ABINIT Discussion Forums

Hello.


I'm working on low-dimensional metallic system and I want to calculate my system's band structure with GW method.


For the first test, I referred to the input file in tutorial gw 1 ( : tgw1_9.in ) , which is for fcc Si, and just added lines like

=================================================
ndtset 5

# Dataset 5: Band structure
iscf5 -2
getden5 -1
kptopt5 -1
nband5 8
ndivk5 50 # 50 points for one segment.
# by 4 points.
kptbounds5 0.0 0.0 0.0 # Gamma point
0.0 0.0 0.5 # Z point
tolwfr5 1.0d-18
enunit5 1 # Will output the eigenenergies in eV

tsmear 0.01
occopt 3


===================================================


( Surely atomic and crystal infos are corrected. )



I wonder that is it OK for getting band structure with GW correction and whether there are some cautions for calculating bandstructure of transition metal low-dimensional systems with GW method by ABINIT.

Thank you for reading this post and your reply.

Hi,

The simple answer is, no, you can not compute the GW corrections on an arbitrary kpt path. You have to use a regular grid.
Then, you can use some interpolation scheme (e.g. wannier functions) to obtain the full GW bandstructure.

Thank you for your reply.

I've used the k mesh with no k-point shift ( nshiftk 1 / shiftk 0.0 0.0 0.0 ), and nkgpt 2 2 10.

Actually I cannot fully understand your reply because I'm not familiar with GW calculation by ABINIT. Could you please note me some materials related to the GW band structure calculation by ABINIT, or give me some sample input files..?

So far, in my understanding, I should fix the kptgw parameter in optdriver=4 dataset, and with the self-energy at certain k-point specified by kptgw, I should interpolate them to get the whole eigenvalues along my k-path ( Gamma to Z ). Is it right...?

Yes, your k-point grid with no shift is fine. Since you only need the bandstructure along Gamma-Z, and this is along one of your primitive reciprocal-space vector,
you can simply use ngkpt 2 2 X, and increase X until you have enough points. You have to enumerate all k-points at which you want to compute corrections in kptgw.

Then, if you want to interpolate the corrected eigenvalues on a finer kpt grid, the most reliable method is to make a wannier interpolation. Unfortunately, I don't think there is any example available and I am not familiar with this myself. But if you need to interpolate only on a single kpt line, you could just do a one dimensional spline with any post-processing tool (python, mathematica, matlab...)

Thank you for your reply.

I had finished the convergence test for # of k-points and ecut with LDA exchange-correlation functional. In my case, about 30 k-points along Gamma-to-Z path was enough for both total energy convergence and Fermi level estimation.

Thank you again for your kind reply.

Hello,

I dont know what your system is but in the SCF calculation, you should converge your k point grid in the 3 directions, not only along gamma-Z.

David