Since a even order central differential scheme has a huge numerical dissipation to the calculation of turbulence boundary layer, it is very difficult to hold the U_tau and turbulent statistical qualities. Structures are decay very fast in my LES simulations, dynamic modeling + LUSGS scheme used on that.

My question is that,

Does anyone encounter it before? Do you have some suggestions to the numerical dissipation?

As far as I know, 2nd order central difference scheme can work fairly well in LES if the desired order of the statistical quantities is not too high ( say 2nd order).

Maybe your mesh is not fine enough, or try employing a wall model in your simulation.

regards