Explicit band structure vs interpolation

Concerns issues with computing quasiparticle corrections to the DFT eigenvalues - i.e., the self-energy within the GW approximation (-g n), or considering the Hartree-Fock exchange only (-x)

Moderators: Davide Sangalli, andrea.ferretti, myrta gruning, andrea marini, Daniele Varsano

Post Reply
TreeEggs
Posts: 4
Joined: Sat Nov 19, 2016 5:15 pm

Explicit band structure vs interpolation

Post by TreeEggs » Mon Mar 07, 2022 12:59 pm

Hi all,

probably this has been discussed before, although I could not find anything in the search, so I apologize in advance.

I need to determine the band structure of a thin film system, and have done the usual steps (QE scf, then nscf calculations at the special points, GW at the special points). As I understand, I am supposed to generate the band structure along high symmetry lines with ypp. So my first check was to compare the interpolated DFT band structure to the explicit one obtained with QE. The agreement is very bad though and changing some parameters in the ypp input file did not really help.

Is there an idea of what the issue could be? My guess is that the number of special k-points is too small to start from?

If the interpolation remains problematic: is there also a way to evaluate the QP corrections along the band structure path explicitly?

Thanks,
Bjoern
Bjoern Baumeier
TU Eindhoven

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Explicit band structure vs interpolation

Post by Daniele Varsano » Mon Mar 07, 2022 1:25 pm

Dear Bjoern,
Is there an idea of what the issue could be? My guess is that the number of special k-points is too small to start from?
Yes, this could be an option, if your k point grid is too coarse this can bring inaccurate results.

If the interpolation remains problematic: is there also a way to evaluate the QP corrections along the band structure path explicitly?
Unfortunately not, as Yambo needs regular grids so explicit calculations along a path are prohibitive.

Another way to perform interpolations (also considering GW energies) that should be more accurate is to go through Wannier90 interpolation.
Here you can find an example of how to interface Wannier90 and Yambo:
https://github.com/wannier-developers/w ... /example23
https://github.com/wannier-developers/w ... c/tutorial

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

TreeEggs
Posts: 4
Joined: Sat Nov 19, 2016 5:15 pm

Re: Explicit band structure vs interpolation

Post by TreeEggs » Mon Mar 07, 2022 3:35 pm

Dear Daniele,

thanks for the quick reply. Indeed, using more kpoints seems to improve the interpolation. Thanks also for the heads up regarding the Wannier Interpolation, will check it out.

I still have one question regarding the ypp input. I thought

Code: Select all

BANDS_steps= 30                  # Number of divisions
would make 30 steps in each segment of the kpath. However, when I use this with:

Code: Select all

%BANDS_kpts                      # K points of the bands circuit
0.0 | 0.0 | 0.5|
0.0 | 0.0 | 0.0|
I end up with fewer points

Code: Select all

#     |k| (a.u.)         b100               b101               b102               b103               b104               b105               b106               b107               b108               b109              k_x (rlu)          k_y (rlu)          k_z (rlu)
#
       0.00000000        -1.14401233       -0.628518105       -0.438549034E-1     0.721207082         1.30482411         2.10452604         2.34852815         2.47630310         2.91414833         2.98115277         0.00000000         0.00000000        0.500000000
      0.979050528E-2     -1.14392805       -0.628578305       -0.437178910E-1     0.721195817         1.30492163         2.10455275         2.34873843         2.47617722         2.91426229         2.98114419         0.00000000         0.00000000        0.428571433
      0.195810106E-1     -1.14374435       -0.628702343       -0.433938354E-1     0.721208334         1.30514252         2.10464907         2.34920573         2.47595191         2.91452813         2.98106289         0.00000000         0.00000000        0.357142895
      0.293715205E-1     -1.14358652       -0.628789306       -0.430470817E-1     0.721319377         1.30535519         2.10483456         2.34963083         2.47589064         2.91480446         2.98082757         0.00000000         0.00000000        0.285714298
      0.391620249E-1     -1.14352655       -0.628784657       -0.427870080E-1     0.721551120         1.30547845         2.10509872         2.34983492         2.47611928         2.91499925         2.98043609         0.00000000         0.00000000        0.214285702
      0.489525311E-1     -1.14357328       -0.628688574       -0.426532105E-1     0.721867859         1.30549359         2.10540056         2.34978604         2.47660327         2.91508222         2.97996402         0.00000000         0.00000000        0.142857149
      0.587430410E-1     -1.14367497       -0.628557324       -0.426261649E-1     0.722167313         1.30543911         2.10565948         2.34959292         2.47713757         2.91507792         2.97955251         0.00000000         0.00000000        0.714285374E-1
      0.685335398E-1     -1.14372849       -0.628494203       -0.426325686E-1     0.722294509         1.30540347         2.10576463         2.34948444         2.47737813         2.91506124         2.97938418         0.00000000         0.00000000         0.00000000
Am I interpreting the option wrong?

Cheers,
Bjoern
Bjoern Baumeier
TU Eindhoven

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Explicit band structure vs interpolation

Post by Daniele Varsano » Mon Mar 07, 2022 7:10 pm

Dear Bjoern,
I still have one question regarding the ypp input. I thought
BANDS_steps= 30 # Number of divisions
would make 30 steps in each segment of the kpath.
Me too, let us check what's the reason of that.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

Post Reply