Comparison of Sub-Grid Scale Models for Large-Eddy Simulation using a High-Order Spectral Element Approximation of the Compressible Navier-Stokes Equations at Low Mach Number

TL;DR

This paper compares various sub-grid scale models for large-eddy simulation of low Mach number compressible flows using high-order spectral elements, evaluating their stability, accuracy, and ability to capture flow features.

Contribution

It introduces and assesses the performance of dynamic SGS models, especially the residual-based variants, in high-order spectral element methods for low Mach flows.

Findings

R-DSGS models are more robust than Vreman and Smagorinsky-Lilly.

Vreman and Smagorinsky models better resolve fine flow structures.

Nodal R-DSGS outperforms other models in most metrics.

Abstract

This study aims to identify the properties, advantages, and drawbacks of some common (and some less common) sub-grid scale (SGS) models for large eddy simulation of low Mach compressible flows using high order spectral elements. The models investigated are the classical constant coefficient Smagorinsky-Lilly, the model by Vreman and two variants of a dynamic SGS (DSGS) model designed to stabilize finite and spectral elements for transport dominated problems. In particular, we compare one variant of DSGS that is based on a time-dependent residual version (R-DSGS) in contrast to a time-independent residual based scheme (T-DSGS). The SGS models are compared against the reference model by Smagorinsky and Lilly for their ability to: (i) stabilize the numerical solution, (ii) minimize undershoots and overshoots, (iii) capture/preserve discontinuities, and (iv) transfer energy across different length scales. These abilities are investigated on problems for: (1) passively advected tracers, (2) coupled, nonlinear system of equations exhibiting discontinuities, (3) gravity-driven flows in a stratified atmosphere, and (4) homogenous, isotropic turbulence. All models were able to preserve sharp discontinuities. Vreman and the R-DSGS models also reduce the undershoots and overshoots in the solution of linear and non-linear advection with sharp gradients. Our analysis shows that the R-DSGS and T-DSGS models are more robust than Vreman and Smagorinsky-Lilly for numerical stabilization of high-order spectral methods. The Smagorinsky and Vreman models are better able to resolve the finer flow structures in shear flows, while the nodal R-DSGS model shows better energy conservation. Overall, the nodal implementation of R-DSGS (in contrast to its element-based counterpart) is shown to outperform the other SGS models in most metrics listed above, and on par with respect to the remaining ones.