q-point reduction

[tolsen]

Hi

I am a bit confused about the way yambo handles k-points and q-points. I have run an abinit calculation with a 2x2x2 Monkhorst-Pack k-point mesh,
which reduces to two k-points in the IBZ. When I then use a2y and initialize yambo the two irreducible k-points are read correctly, but the r_setup file says the k-point grid latttice is 4x4x4 with 32 k-points in the full Brillouin Zone? However, the weights of the two IBZ k-points are still given correctly by 0.25 and 0.75.

In the q-point part of r_setup, it also says that there is 32 q-points in the full Brillouin zone (and 6 in the IBZ), and the Gamma q-point has a weight of 1/32. I would expect that the q-point grid is identical to the k-point grid but shifted rigidly to the Gamma point and if the k-point grid is really 2x2x2 there should not be any irreducible q-points with weights larger than 1/8. Anyway, I do not understand the 32 grid points in both the q and k mesh. I have attached some part of the r_setup file below.

Hope to hear from you

BR
Thomas Olsen
Post Doc
Technical University of Denmark

[02.04] K-grid lattice
======================

Compatible Grid is 3D
B1 [rlu]= 0.250000 0.000000 -0.250000
B2 = 0.000000 0.250000 0.250000
B3 = 0.250000 -0.250000 0.000000
Grid dimensions : 4 4 4
K lattice UC volume [au]: 0.02870

[02.05] Energies [ev] & Occupations
===================================

Fermi Level [ev]: 5.514878
Electronic Temp. [ev K]: 0.000 0.000
Bosonic Temp. [ev K]: 0.000 0.000
States summary : Full Metallic Empty
0001-0004 0005-0050
Indirect Gaps [ev]: 2.189679 2.667022
Direct Gaps [ev]: 2.667022 3.636977
X BZ K-points : 32

Energy unit is electronVolt [eV]

*X* K [1] : 0.250000 0.250000 0.250000 (iku) * Comp.s 1239 * weight 0.250000
E -10.41180 -3.18791 0.00000 0.00000 2.66702 4.23645 4.23645 7.45231
E 7.94651 7.94651 9.49662 12.32970 15.30944 15.30944 18.40633 18.48032
E 18.48032 20.43920 24.59591 24.59591 26.81704 26.81704 28.16491 29.34568
E 30.83834 32.43739 32.43739 33.86847 33.86847 34.88784 34.88784 37.82172
E 38.72376 38.87284 38.87284 40.19839 40.58587 41.63432 41.63432 42.41716

*X* K [2] : 0.750000 -0.250000 -0.250000 (iku) * Comp.s 1239 * weight 0.750000
E -8.487826 -5.636255 -2.799987 -1.447297 2.189679 5.123933 6.387629 6.637626
E 10.80155 11.25749 12.87792 13.21874 13.51601 17.33978 18.39631 19.08792
E 19.96195 20.07125 20.42633 23.34865 24.23030 25.70242 25.91661 28.12370
E 29.54890 29.82261 29.91684 31.20166 32.50274 34.73485 35.34467 36.54069
E 37.04201 37.84974 39.19511 40.69464 42.50407 43.69879 44.95288 44.98957

[03] Transferred momenta grid
=============================


IBZ Q-points : 6
BZ Q-points : 32

Q [00001] : 0.000 0.000 0.000 (iku) * weight 0.03125
Q [00002] :-0.500000 0.000000 -0.500000 (iku) * weight 0.375000
Q [00003] :-0.500000 -0.500000 -0.500000 (iku) * weight 0.125000
Q [00004] : 0.000000 -0.500000 0.000000 (iku) * weight 0.187500
Q [00005] :-1.000000 -0.500000 0.000000 (iku) * weight 0.187500
Q [00006] :-1.000000 0.000000 0.000000 (iku) * weight 0.09375

:: Indices: polarization function

[tolsen]

Ups.
Perhaps you would also like to know that the system is bulk Si, using the minimal unit cell.

BR
Thomas

[claudio]

Dear Thomas

did you remove the non-symmorphic symmetries putting the flag
symmorphi 0

in you abinit input?

Claudio

[tolsen]

Hi Claudio

I do not include non-symmorphic symmetries. Anyway, I believe this only effects the number of irreducible k-points, and I am mostly puzzled about the full Brillouin zone k-point mesh.

BR
Thomas

[claudio]

Dear Thomas

I don't know how you produce that error in Yambo,
I put in attachment my input file for Silicon with the 2x2x2 kpoint grid and the r_setup file produce by Yambo.
As you can see in the report yambo says:

IBZ Q-points : 3
BZ Q-points : 8

let me know if it is ok

Ciao
cla

[tolsen]

Dear Claudio

Thank you for the effort. I should have mentioned earlier that I obtain the "right" result if I use a Gamma centered k-point mesh, which is also what is used in the example you have attached (I think) due to the (shiftk 0 0 0) keyword. If this keyword is commented out I think you will get the standard 2x2x2 Monkhorst-Pack sampling which reduces to two k-points (and yambo then states 32 k(q)-points in the Brillouin zone).

BR
Thomas

[claudio]

Dear Thomas

I reproduced the bug,
I have no time to fix it now, so I advise you to use non-shift grid for the moment. (see this post)
I will make you know when it will ok.

Claudio

[tolsen]

Dear Claudio

Thanks a lot. Just so you know, I usually use gamma centered grids and have no urgent need to use shifted grids. The present case was merely a test to understand how the q-point reduction works.

BR
Thomas

[myrta gruning]

Hallo Thomas

We looked at the matter and we concluded that Yambo is working fine and you can use shifted grid.

What Yambo does it reads the k points in the IBZ from Abinit and using the symmetries it expands the points in the full BZ. Then it checks that the grid in the BZ is uniform (that is it forms a cubic grid). It could be that in difficult cases (e.g. surfaces) Yambo does not recognize as uniform a uniform grid, but the contrary is not happening, it always detects if a grid is not uniform.

About the number of k-points in the full BZ that you can expect when choosing a MP LxMxN grid it depends on the crystal structure, but it is not simply the product LMN except in particular cases. The same it is true for the q-point grid.

What can be confusing is that yambo report that the grid is 4x4x4. Note that this is not referring to the MP grid, but to an equivalent grid that Yambo uses internally for generating the q-points..

m