I wonder why the default nempty does not depend on the number of atoms in the unit cell. How about something like nempty = 5 * natmtot * nspinor or so? Or is the large number of empty states only required for calculations in which we use the second-variational eigenvectors? At least for metals the number of empty states to act as a buffer should also depend on the actual number of electrons in the system.
Further, the convergence criteria are not normalized; thus, the relative convergence gets tighter with increasing number of atoms. Is there a reason for this?
after some more code reading I found that the RMS change in effective potential is actually kind of normalized with the size of the mixing vector, which of course increases with system size. However, the total energy is not normalized, so my above question still holds for epsengy.
John Kay Dewhurst
Your first request (nempty = 5 * natmtot * nspinor) has now been implemented.
I'll think about normalizing the total energy and talk to SS and LN about it.