TL;DR
This paper introduces a novel, highly efficient numerical method for simulating incompressible turbulent flows. By combining a high-order compact difference scheme with an approximate factorization approach and a fast Fourier transform-based pressure solver, the method achieves greater accuracy and efficiency than traditional second-order schemes, particularly for capturing turbulence statistics and energy spectra in wall-bounded flows.
What Problem Does This Paper Address?
Simulating turbulent flows, especially those near solid boundaries (wall-bounded turbulence), is crucial for understanding and designing many engineering systems. While high-fidelity methods like Direct Numerical Simulation (DNS) and Large Eddy Simulation (LES) offer accuracy, they are computationally expensive. Traditional finite difference (FD) and finite volume (FV) methods, often used for their flexibility in complex geometries, typically employ second-order schemes. However, the truncation error of these low-order schemes can be comparable to or even larger than the subgrid-scale terms in LES, limiting their accuracy. The challenge, therefore, is to develop numerical methods that are both highly accurate (high-order) and computationally efficient, especially for the complex velocity-pressure coupling inherent in incompressible flows.
Methods
The core of this work is the development of a new numerical solver that integrates several key techniques:
- High-Order Compact Difference Scheme: The paper utilizes a fourth-order compact difference scheme for discretizing spatial derivatives and interpolations in the incompressible Navier-Stokes equations. Compact schemes use a smaller stencil of grid points but achieve higher accuracy compared to traditional schemes with larger stencils.
- Approximate Factorization: This technique is employed to simplify the implementation of a semi-implicit time-advancement scheme. It breaks down a complex implicit system into a series of simpler, sequential steps, making the overall solver more manageable and efficient.
- Semi-Implicit Time Advancement: The method uses an explicit Adams-Bashforth scheme for convection terms and a semi-implicit Crank-Nicolson scheme for viscous terms. This combination allows for larger time steps than fully explicit methods, improving computational efficiency.
- Fully Staggered Grid: The solver is implemented on a fully staggered grid, which is known to prevent spurious pressure oscillations in incompressible flow simulations.
- Fast Fourier Transform (FFT) for Pressure Poisson Equation: The pressure Poisson equation, a computationally intensive part of incompressible flow solvers, is solved efficiently using the FFT. This is a standard but highly effective approach for problems with periodic boundary conditions.
The paper also details specific discretizations for first-order derivatives, interpolations, and second-order derivatives, including one-sided schemes for boundary conditions. For the semi-implicit linear system, an approximate factorization approach is used to decompose the inversion into solving three tridiagonal systems, significantly reducing computational cost.
Datasets
The paper does not introduce new datasets. Instead, it validates its numerical method against established benchmark problems and compares its results with existing DNS databases:
- Benchmark Problems: Taylor-Green vortices, Burggraf flow (forced lid-driven cavity), and lid-driven cavity flow.
- DNS Databases: Comparisons are made with spectral DNS data for turbulent channel flows from sources like Moser et al. [5] and Lozano-Durán & Jiménez [8], and for high Reynolds number flows from Lee & Moser [9].
Key Insights and Ideas
- High-Order Accuracy Pays Off for Turbulence: The study strongly suggests that high-order schemes are crucial for accurately capturing turbulence statistics, particularly energy spectra. While low-order schemes might suffice for basic flow features, they struggle with the finer scales of turbulence.
- Compact Schemes Enhance Spectral Accuracy: The compact difference scheme, due to its spectral-like properties, demonstrates superior performance in resolving higher wavenumbers (smaller scales) compared to traditional second-order schemes, leading to more accurate energy spectra.
- Approximate Factorization Simplifies Implicit Schemes: The paper highlights the effectiveness of approximate factorization in making semi-implicit time-advancement schemes with high-order compact spatial discretizations practical to implement.
- Trade-offs in Projection Methods: The authors discuss the challenges of achieving an exact projection method with high-order compact schemes on a fully staggered grid, leading to the proposal of approximate projection methods (U4P4-AP) as a compromise.
Key Results
- Validation: The numerical method was validated against benchmark problems (Taylor-Green, Burggraf, lid-driven cavity flows), showing high-order accuracy for velocity and pressure.
- Turbulence Statistics: For turbulent channel flows at low and moderate Reynolds numbers (Reτ = 180, 590), the high-order compact scheme provided more accurate predictions of turbulence statistics (e.g., r.m.s. velocity fluctuations, Reynolds stresses) and, notably, energy spectra compared to the second-order central difference scheme.
- Spectral Accuracy: The paper demonstrates that the high-order scheme significantly improves the prediction of one-dimensional streamwise pre-multiplied spectra, capturing finer details of turbulent energy distribution, especially at higher wavenumbers.
- Efficiency: While not explicitly quantified in terms of speedup, the use of semi-implicit time stepping and FFT for the pressure Poisson equation implies improved computational efficiency over fully explicit or iterative solvers. The ability to use larger time steps is a key benefit.
- WMLES Performance: For high Reynolds number wall-modeled LES (Reτ = 1000) with a coarse grid, the high-order scheme showed better agreement with DNS data for mean velocity and spectral densities, suggesting its necessity when numerical errors might be comparable to SGS stresses.
Limitations and Future Work
- Vertical Direction Discretization: The paper notes that in the vertical direction of channel flow simulations, a second-order central difference scheme was used. This means the overall method is not strictly fourth-order accurate in all directions for these specific applications.
- Approximate Projection for DCT: For Neumann boundary conditions in the pressure Poisson equation, an exact projection method with high-order compact schemes on a fully staggered grid was found to be challenging. This led to the development of approximate projection methods (U4P4-AP).
- Future Work: The authors suggest that the code is parallelized using OpenMP, implying potential for further parallelization for larger-scale simulations. The paper also implicitly points towards further exploration of high-order schemes for WMLES, where coarse grids make numerical errors significant. The authors also mention that while low-order statistics might not always improve, higher-order statistics and spectra do benefit from high-order schemes, suggesting further investigation into these aspects.
Summary updated