could someone give me a good method to set a correct time step in LES simulations. The only infos given by the manual says : "under 20 iteration to converge"

i'm not sure this is a good argument to put in a scientific work

I've tried using a CFL condition witch could be interpreted as : "the information cannot jump to the next cell during one time step" but the CFL conditions given in the manual refers to the coupled solver and i use the segregated one.

any idea ?

You can just consider the max. streamwise velocity and the smallest cell size in the same direction to evaluate a time step (this will be cfl~1).

For exemple channel flow with streamwise x direction with vmoy ~1 and DeltaX =0.01 => Delta t= 0.01 ....

that's what i'm actually using, while adapted after some readings to CFL=0.3

but this is true only for stability in explicit time integration, is there a good justification for this in an implicit second order time integration ?