convergence problem

[apple]

hi,
I have been trying to run a single point calculation with ISMEAR=-5 and ISPIN=2 for a metallic structure converged with ISMEAR=1, SIGMA=0.1, and ISPIN=2, but cannot achieve the electronic convergence. I tried to set ALGO=48 from 38 and increase EDIFF from 1E-05 to 1E-06, but it did not help. What could be a reason for the bad convergence with ismear=-5 and how to overcome it?
Can someone please comment on my question.
Thanks.

[jlbettis]

Apple,
I think your choice of ISMEAR may be the cause of your problem. In metals the energy difference between the conduction and valence bands is infinitesimally small. Therefore, in general the energy minimisation will take a long time to resolve. Setting ISMEAR = -5 imposes the tetrahedron method with Blochl corrections. The VASP manual states not to use the tetrahedron method when dealing with metals because this method is not variational with respect to the partial occupancies. More bands are necessary, specifically unoccupied bands, for proper convergence of metallic systems.

If you use the CHGCAR and WAVECAR from your previously converged calculation, keep ISMEAR = 1, and slowly decrease SIGMA from 0.1 --> 0.05 --> 0.01. Then use the CHGCAR and WAVECAR from the converged SIGMA = 0.01 calculation as a starting point for the ISMEAR = -5 calculation. The trick is to slowly decrease SIGMA, in a stepwise manner, so that upon running the ISMEAR = -5 calculation you are very close to the correct electronic structure.






<span class='smallblacktext'>[ Edited Tue Jul 03 2012, 08:45PM ]</span>

[apple]

Thank for the explanation.