BSE convergence

[Laura Caputo]

Ciao!

So from this input, I should convergence all the parameters (for screening and BSE kernel) one by one and check for each of them the spectra to see if they are converged. Right?

[Daniele Varsano]

Dear Laura,
yes essentially the bands in the screening and its dimension (the same dimension, or smaller need to be considered in BSENGBlk), and the other parameters listed here:

http://www.yambo-code.org/wiki/index.ph ... ter_kernel

Best,

Daniele

[Laura Caputo]

Ok thanks! One more thing, the parameter
% BLongDir
1.000000 | 0.000000 | 0.000000 | # [BSS] [cc] Electric Field
%

should still be set as
1.0000 1.00000 0.000000

for 2D materials?

[Daniele Varsano]

Dear Laura,
this is the polarization of the light, if the system has anisotropy you may want to look the absorption in one direction (100) and in the other one (010).
Best,

Daniele

[Laura Caputo]

Hello,

I have done some BSE calculations and I have the k-point convergence in attachment. I think that the 12x12 the sharpest peak is converged (in position and shape). Is it correct?
Then, I sorted the electrons, having:
# Maximum Residual Value = .66719E+00
#
# E [ev] Strength Index
#
1.25849969 0.739326715E-6 1.00000000
1.30254217 1.00000000 2.00000000
1.30455674 0.444401161 3.00000000
1.32095464 0.111625433 4.00000000
1.47684021 0.610957951E-3 5.00000000
1.47878403 0.108990515 6.00000000

Thus I've continued studying the brightest exciton, which should be doubly degenerate (2nd and 3rd).
Then, I have studied the amplitude of the 2nd and 3rd exciton listed, having in both cases:
# Band_V Band_C Kv-q ibz Symm_kv Kc ibz Symm_kc Weight Energy
#
36.0000000 37.0000000 1.00000000 1.00000000 1.00000000 1.00000000 0.306682619 1.54513061
36.0000000 38.0000000 1.00000000 1.00000000 1.00000000 1.00000000 0.185729111 1.55102908
35.0000000 37.0000000 1.00000000 1.00000000 1.00000000 1.00000000 0.185607914 1.55089469
35.0000000 38.0000000 1.00000000 1.00000000 1.00000000 1.00000000 0.694424796E-1 1.55679316
36.0000000 37.0000000 2.00000000 2.00000000 2.00000000 2.00000000 0.641617854E-1 1.66334865
36.0000000 37.0000000 2.00000000 1.00000000 2.00000000 1.00000000 0.641358864E-1 1.66334865
36.0000000 37.0000000 9.00000000 2.00000000 9.00000000 2.00000000 0.634332189E-1 1.66363825
36.0000000 37.0000000 9.00000000 1.00000000 9.00000000 1.00000000 0.634120914E-1 1.66363825
35.0000000 38.0000000 23.0000000 1.00000000 23.0000000 1.00000000 0.550116558E-1 1.68587457
35.0000000 38.0000000 23.0000000 2.00000000 23.0000000 2.00000000 0.550109311E-1 1.68587457


for which I conclude that the main contribution comes from Gamma (kpoint 1), where the weights are the largest.
Lastly, I wanted to plot the amplitude and I used something similar to what suggested in the tutorial:
$ paste o-3D_BSE.exc_qpt1_amplitude_at_2 o-3D_BSE.exc_qpt1_amplitude_at_3 > o-3D_BSE.exc_qpt1_amplitude_at_2_3
gnuplot> p 'o-3D_BSE.exc_qpt1_amplitude_at_3_4' u 1:($2+$4)/2 w l t 'Bright exciton'

however, I obtain the spectra in the attachment. I don't find any relation between these and the peaks in the absorption spectra: am I missing something?

Lastly, I wanted also to do spatial plotting and I've read that the best position for the hole is where the valence electron contributing to the exciton is. How can I assess this?

Thanks,

[Daniele Varsano]

Dear Laura,

the amplitude does not provide you the absorption spectrum but the independent particle transitions forming the exciton:
Amplitude_i(w)=\sum A^icv delta(w-Ec-Ev)
the delta then is smoothened as Lorentzian, you have correctly the peaks at 1.55 and 1.66 eV as indicated in the weight file.

Code: Select all

Lastly, I wanted also to do spatial plotting and I've read that the best position for the hole is where the valence electron contributing to the exciton is. How can I assess this?

The suggestion, if possible is to place in or nearby a symmetric position being sure the valence charge is not negligible.
You can asses plotting the density of the relevant valence band contributing to the exciton.

Best,
Daniele

[Laura Caputo]

Thank you Daniele. Also, I've seen that there is also in this BSE case to use the BG termination which should allow using fewer bands. Is it safe to use for BSE calculations? Should one take into account other variables?
I've calculated the convergence of the bands, having very slow convergence considering valence/conduction bands. On the other hand, the gap is converged at 250 bands. Which one should I consider?

Lastly, in the calculation of GW gap the kpoints needed for convergence were quite more than 12x12 (as reported here in the spectra), on the tutorial I've read that it should be the contrary. Is it because I've just considered the sharpest peak?

[Daniele Varsano]

Dear Laura,

Is it safe to use for BSE calculations? Should one take into account other variables?

There is no band terminator for BSE, and it should be not present in the input, unless you have also the GW runlevel in the same input file.

I've calculated the convergence of the bands, having very slow convergence considering valence/conduction bands. On the other hand, the gap is converged at 250 bands. Which one should I consider?

These are different convergences, in GW you have a sum over states, in BSE you need to include bands across the gap in order to have enough transitions to converge the spectrum in the range of energy you are interested in. The valence and conduction bands are fewer than the ones used for GW calculation (all occupied and many empty bands).

Lastly, in the calculation of GW gap the kpoints needed for convergence were quite more than 12x12 (as reported here in the spectra), on the tutorial I've read that it should be the contrary. Is it because I've just considered the sharpest peak?

The plot you included is not a convergence test on kpoints but on the dimension of the screening matrix, at least looking at the labels. Usually, the kpoint grid needed to converge BSE spectra are larger than the one of GW, but this is system dependent.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

Thanks for the clarification. Yes, I'm doing also the GW runlevel and that's why maybe I have the terminator. Regarding the band convergence for the GW level, I guess then that it could be enough to converge the quasi particle gap, right?

For the k-point, I'm sorry I wasn't clear. I meant the convergence of the k-point in the spectra included a few replies ago.

[Daniele Varsano]

Dear Laura,

Regarding the band convergence for the GW level, I guess then that it could be enough to converge the quasi particle gap, right?

I'm not sure I've got your point, anyway if you are asking if for what concern BSE spectra it is enough to converge the QP gap instead of the single QP level, the answer is yes as you are interested in energy differences (transitions).
Best,
Daniele