fermi level problem

[Cyrille]

Dear all

I am performing calculations on a structurally very simple but magnetically very complicated system: FeCr in B2 structure.
B2 structure is just based on bcc structure but Fe is sitting at the corner of the cube and Cr at the centre (or vice versa:-) so it is simple cubic.
Since I realized that FHI pseudo/basis was doing a very bad job (in fact totally wrong) I am using HGH pseudo (Tier0)
Now the magnetization of the system is very reasonable (in agreement with other DFT results) but when I plot the band structure I am convinced that the Fermi level is not at the right position.
Indeed the band structure looks very reasonable but the Fermi level (zero of energy) is clearly too high.
I wonder if someone has experienced this type of problem.

I have attached my input and log file..

Thanks in advance

Cyrille

[Cyrille]

It's me again..
Forget about my post it seems to be again related with the pseudo/basis...

I will do some additional tests..

Cyrille

[Jess Wellendorff]

Just out of curiosity, was it the use of the Tier0 basis set for HGH that caused problems?

[Petr Khomyakov]

I have looked at your files. As a general suggestion, you may want decreasing the tolerance parameter for the total energy from 10^-4 to 10^-7 Hartree. In addition, increasing the number of k-points might also be helpful, and try using an odd k-point grid, e.g., 31x31x31, to shift the mesh to the Gamma point as that is usually considered as a good choice for cubic and hexagonal crystal structures.

[Cyrille]

Tier0 has some localized 3s and 3p.
But it only includes 4s and no 4p.
I do not know if it is the origin of my problem but the strongly dispersive "4s" parabolic band, crossing the 3d has a minimum (below -10eV) which seems to me too low. It is typically 2 eV below  other DFT results..

Of course it would much preferable not to have the costly 3s and 3p but having 4s and 4p...
However the FHI pseudo/basis give very bad results for Cr.

Cyrille

[Anders Blom]

Note that the HGH potentials are very hard, check your mesh cut-off also, you may need 200-300.
New PPs and basis sets are coming in the 2016 version!

Yes Cr is bad in our FHI set, it's been known for long

[Cyrille]

I checked OMX and it seems much better.... but also quite hard pseudo
Waiting impatiently for new pseudo release:-)

Thanks for your help
Cyrille