The Turbulence Modeling Resource for RANS models provides code-verification cases where one can check that ones code converges to the correct solution of the specific RANS equations that one has tried to implement. This type of code-verification is difficult in LES due to the chaotic nature of the equations. The cases listed here should thus be viewed as validation (rather than verification) cases. However, it is still absolutely crucial to test a code on several different grids, both with different grid-spacings (to check convergence) and with different grid-anisotropies (to check the robustness). Note that LES (and WMLES) is notorious for being very sensitive to the grid-anisotropy (i.e., the cell aspect ratio) for wall-bounded turbulence (cf. the paper by Meyers and Sagaut, 2007, which is required and cautionary reading for every practitioner of LES).
It is important to recognize that these types of validation tests inherently test the full WMLES method, including the wall-model, the LES model (away from the walls), the numerics of the code, the inflow boundary conditions, the response of the flow to the domain size, etc. To validate the wall-model itself, we thus need to minimize the effect of all other errors. This favors cases with periodic boundary conditions and approaches where the grid and the wall-model are decoupled (i.e., where the wall-model can remain fixed under grid-refinement). For those errors that can’t be eliminated, one should instead systematically vary those parameters: e.g., test the wall-model with different subgrid models, with different numerics, with different distances from the inflow boundary, etc.
Since the validation tests inherently test the full WMLES method, one must also be judicious about which quantities to analyze. For example, the largest eddies in a channel flow are sensitive to the domain size, and therefore the wake region in a channel is sensitive to the domain size. Therefore, discrepancies between DNS and WMLES in the core of a channel are not a sign of wall-model inadequacy. Similarly, it is actually easier to get the velocity fluctuations right than the mean velocity profile — this is a bit counter-intuitive at first, but is related to the fact that the turbulence is primarily produced locally. Therefore, an error near the wall will affect the mean velocity profile throughout the boundary layer, but will only affect the Reynolds stresses locally. The main validation quantity for WMLES is, therefore, the mean velocity profile in the region above the wall-model and below the outer part of the boundary layer.
The validation cases included here are:
- Incompressible channel flow at friction Reynolds number of 5200, with DNS data from Lee and Moser (2015). With periodic boundary conditions and the large body of work on turbulent channels, this is close to a verification case as one can get.
- Separation over a smooth ramp, with DNS data from Ali Uzun and Mujeeb Malik. This case was part of the High-Fidelity CFD Verification Workshop in 2022 and 2024.