Band convergence in the d/dk calculation with occopt=7

Phonons, DFPT, electron-phonon, electric-field response, mechanical response…

Moderators: mverstra, joaocarloscabreu

Locked
Boris
Posts: 128
Joined: Tue Feb 16, 2010 10:13 am
Location: France

Band convergence in the d/dk calculation with occopt=7

Post by Boris » Wed Mar 09, 2011 3:06 am

Dear all,

I'm performing several calculations on phonon frequencies in CdS. Following what Matthieu said in this post, I used occopt= 7 with a low tsmear (0.005 Ha) in order to improve the convergence in the d/dk calculation.

This indeed improved the convergence but triggered another problem: my last two bands are not converged, whatever the number of bands in my calculations (I tried nband=18, 40 and 45). This can be seen on the "res" block in the ouput file:

Code: Select all

 Mean square residual over all n,k,spin=  -4.4444E-03; max=  9.9832E-21
   0.0000  0.0000  0.0833    1  8.82866E-21 kpt; spin; max resid(k); each band:
  1.51E-21 8.18E-21 4.06E-21 5.17E-21 5.17E-21 4.58E-21 4.58E-21 1.50E-21
  1.50E-21 3.00E-21 3.00E-21 1.99E-21 7.97E-21 8.78E-21 8.78E-21 5.62E-21
  2.97E-21 2.97E-21 5.66E-21 1.89E-21 7.40E-21 7.40E-21 4.03E-21 1.62E-21
  3.26E-21 3.26E-21 3.83E-21 3.83E-21 3.95E-21 6.61E-21 8.08E-21 8.08E-21
  1.98E-21 3.94E-21 6.46E-21 6.46E-21 8.51E-21 5.61E-21 5.61E-21 5.73E-21
  5.73E-21 8.83E-21 3.64E-21-1.00E-01-1.00E-01


You can see that the last two bands are not converged, at all.

When I use 18 bands (only filled bands), with occopt=1, the convergence is a bit slower but:

Code: Select all

 Mean square residual over all n,k,spin=   5.2588E-21; max=  9.9991E-21
   0.0000  0.0000  0.0833    1  9.51723E-21 kpt; spin; max resid(k); each band:
  3.34E-21 6.98E-21 3.24E-21 8.02E-21 7.93E-21 2.40E-21 8.06E-21 7.28E-21
  4.27E-21 3.24E-21 3.57E-21 2.93E-21 8.55E-21 3.72E-21 9.46E-21 9.52E-21
  4.04E-21 6.50E-21


Now all the bands are converged.

Is that a problem?

Thank you for your anwers

Boris
----------------------------------------------------------
Boris Dorado
Atomic Energy Commission
France
----------------------------------------------------------

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Band convergence in the d/dk calculation with occopt=7

Post by mverstra » Wed May 25, 2011 5:34 pm

Hi Boris,

no - looks ok, the last bands in the first calculation are probably degenerate with further bands you do not take into account. Remove them and the lower bands converge normally. Add a few bands more and they should be able to converge normally as well (you need the full degenerate subspace).

Another possibility is neglecting the last bands in the convergence criterion (nbdbuf)

Matthieu
Matthieu Verstraete
University of Liege, Belgium

Locked