Orthonormality of QP-wave functions with PAW

[loefaas]

I am trying to do a band structure calculation using PAW+GW+MLWF, but have run into some troubles. The PAW+GW calculation seems to go ok, and gives ok results for the gap at Gamma etc. But when I am trying to call the wannier plugin I always get (have tried several different methods and parameters for GW)

Code: Select all

mlwfovlp_qp.F90:361:ERROR
   orthonormality error for quasiparticle wave functions.
  spin=   1  k point=     1  ortho_err= 6.3E-01 >1E-6
  Action : Be sure input nband>=maxval(bndgw)

 leave_new : decision taken to exit ...

I can't figure out what is going wrong, are the QP-WF when using the PAW-formalism not orthonormal?

Best Regards
Henrik

[gmatteo]

The orthonormality test is not correct in the case of PAW calculations since the onsite contribution is
missing. For the time being one has to skip the test on the orthonormality in 68_gw/mlwovlp_qp.F90
when PAW is used:

! if (ortho_err>tol6) then OLD VERSION
if (ortho_err>tol6.and. Dtset%usepaw==0) then NEW VERSION
write(msg, '(3a,i4,a,i6,a,1p,e8.1,3a)' )&
& ' orthonormality error for quasiparticle wave functions.',ch10,&
& ' spin=',isppol,' k point=',ikpt,' ortho_err=',ortho_err,' >1E-6',ch10,&
& ' Action : Be sure input nband>=maxval(bndgw)'
MSG_ERROR(msg)
end if

This minor bug will be solved in v642.
You might look at the enclosed patch file to understand how to change the routine so
that the PAW scalar product is calculated correctly.

Matteo

[gmatteo]

Here is the patch file.

[loefaas]

Thank you! Now it works.

[loefaas]

Hi again!

I am still in doubt, about the functionality of paw+GW+wannier.

I am now running with the new 6.4.2 version and have similar problems, or at least what I think. I am using the files for t03 in the wannier folder.

When just running the test it works fine and give a good result. But when I just change the psuedo to a Si-PAW the calculation crashes just before the calculation of the overlaps with a segmentation fault. I also tried running with 6.2.2 with the above suggested correction to skip the orthonormality test of the QPS, this run completes but gives just rubbish for the bandstructure. This seems to be a common feature for all my tries with PAW+GW+wannier, there are a lot of fluctuations in the bands from the wannier interpolation, and also when the wannier functions are visualized the symmetry of the lattice is for sure broken.

I don't know if I have something completely wrong in my inputs, or if my interpretation is wrong or if it is some problems with the implemetation? Any suggestions?

Best Regards
Henrik Löfås

[igor.favorskiy]

Hello

I have the same problem while running a GW + PAW + Wannier90 band structure computation for Si
(abinit-6.6.1/tests/Psps_for_tests/Si.GGA_X_PBE_SOL+GGA_C_PBE_SOL-paw.abinit)
on abinit 6.6.1:
segmentation fault at the w90 dataset
Currently, I haven't found any solution to this problem neither.
Thanks in advance for your help.

Best regards
Igor Favorskiy