Density of empty states sensitive to NBANDS parameter

[ye]

Recently in attempting to calculate the d-band center of a late transition metal surface, we realized that the number of empty states (i.e., those above the Fermi level), although small, nonetheless increases with the number of bands specified through the NBANDS parameter, so that the d-band center when integrating over the entire d-band almost linearly upshifts with increasing NBANDS. We are using PBE-PAW and have taken care that all other parameters (k-points, smearing, enough bands etc.) are all converged. Does anyone have an explanation for this phenomenon? Thanks in advance.

[forsdan]

I think I am missing the point. It is obvious that you will get more empty states in the DOS if you solve for more eigenvalues (i.e. by increasing NBANDS). Then you also need to know the where the entire d-band is localized in order to determine center of gravity. So if you don't have the entire d-band from the start and then gradually include more d-staes in the determination, then you will get an increase in the center of gravity.

Have I missunderstood your question?

Cheers,
/Dan
<span class='smallblacktext'>[ Edited Thu Oct 21 2010, 12:44PM ]</span>

[ye]

Thanks, Dan. I think I did capture the entire d-band. There is a wide range of energy at the top of the energy scale in which there is zero DOS, as shown in DOSCAR. The problem is, when I increased NBANDS, the last non-zero state, which e.g. used to appear at +6 eV vs. Ef, now appears at +8 eV vs. Ef. My EMIN,EMAX are -20 and +10 eV. There is 0 DOS between the last non-zero state and +10 eV.

Is this phenomenon something to be expected?

[forsdan]

The DOS at higher energies is not evaluated if you don't include sufficiently many bands. Therefore an empty DOS in the upper energy range doesn't necessarily imply that you have a band gap. The accuracy of the unoocupied bands is, however, another question.

/Dan