Dongsu Ryu: What I found interesting and also worrisome is the following. The agreement among my three runs is fair (comp_dmpos_tvd.gif) and the agreement among your two runs is also fair (comp_dmpos_gadget.gif). But the agreement between yours and mine is not satisfactory (comp_dmpos_[64,128256].gif). Do you think that it is due the different schemes used for dm (PM for mine and Tree or P^3M (?) for yours)? But I expect the difference in the scheme should cause differences in the structures of collapsed objects (so would cause differences in the mass function), not differences in the distribution of those objects. Klaus Dolag: I think i found the origin. If i setup the gas particles for the Gadget runs on a unperturbed grid with zero velocity (to mimic the setup of the grid codes ICs) i reproduce the positions of the objects quite well. Moreover, in doing so i even get similar FoF mass functions than the PM run, showing the same lack of medium mass objects. However such runs differ in the phase diagrams from the one with perturbed gas density and velocity field, specially at high redshift. This has to be investigated specially as i expect the shock distribution to be influenced by this. Dongsu Ryu: Just one correction. We always perturb gas initially in grid-code simulations. The only simulations done without perturbations in gas were for the purpose of comparison with sph (there are already couple of papers about it before). Anyway, as we are discussing, imposing an exactly same initial condition is important in comparison studies. Since it is difficult to perturb gas in an exactly same way in grid codes and sph, I prefer not to perturb gas initially in our comparison study...