nbnd in PWscf calculations

[8813204602]

Dear developers
I want to calculate the band structure for a semiconductor material with a 2eV gap. When perform nscf calculation to make a Yambo input file,using nbnd=20, the output is correct and show a semiconductor material with 15 occupied bands. But when increase nbnd, for example nbnd=400, the output shows a half-metall material,that means 16 and 17 bands are partially occupied. Why does this happen by changing the value of nbnd?

Best regards,
Mitra Helmi

[andrea.ferretti]

Dear Mitra,

as you mention, such a behaviour is not expected (the only reason I can come up with is e.g. a problem in the nscf diagonalization with 400 bands).
In order to go any further, it would be useful if you could send scf+nscf input and output files showing the problem such that we can try to reproduce it.

take care
Andrea

[8813204602]

Dear Andrea

scf and nscf input and output files are attached.

There is no problem up to nbnd=75, but from nbnd=80,partial occupation of 16 and 17 bands is seen in the nscf output and value of Fermi level also changes.

Thanks a lot
Mitra

[andrea.ferretti]

Dear Mitra,

I had a look at your inputs in order to reproduce the problem. Here are some comments

• since you are using norm-conserving pseudo potentials, ecutrho should be exactly 4 times ecutwfc (or simply
  unset, leading to the default that is 4*), while you are setting 8 (no reason to do that, but calculations are slower).
• force_symmorphic=.true. is only needed in the last pw.x calculation before p2y (say, the nscf); no need to set it
  at the scf level.
• when using smearing="mp" I get spurious (and negative) occupations for bands 16,17 already at the scf level;
• using smearing="gaussian" solves the problem in my case
• using smearing for yambo calculations may be somehow problematic (eg the smearing shape in yambo is only Fermi-Diract, etc);
  Ideally you want to use a small value in the scf calculation in order to be able to use starting_magnetization and get to the magnetic solution you are looking for;
  Then I would run the nscf calculation without any smearing (after all, you have a semiconductor)
• all tests were done using qe-6.7, though this is probably not causing any problem

take care
Andrea

[8813204602]

Dear Andrea

Thank you very much for your useful guidances.

Best regards,
Mitra