Effect of smearing method on band gap

[Neutrino]

I'm studying a semiconductor for which GGA predicts a metallic ground state. Now I'm testing the effect of Spin-Orbit-Coupling on improving the GGA solution. If I use ISMEAR=0 with SIGMA=0.05 eV (recommended VASP setup for semiconductors) the calculations never converge and the OUTCAR shows a oscillations of about 0.1 electrons going back and forth between "the supposedly valence band" and "the supposedly conduction band" which are separated by ~0.1 eV. Then I thought I may test the recommended smearing for metals (ISMEAR=1, SIGMA=0.2 eV). Surprisingly everything converges very fast (I can even do ISIF=3 to get lattice constant), but the big surprise is that OUTCAR indicates a gap of 1 eV! The system has 228 electron and I examine the band gap by

grep " 228 " OUTCAR
and
grep " 229 " OUTCAR

any hints about the reasons for this observed gap with ISMEAR=1?

[Neutrino]

I think I resolved this. The metallicity at the GGA level is due to an isolated narrow d-band in the middle of which the Fermi energy lies. Spin-orbit coupling splits this narrow band and opens a gap in the middle of this band in the order of 0.1 eV. This gap separates the HOMO from the LOMO. Now Gaussian smearing with sigma=0.05eV fails completely to treat this situation and no convergence is achieved. Methfessel-Paxton with smearing of 0.2 eV fills the segment the original narrow d-band which is supposed to be unoccupied but because of the large smearing value, it gets populated with electrons. So after applying the MP smearing it looks like that the whole narrow d-band became part of the valence band. The next empty band (which is apparently due to anion states) is far away and looks like a conduction band. Thus, in total the MP smearing gave me the impression that I have a semiconductor which is of course not the case.

I concluded that I need to converge this case with Gaussian smearing of 0.001 eV but probably with a denser k-point grid

Comments are appreciated and I hope this explanation helps other who may face the same situation.