negative occupancies with tetrahedron method

Message

eariel · #1 Post by **eariel** » Tue Feb 05, 2013 12:31 pm

I find several bands with occupancies different from 0 and 1 while using ISMEAR = -5. I have values smaller than zero and larger than 1. I have always understood that with ISMEAR = -5 one has occupantions 0 or (or 2 for non-spin-polarized calcualtions).

I read the articule by Bl?chl et al (PRB 49, 16223 (1994)) and there the occupations are 0 and 1.

In a different calculation, after obtaing the SCF state using ISMEAR=2, I ran a calculation using ALGO=None with ISMEAR = -5. Then I see the following line in the OUTCAR

None: do nothing, only one-electron occupancies are recalculated

and again I got strange occupations
grep ' 484 ' OUTCAR-none
484 3.1231 1.00000
484 3.6826 0.79753
484 3.6823 0.79795
484 4.3278 -0.30484
484 3.6901 0.78905
484 3.2231 1.00000
484 3.8080 0.60281
484 3.8083 0.60224
484 4.4551 -0.39050
484 3.8063 0.60579
484 3.1231 1.00000
484 3.6826 0.79753
484 3.6823 0.79794
484 4.3278 -0.30485
484 3.6901 0.78904
484 3.2231 1.00000
484 3.8080 0.60280
484 3.8083 0.60223
484 4.4551 -0.39051
484 3.8063 0.60579

Is it a bug? I am using VASP 5.2.12

Regards, Eduardo

#2 Post by **admin** » Tue Feb 05, 2013 2:56 pm

This is not a bug. Bloechls correction to the tetrahedron method can lead to occupancies >2 or <0.

eariel · #3 Post by **eariel** » Tue Feb 05, 2013 5:42 pm

Thank you. Where can I read the method to obtain the occupancies? In Bloechl's paper the occupancies are fixed.

eariel · #4 Post by **eariel** » Thu Feb 07, 2013 1:46 pm

So, having no answer, I wonder if the optimization of the occupancies is an unpublished recipe ? Is there any keyword in VASP to enforce the occupancies to their physical values at 0 K ?

#5 Post by **admin** » Thu Feb 07, 2013 2:24 pm

Modified tetrahedron method was published by Bloechl in PRB 49(1994)16223. The method itself allow only slight deviation of occupancies from phasical values. The values like 0.5 are caused by symmetry of energy levels.
Try another smearing, e.g. Gaussian (ISMEAR=0) and tune the SIGMA to small values until you will get occupancies only 0 and 1.

eariel · #6 Post by **eariel** » Thu Feb 07, 2013 2:49 pm

I began this thread mentioning that I had read Bloechl in PRB 49(1994)16223 (see my first post), and I found no mention to occupations within the tetrahedron method. I see a discussion around Eq. (27), on page 16231, but I understand that this concerns smearing methods and their difference with the tetrahedron method. Yes, I understand the other smearing methods, and the way to tune SIGMA. I also read in the manual that the tetrahedron method is the most precise method for total energies, even in metals if the forces are not need. There is mention to unphysical occupations if the Methfessel-Paxton method is used with non-metals, but no mention to unphysical occupations (i.e., negative) with the modified tetrahedron method.

eariel · #7 Post by **eariel** » Thu Feb 07, 2013 2:54 pm

In Bloechl's paper it is mentioned a possibility to make the method variational. It is juat a possibility mentioned in Ref. 21. Is this comment related with the optimization of occupation numbres in VASP?