Key ingredients for wall-modeled LES with the Lattice Boltzmann method: Systematic comparison of collision schemes, SGS models, and wall functions on simulation accuracy and efficiency for turbulent channel flow

Abstract

In this study, we consider different combinations of collision schemes, wall functions, and subgrid scale (SGS) models to simulate the bi-periodic turbulent channel flow at . The study is carried out on a lattice stencil, where the considered collision schemes are the Multiple Relaxation Times (MRT), the Hybrid Recursive Regularized Bhatnagar-Gross-Krook (HRR), and the parameterized Cumulant scheme. The considered SGS models are the Smagorinsky, the Wall-Adapting Local Eddy-viscosity (WALE), and the Vreman model. The Cumulant scheme utilizes its intrinsic implicit SGS model. The turbulent velocity profile is modeled with the following wall functions: the Reichardt, the Musker, and a combination of the Werner and Wengle and the Schmitt (Power-law) function. To assure an impartial comparison, all these ingredients are implemented in the same infrastructure, the open-source software Musubi. The comparison of the considered wall functions shows that the Musker function offers a good compromise between accuracy and performance. When comparing the considered SGS models, although the WALE model delivers the most accurate results, the Vreman model offers the fastest computation. On average, the parallel performance improves by . Amongst the considered collision schemes, the Cumulant outperforms the competitors accuracy-wise. However, for the considered test case, the fastest collision scheme is the MRT. Our investigations show that for the considered test case, the best results in terms of accuracy and performance are delivered by the combination of the Cumulant scheme with its implicit SGS model and the Musker wall function.