How to find appropiate SIGMA value

[daniela_herrera-molina]

I'm doing calculation in a single molecule, that is, I constructed a vacuum box and placed my molecule in the middle. This molecule have a 4d (transition metal) atom.

So, I'm using Gaussian smearing.

With default SIGMA (0.2) I get nice peaks, but when I decrease the SIGMA value, those peaks changed in energy and some of them dissapear. Not sure if this problem is related to the fact I'm doing the calculations only in gamma point, because I have a big supercell (25x25x25 A).

So, my question is: How should I select the SIGMA value?

My other question is: When I run a GGA calculation, I can see the d-orbitals spin up and down in the PDOS, but when I run a GGA+U calculation I mised one d-orbital (spin down) above fermi level. What should I do in order to avoid this problem?

[ferenc_karsai]

Please have a look at the following page:
https://www.vasp.at/wiki/index.php/Smearing_technique

Regarding your second question, you can try to vary/optimize "+U" parameters. Please have a look at this site:
https://www.vasp.at/wiki/index.php/DFT%2BU:_formalism

[daniela_herrera-molina]

Thanks for you reply.

According to the problem with GGA+U, maybe I was not clear about that.

My problem is that when I plotted the PDOS of d-orbitals, with the GGA calculation I get the full d-levels visible there (up and down spin), but when I added U (U = 5 eV), one of the down spin associated with this level disappears from the PDOS in the entire energy window.

I tried different transition metals. For some of them (as Mo), I don't have this problem, but for some of them (as Nb) it shows that problem.

I'm including the PDOS for both of them, GGA and GGA+U (comparison between transition metal Nb and Mo).

[ferenc_karsai]

Please upload your calculation according to the forum guidelines.

[daniela_herrera-molina]

Thank you.

Here, I am attaching the input and output files for the calculation with U = 5 eV. Additionally, I performed some calculations by varying U, and I noticed that the peak associated with the d-orbital (down spin above the Fermi level) starts to disappear for U > 2.5 eV, which should not happen. I am also attaching an image showing how the peak disappears as U increases. Could there be an error in my approach, or is there a recommended value of U in this case?

[ferenc_karsai]

I've rerun your calculation and talked to a colleague. I can reproduce your results, but we concluded that everything is alright.

The +U method can lead to quite strong redistribution of the orbital characters. You can see that by the d_xy peak, who's intensity increases with increasing U.
If you look at your total density of states (in the calculation with U=5ev) you would see that the last peak ends at 4.7eV. The spin down peak of the last peak should be around that energy. The spin down states corresponding to this peak would need probably orbitals from the neighboring C and H atoms to hybridise with. Since not enough bands are set in the calculation there are not enough "partners" to hybridise with and the bands vanish.
We think that everything is fine with the calculation and if you don't necessarily need to show peaks at that energy then just leave it as it is. Otherwise you can try to increase NBANDS and see if the peaks appear again.