This is stimulated by the last message. I am also interested in LES. In my opinion, LES is the simplist model for (large scale) turbulence. The only thing one needs to do is to write an effecient code, and get enough computer power (that's not easy though). However, I have always heard that one needs small enough grid size in order to get some meaningful results - if the grid size is too large, not enough sink provided, grid energy gets equipartitioned too all the wavelengths. The question is then, what is the largest grid size one can use in order not to get energy equipartition? Intuitively, one would think that the criterion must have something to do with the grid Rynolds number (peclet number). Anyone knows the conclusion or literatures about this? Discussions appreciated.

Li Jin

To compute a LES you therocally have to resolve the field in order to calculate all the anisotropic turbulence, because most of the LES models represent only isotropic turbulence. So the size of the mesh is linked to this notion. Some autors connect this with the Taylor microscale. This scale implies the size of the mesh to be proportional to Re^(-0.5). Practically, the size of the mesh is rather connected whith the characteristic turbulent structures of the flow you study. For instance, for a boundary layer, you have to give a good representation of the streaks. For further details see for example : "Turbulence in fluids" by M. Lesieur.