with gradients driven by the time evolution of the gas distribution function.

Both schemes employ the GMRES method for solving the linear systems, with Jacobi iteration used as a preconditioner to accelerate convergence. For computational efficiency, the schemes are implemented using Graphics Processing Units (GPUs) with CUDA.

Datasets

Not discussed in the paper. The paper focuses on numerical method development and validation using established benchmark problems.

Key Insights and Ideas

  • Implicit HGKS for Steady Flows: The development of implicit HGKS is a significant step towards efficiently simulating steady-state compressible flows, overcoming the slow convergence of explicit methods.
  • Compact vs. Non-Compact Trade-offs: The paper explores the trade-offs between non-compact (wider stencils, potentially better resolution) and compact (narrower stencils, better parallelism) HGKS, demonstrating that both can achieve high accuracy and efficiency.
  • GMRES with Preconditioning: The use of GMRES with Jacobi preconditioning proves effective in solving the large, sparse linear systems arising from implicit schemes, balancing accuracy and computational cost.
  • GPU Acceleration: Implementing these implicit HGKS on GPUs significantly boosts computational speed, making complex simulations more feasible.
  • Gradient Update Strategy: A novel strategy for updating cell-averaged gradients using the time evolution of the gas distribution function simplifies the implicit update process.

Key Results

  • Accuracy and Robustness: The developed implicit HGKS (both non-compact and compact) demonstrate good accuracy and robustness across a range of flow regimes, from subsonic to supersonic, and for both inviscid and viscous flows.
  • Faster Convergence: Numerical tests show that the implicit GMRES-based methods converge significantly faster than the traditional LUSGS method for steady-state problems.
  • GPU Speedup: The GPU implementation achieves an impressive 8x speedup compared to the CPU implementation for the implicit HGKS. When considering the number of CPU threads, the effective speedup reaches approximately 130x.
  • Validation: The schemes are validated against benchmark problems like the lid-driven cavity flow and flow past a sphere, showing good agreement with experimental data and other numerical results.

Limitations and Future Work

  • Memory Constraints for Compact Schemes: The paper notes that compact schemes, while offering better parallelism, require storing more local information, which can be a limitation for single GPU memory capacity.
  • Future Work: The authors suggest future work will involve:
  • Upgrading the implicit compact HGKS for unstructured meshes to utilize multiple GPUs with MPI and CUDA for even larger-scale simulations.
  • Investigating more challenging problems, particularly for turbulent flows.
  • Exploring optimizations for the implicit update strategy for unsteady flows, potentially by designing nonlinear weights.
  • Further investigation into stencil selection and the temporal evolution of cell-averaged gradients.

Summary updated

BibTeX
@article{yang2024implicit,
  author = {Yaqing Yang and Liang Pan and Kun Xu},
  title = {{Implicit high-order gas-kinetic schemes for compressible flows on three-dimensional unstructured meshes I: Steady flows}},
  journal = {Journal of Computational Physics},
  year = {2024},
  doi = {10.1016/j.jcp.2024.112902},
}
Figure