TL;DR
This paper introduces a high-order gas-kinetic scheme (HGKS) for simulating turbulent flows, particularly at supersonic speeds, and demonstrates its parallel implementation for large-scale direct numerical simulations (DNS). The HGKS shows comparable accuracy to existing high-order methods while offering improved robustness for supersonic regimes and outperforming other mesoscopic methods like LBM and DUGKS.
What Problem Does This Paper Address?
Direct numerical simulation (DNS) is a powerful tool for understanding turbulent flows, but accurately capturing all scales of turbulence, especially in compressible and supersonic regimes, remains a significant challenge. Traditional methods like spectral and pseudo-spectral methods struggle with discontinuities like shocks, while finite volume schemes often suffer from excessive numerical dissipation. High-order finite difference schemes have shown promise, but simulating flows with strong shocks can still lead to instability. The authors aim to develop and validate a high-order gas-kinetic scheme (HGKS) that can accurately and robustly simulate turbulent flows across a wide range of Mach numbers, from nearly incompressible to supersonic, and to enable large-scale simulations through parallelization.
Methods
The core of the paper is the development and application of a high-order gas-kinetic scheme (HGKS). This scheme is based on the Bhatnagar-Gross-Krook (BGK) equation, which models gas behavior from a kinetic to a hydrodynamic scale. Key methodological aspects include:
- Gas Distribution Function: The scheme recovers inviscid and viscous fluxes from a time-dependent gas distribution function at cell interfaces.
- High-Order Spatial Reconstruction: For spatial accuracy, the paper employs a fifth-order WENO (Weighted Essentially Non-Oscillatory) method, with dimension-by-dimension reconstruction for 3D cases. For smoother flows, a simplified smooth second-order GKS flux and WENO with linear weights are used to minimize numerical dissipation.
- Two-Stage Fourth-Order Time Discretization: A temporal discretization scheme is used to achieve fourth-order accuracy in time.
- Parallelization: A popular parallel strategy using two-dimensional domain decomposition and Message Passing Interface (MPI) is implemented to enable large-scale computations. The domain is divided into sub-domains, and data communication occurs between neighboring sub-domains.
Datasets
The paper does not introduce new datasets but rather uses established benchmark problems for validation:
- Taylor-Green Vortex: A classic problem for studying vortex dynamics, turbulent transition, and decay, used to validate the scheme for nearly incompressible turbulent flows.
- Turbulent Channel Flow: A standard test case for understanding wall-bounded turbulent flows, used to validate the scheme for low Reynolds number turbulent flows.
- Isotropic Compressible Turbulence: Used to test the scheme's performance in simulating turbulence across different Mach number regimes, including supersonic.
Key Insights and Ideas
- HGKS as a Unified Tool: The HGKS demonstrates its capability as a single framework to handle turbulent flows across a wide range of Mach numbers, from subsonic to supersonic, without requiring special treatments for shock capturing.
- Robustness in Supersonic Regimes: The kinetic formulation of HGKS provides inherent advantages in accuracy and robustness for supersonic turbulent flows, where traditional schemes might struggle with shock instability.
- Mesoscopic Advantage: As a mesoscopic method, HGKS outperforms other mesoscopic methods like Lattice Boltzmann Method (LBM) and Discrete Unified Gas-Kinetic Scheme (DUGKS) due to its higher-order accuracy.
- Quantitative Dissipation Analysis: The paper provides a quantitative evaluation of numerical dissipation in turbulence simulations, which is less commonly reported.
- Scalability of Parallel Implementation: The MPI-based parallel implementation shows good scalability, indicating that the computational cost of data communication across nodes is manageable compared to the flow field computation.
Key Results
- Taylor-Green Vortex: The HGKS achieved good agreement with reference solutions, especially with finer grids. The results were comparable to a high-order finite difference method (BB13 DRP scheme).
- Turbulent Channel Flow: The HGKS results showed reasonable agreement with spectral, LBM, and DUGKS results, particularly in the near-wall region. The scheme's performance was found to be better than LBM and DUGKS, and it could achieve comparable results with coarser grids than LBM.
- Isotropic Compressible Turbulence: The HGKS successfully simulated isotropic compressible turbulence up to supersonic regimes, which is a challenging area for many existing methods. The results provided benchmark data for these regimes.
- Parallel Efficiency: The parallel implementation demonstrated good scalability, with computational costs being comparable to high-order finite difference methods, albeit slightly higher (around 1.4 times).
Limitations and Future Work
- Grid Resolution: While the HGKS performs well, achieving very high Reynolds numbers or resolving extremely fine turbulent structures might still require significant computational resources and finer grids.
- Uniform Meshes: The current spatial reconstruction is based on WENO on uniform meshes. The authors acknowledge that for complex geometries or flows with significant variations in neighboring cells, this might affect accuracy.
- Future Work: The authors plan to investigate genuinely high-order gas-kinetic schemes on non-uniform and curvilinear meshes to eliminate geometrical errors. They also aim to explore more challenging compressible turbulence problems, such as shock-boundary interaction and supersonic turbulent boundary layers. Further investigations into higher Reynolds numbers and different flow configurations are also planned.
Summary updated