De-aliasing on non-uniform grids: algorithms and applications

Robert M. Kirby, George Em KarniadakisJournal of Computational Physics 2003 · DOI

TL;DR

This paper introduces a "super-collocation" technique for de-aliasing non-linear terms in spectral methods on non-uniform grids, analogous to Fourier spectral methods. By using a higher order of quadrature for non-linear terms than for linear ones, they significantly enhance the accuracy and stability of simulations, particularly for challenging fluid dynamics problems.

What Problem Does This Paper Address?

Spectral methods are powerful tools for simulating fluid dynamics, but they often struggle with non-linear terms (like those in the Navier-Stokes equations). When evaluating these non-linear terms, a common practice is to use quadrature points in physical space. If the number of quadrature points is insufficient, it leads to "aliasing errors," which can cause numerical instability and inaccurate results, especially in simulations with complex geometries or at higher Reynolds numbers. Standard spectral methods often use the same number of quadrature points for both linear and non-linear terms, which is only sufficient for exact integration of linear terms. This paper addresses the problem of how to accurately and stably compute non-linear terms in spectral methods, particularly when using non-uniform grids, which are common in real-world applications.

Methods

The core of the paper is the development and application of "super-collocation" for de-aliasing. This technique involves using a higher number of quadrature points (Q) for evaluating the non-linear terms than what is strictly required for exact integration of the polynomial basis functions (M). Specifically, for quadratic non-linearities, they propose using $3M/2$ quadrature points, and for cubic non-linearities, $2M$ points. This is achieved by decoupling the number of quadrature points used for the Galerkin projection of non-linear terms from those used for linear terms.

The authors derive the necessary number of quadrature points based on the order of the integrand and the polynomial degree of the basis functions. They then test this approach using:

  • A simple test case: Evaluating the modal representation of a polynomial, squaring it, and projecting back to modal space to observe the effect of under-integration.
  • The 1D viscous Burgers equation: Solved using both continuous and discontinuous Galerkin methods to demonstrate the impact on stability and accuracy.
  • Direct Numerical Simulations (DNS): Applied to three complex fluid flow problems:
  • Incompressible turbulent flow in a triangular duct.
  • Incompressible turbulent flow in a channel at a moderate Reynolds number ($Re_c = 395$).
  • Compressible flow around a pitching airfoil.

Datasets

Not discussed in the paper. The paper focuses on numerical methods and their application to established fluid dynamics problems, using analytical solutions or benchmark DNS results for comparison.

Key Insights and Ideas

  • "Under-integration" is a significant source of aliasing errors: Using the minimum number of quadrature points required for linear terms is insufficient for non-linear terms, leading to instability and inaccurate results.
  • Super-collocation effectively de-aliases: By increasing the quadrature points for non-linear terms, aliasing errors can be eliminated, leading to stable and accurate simulations.
  • The rules are generalizable: The derived rules for quadratic ($3M/2$) and cubic ($2M$) non-linearities are presented as semi-empirical guidelines for spectral methods on non-uniform grids.
  • Over-integration is a viable alternative to filtering: While techniques like Spectral Vanishing Viscosity (SVV) can stabilize simulations, over-integration directly addresses the root cause of aliasing errors by ensuring accurate computation of non-linear terms.
  • Stability is paramount for high-Reynolds number flows: For simulations where viscosity is insufficient to suppress instabilities, over-integration becomes a crucial tool.

Key Results

  • Burgers Equation: For the viscous Burgers equation, using $Q=M+1$ quadrature points led to unstable solutions, while $Q=3M/2$ resulted in stable solutions. This was demonstrated for both continuous and discontinuous Galerkin methods.
  • Triangular Duct Flow: Simulations at $Re=1250$ showed that using $Q=M+1$ quadrature points led to less stable wall shear force histories compared to $Q=2M$ or $Q=3M/2$, which yielded statistically identical and stable results.
  • Channel Flow ($Re_c=395$): For a coarse resolution simulation, over-integration with $3M/2$ quadrature points achieved asymptotic stability, comparable to results obtained with SVV filtering, and produced turbulent intensities that agreed well with benchmark DNS.
  • Pitching Airfoil: Over-integration stabilized the simulation of compressible flow around a pitching airfoil, allowing for accurate lift coefficient predictions up to higher angles of attack, whereas SVV alone eventually became unstable.

Limitations and Future Work

  • Computational Cost: The primary drawback of over-integration is the increased computational cost due to the higher number of quadrature points used for non-linear terms. While the asymptotic scaling remains the same, the constant factor is larger.
  • Theoretical Justification: The authors acknowledge the need for a rigorous theory to provide deeper justification and insight into the stability and accuracy enhancements observed with their proposed algorithms.
  • Future Work: The paper suggests that future research should focus on developing this rigorous theory.

Summary updated

BibTeX
@article{kirby2003de,
  author = {Robert M. Kirby and George Em Karniadakis},
  title = {{De-aliasing on non-uniform grids: algorithms and applications}},
  journal = {Journal of Computational Physics},
  year = {2003},
  doi = {10.1016/s0021-9991(03)00314-0},
}

Loading related content...

Figure