Simulating turbulence in complex geometries

We first review the state-of-the art in direct numerical simulation and present a new class of spectral methods on unstructured grids for handling complex-geometry domains. Subsequently, we concentrate on the classical problem of the turbulent wake behind a circular cylinder and compare the accuracy of spectral DNS versus other LES results available in the literature. We find that DNS provides consistent agreement with the experimental results, but that LES predictions are inconsistent and depend strongly on the interaction between numerical discretization and the subgrid model. We also demonstrate via a simple vorticity-based analysis of the turbulent near-wake that eddy-viscosity models are inappropriate for sudgrid modeling. In contrast, preliminary a priori tests suggest that scale-similarity models may be a good candidate. We close the paper by forecasting the use of dynamic DNS and comment on its role in simulating turbulence in complex geometries.