Abstract
Most applications of the dynamic subgrid‐scale stress model use volume‐ or planar‐averaging to avoid ill‐conditioning of the model coefficient, which may result in numerical instabilities. Furthermore, a spatially‐varying coefficient is mathematically inconsistent with the original derivation of the model. A localization procedure is proposed here that removes the mathematical inconsistency to any desired order of accuracy in time. This model is applied to the simulation of rotating channel flow, and results in improved prediction of the turbulence statistics. The model coefficient vanishes in regions of quiescent flow, reproducing accurately the intermittent character of the flow on the stable side of the channel. Large‐scale longitudinal vortices can be identified, consistent with the observation from experiments and direct simulations. The effect of the unresolved scales on higher‐order statistics is also discussed.