TL;DR
This paper introduces a family of "all-Mach" number numerical schemes for simulating compressible fluid flow governed by the Euler equations. These schemes effectively handle flows across a vast range of speeds, from very slow (low Mach) to supersonic, by implicitly treating acoustic waves while explicitly handling slower material waves. This approach overcomes the computational limitations imposed by acoustic waves in explicit methods, leading to more efficient and accurate simulations, especially in low-Mach regimes.
What Problem Does This Paper Address?
Simulating compressible fluid dynamics using numerical methods often faces a significant challenge: the vast difference in speeds between different types of waves. The Euler equations, which describe this behavior, contain both slow-moving "material" waves (like the bulk flow of the fluid) and extremely fast "acoustic" waves (sound waves). Explicit numerical schemes, which are generally simpler to implement, are constrained by the Courant-Friedrichs-Lewy (CFL) condition, requiring time steps to be small enough to resolve the fastest waves. In low-Mach number flows, where the fluid moves slowly but sound travels at its usual speed, this means the time step is dictated by the fast, often less important, acoustic waves, leading to computationally prohibitive simulation times. Furthermore, standard numerical methods designed for high-Mach flows can introduce excessive numerical dissipation when applied to low-Mach flows, degrading accuracy. This paper aims to develop schemes that are accurate and efficient across the entire spectrum of Mach numbers, from very low to high.
Methods
The core innovation lies in a "linearly implicit" approach combined with a "pressure splitting" or "flux splitting" strategy. The key idea is to treat the terms responsible for acoustic waves implicitly, thereby removing the acoustic CFL restriction. These terms are typically discretized using central differences. Conversely, terms related to material waves, which are less stiff and more relevant in low-Mach flows, are treated explicitly using upwind schemes or local Lax-Friedrichs fluxes. This selective implicit treatment of stiff acoustic terms allows for much larger time steps without sacrificing stability or accuracy. The paper explores two main splitting strategies:
- Pressure Splitting: This approach explicitly treats convective terms and implicitly handles pressure terms.
- Flux Splitting: This method splits the fluxes based on their characteristic speeds, treating the material wave component explicitly and the acoustic wave component implicitly.
The paper also discusses the implementation of second-order accurate schemes in space, achieved through piecewise linear reconstructions, and explores high-order time discretization using implicit-explicit Runge-Kutta methods.
Datasets
Not discussed in the paper. The paper focuses on numerical methods and their validation using established test cases.
Key Insights and Ideas
- Implicit Treatment of Acoustic Waves is Key: The fundamental insight is that by treating the fast acoustic waves implicitly, the severe time step restriction imposed by the CFL condition can be lifted, enabling efficient simulations of low-Mach flows.
- Splitting Strategies Matter: Both pressure splitting and flux splitting approaches demonstrate effectiveness, with the choice potentially influencing accuracy and robustness in different flow regimes.
- Low-Mach Accuracy and Efficiency: Implicit schemes offer a significant advantage in low-Mach number flows, providing both higher accuracy and computational efficiency compared to explicit methods for the same level of precision.
- Robustness Across Mach Numbers: The developed "all-Mach" schemes show promise in handling a wide range of flow speeds, from very low to moderate Mach numbers, without excessive numerical dissipation.
Key Results
The paper presents numerical results for various test cases, including:
- Sod and Lax Shock Tubes: These tests, at intermediate Mach numbers, demonstrate the robustness of the proposed schemes in capturing shocks.
- High Mach Test Case: This extreme test highlights the stability of the P-split LF scheme, while other semi-implicit schemes become unstable.
- Acoustic Waves: Tests with acoustic waves at low Mach numbers show that the schemes can effectively filter out these waves when a material CFL restriction is adopted, while maintaining stability.
- Nozzle Flow: Simulations of subsonic and transonic flows in a nozzle demonstrate the schemes' ability to accurately capture flow behavior across different Mach regimes. The P-split LF scheme is noted for better resolution than explicit LF.
- Convergence Studies: For nozzle flows, convergence tests show second-order accuracy for all methods, with implicit schemes generally being more accurate and efficient in low-Mach regimes.
- 2D Euler Equations: Initial tests with the 2D Euler equations on a Sod shock tube problem show that the 2D code is more accurate than the 1D code for comparable computational effort.
- Gresho Vortex: This test case, used for convergence analysis, shows that the scheme preserves the stationary solution for a wide range of Mach numbers.
- Vortex Dipole: Comparison with a spectral solution for incompressible flow shows good agreement, indicating the scheme's capability in the low-Mach limit.
Limitations and Future Work
- No Single Best Scheme: The authors acknowledge that no single scheme outperforms all others across the entire Mach number range.
- Computational Cost in 2D/3D: While the implicit schemes are efficient in 1D, the need to solve linear systems can become computationally costly in multi-dimensional problems. This is identified as an area for future work.
- Multi-Material Flows: The authors plan to extend their work to multi-material flows, which introduce additional time scales and complexities.
- Space-Time Adaptivity: Future research will explore space-time adaptivity in multi-dimensional simulations.
- Higher-Order Schemes: The paper focuses on first and second-order schemes but notes that the underlying techniques can be extended to higher orders.
Summary updated