A mathematical flavor of large-eddy simulation of turbulence

    Activity: Talk or presentationOral presentation


    Modern strategies for the computation of turbulent flow are aimed at reducing the dynamical complexity of Navier-Stokes solutions, while reliably retaining their primary flow phenomena. In large-eddy simulation (LES) a coarsened numerical description is achieved by spatial filtering which allows to externally select the physical detail that will be retained in the LES solution.
    Moreover, filtering gives rise to the central closure problem in LES in which the turbulent stresses need to be parameterized by an appropriate ‘subgrid’ model. The spatial filter defines all aspects of the flow smoothing. In particular, the filter should identify the subgrid model. However, subgrid models for LES have mainly been obtained through physical or mathematical reasoning that may be only loosely connected to a specified filter. For example, traditional subgrid modeling approaches are primarily based on dissipation and similarity considerations.
    We review the spatial filtering approach to large-eddy simulation and describe the intuitive dissipation and similarity requirements, commonly imposed on models for the turbulent stress. Then we present direct regularization of the nonlinear convective flux which provides a systematic framework for deriving the implied subgrid model. Regularization maintains the central transport structure of the governing equations. We illustrate the approach with Leray regularization and the Lagrangian averaged Navier-
    Stokes- model applied to turbulent mixing. These models display a strongly improved accuracy of predictions compared to dynamic subgrid models, as well as robustness at high Reynolds number. The treatment is finally extended to turbulent flows in which an additional self-organizing tendency competes with three-dimensional turbulence. E ects of rotation, buoyancy and interactions between embedded particles will be presented in a large-eddy context.
    In total, regularization modeling combines elements of mathematical analysis with smoothed flow descriptions that accurately represent multi-scale turbulent flow physics under widely varying conditions without having to resort to additional ad hoc modeling steps.
    Period16 Mar 2005
    Held atCentre for Analysis, Scientific computing and Applications (CASA), Netherlands
    Degree of RecognitionNational