A comparative study on polynomial dealiasing and split form discontinuous Galerkin schemes for under-resolved turbulence computations

Andrew R. Winters, Rodrigo Costa Moura, Gianmarco Mengaldo, Gregor J. Gassner, Stefanie Walch, Joaquim Peiró, Spencer J. Sherwin

TL;DR

This paper investigates two methods for improving the accuracy and stability of high-order numerical simulations of turbulent flows when computational resources limit the resolution of small-scale features. The authors compare a standard "consistent/over-integration" technique with a newer "split-form" approach, finding that split forms offer comparable accuracy and improved robustness, particularly for challenging test cases.

What Problem Does This Paper Address?

Simulating turbulent flows at high Reynolds numbers is computationally expensive, often requiring "under-resolved" simulations where the smallest turbulent scales cannot be explicitly captured. This lack of resolution can lead to numerical errors, particularly aliasing errors in non-linear terms, which can destabilize simulations and corrupt results. Traditional methods like Discontinuous Galerkin Spectral Element Methods (DGSEM) face challenges in accurately and stably simulating these flows. This paper addresses the problem of suppressing these aliasing errors and improving numerical stability in under-resolved turbulence simulations using high-order DGSEM. Specifically, it compares two dealiasing strategies: consistent/over-integration (increasing quadrature points) and split-form discretizations (reformulating the governing equations).

Methods

The authors employ the Discontinuous Galerkin Spectral Element Method (DGSEM) for their simulations. They compare two dealiasing strategies:

  • Consistent/Over-integration: This involves using a larger number of quadrature points than strictly necessary for exact integration of the polynomial terms. This aims to reduce aliasing errors introduced by under-integration of non-linear terms.
  • Split-Form DGSEM: This approach reformulates the governing equations by averaging conservative and non-conservative forms of the non-linear terms. This is expected to have inherent dealiasing properties. The paper specifically examines the "DU" and "KG" split forms.

The study uses both a simplified 1D Burgers' equation to analyze the fundamental dealiasing mechanisms and the more complex inviscid Taylor-Green vortex (TGV) flow, a standard benchmark for turbulence simulations, to assess accuracy and stability. They utilize both local Lax-Friedrichs (LLF) and Roe-type Riemann solvers for interface flux calculations.

Datasets

Not discussed in the paper. The study uses a standard benchmark problem (inviscid Taylor-Green vortex) and a simplified 1D Burgers' equation for numerical experiments.

Key Insights and Ideas

  • Split forms offer a built-in dealiasing mechanism: Unlike consistent/over-integration, which relies on increased computational cost (more quadrature points), split forms appear to inherently suppress aliasing errors by averaging opposing numerical tendencies.
  • Split forms enhance robustness: The study demonstrates that split forms can stabilize challenging test cases (like the inviscid TGV) that would otherwise become unstable with consistent integration, especially at higher polynomial orders.
  • Comparable accuracy at moderate orders: For moderately high polynomial orders (around sixth order), split-form DGSEM schemes provide solutions with accuracy comparable to consistent/over-integration.
  • Potential for artificial energy conservation: At very high polynomial orders, both methods can exhibit an "energy-conserving bias," leading to unphysical features. This bias is particularly pronounced with the LLF flux due to its over-upwind nature.
  • Low-order solutions are less accurate: For low-order discretizations, significant differences emerge between split-form and consistent integration, with both yielding reduced solution quality compared to higher-order results.

Key Results

  • The KG split form, when coupled with the Roe numerical flux, provided stable and accurate solutions for the inviscid TGV flow at polynomial orders where consistent integration alone failed.
  • At moderately high polynomial orders (around sixth order), energy spectra and QR diagrams from split-form schemes closely matched those from consistent integration, indicating similar solution quality.
  • At very high polynomial orders, stable solutions obtained with split forms exhibited unphysical features, attributed to an energy-conserving bias, particularly when using the LLF flux.
  • For low-order cases (m ≤ 3), noticeable differences were observed between split-form and consistent integration results, with both showing lower solution quality.

Limitations and Future Work

  • High-order unphysical features: The paper notes that at very high polynomial orders, both methods can exhibit unphysical features related to an energy-conserving bias, especially with the LLF flux. This suggests that while split forms improve robustness, achieving physically accurate high-order solutions still requires careful consideration of numerical dissipation.
  • Curvilinear elements: The authors mention that future work could investigate the performance of split-form DGSEM for cases involving curved elements, where geometric aliasing errors have a different source than under-resolved turbulence.
  • Optimization of implementations: The paper states that optimizing the implementation of both dealiasing strategies is an ongoing research area.
  • Further investigation of split forms: While the KG split form was chosen as representative, further exploration of other split forms and their specific properties (e.g., kinetic energy preservation) could be beneficial.

Summary updated

Abstract

This work focuses on the accuracy and stability of high-order nodal discontinuous Galerkin (DG) methods for under-resolved turbulence computations. In particular we consider the inviscid Taylor–Green vortex (TGV) flow to analyse the implicit large eddy simulation (iLES) capabilities of DG methods at very high Reynolds numbers. The governing equations are discretised in two ways in order to suppress aliasing errors introduced into the discrete variational forms due to the under-integration of non-linear terms. The first, more straightforward way relies on consistent/over-integration, where quadrature accuracy is improved by using a larger number of integration points, consistent with the degree of the non-linearities. The second strategy, originally applied in the high-order finite difference community, relies on a split (or skew-symmetric) form of the governing equations. Different split forms are available depending on how the variables in the non-linear terms are grouped. The desired split form is then built by averaging conservative and non-conservative forms of the governing equations, although conservativity of the DG scheme is fully preserved. A preliminary analysis based on Burgers' turbulence in one spatial dimension is conducted and shows the potential of split forms in keeping the energy of higher-order polynomial modes close to the expected levels. This indicates that the favourable dealiasing properties observed from split-form approaches in more classical schemes seem to hold for DG. The remainder of the study considers a comprehensive set of (under-resolved) computations of the inviscid TGV flow and compares the accuracy and robustness of consistent/over-integration and split form discretisations based on the local Lax–Friedrichs and Roe-type Riemann solvers. Recent works showed that relevant split forms can stabilize higher-order inviscid TGV test cases otherwise unstable even with consistent integration. Here we show that stable high-order cases achievable with both strategies have comparable accuracy, further supporting the good dealiasing properties of split form DG. The higher-order cases achieved only with split form schemes also displayed all the main features expected from consistent/over-integration. Among test cases with the same number of degrees of freedom, best solution quality is obtained with Roe-type fluxes at moderately high orders (around sixth order). Solutions obtained with very high polynomial orders displayed spurious features attributed to a sharper dissipation in wavenumber space. Accuracy differences between the two dealiasing strategies considered were, however, observed for the low-order cases, which also yielded reduced solution quality compared to high-order results.
BibTeX
@article{winters2018comparative,
  author = {Andrew R. Winters and Rodrigo Costa Moura and Gianmarco Mengaldo and Gregor J. Gassner and Stefanie Walch and Joaquim Peiró and Spencer J. Sherwin},
  title = {{A comparative study on polynomial dealiasing and split form discontinuous Galerkin schemes for under-resolved turbulence computations}},
  journal = {Journal of Computational Physics},
  year = {2018},
  doi = {10.1016/j.jcp.2018.06.016},
}
Figure