optical absorption peak of GW+BSE is on lower energy than that of IP and rim_cut band structure

[AmberLee123]

Hello,

I am facing a quite weird problem in BSE optical absorption calculation with rim_cut in 2D system.

In tutorials, the first peak of optical absorpiton locates at different energy based on calculation methods. Typically, IP(Independent Particle) <= GW+BSE <= QP(Quasiparticle, or GW only). However I got a GW+BSE abs. with first peak locating at energy smaller than IP band gap. Specifically, IP bandgap(E_IP) = 1.797697, QP scissor = 3.591969, ie. GW bandgap(E_GW) = 5.389666, the first peak of GW_BSE ~= 1.3, even smaller than E_IP. Do yambo also output optical bandgap(E_opt), or exciton binding energy(E_b)? As It should be E_opt + E_b = E_GW, probably we can figure out what the problem is.

The corresponding GW and BSE calculation file is attached in report.tar_.gz (five splitted files, whole file can be downloaded here, link: https://pan.baidu.com/s/17k6XewCCsNvTYMhJcjchmw PW: 5vlo ), and the input parameters were converged without rim_cut.

Also, the process of plotting band structre of 2D materials with rim_cut is the same as that of 3D in tutorial? As the band gap calculated by rim_cut is slightly different, the band structure and the optical absorption should be coherent in my opinion.

Regards,

[AmberLee123]

another two splitted files.

[Daniele Varsano]

Dear Guoyu Yang,

unfortunately I was not able to open the zipped files you post, beside the report_aa (it seems they are corrupted).
Also the link you provided is in chinese and it seems it requires some password.

Do yambo also output optical bandgap(E_opt), or exciton binding energy(E_b)?

You can look at it by looking at the first bright excitation in o.exc_qpt1_E_sorted, and the bidinng energy can be extracted from the knowledge of the band gap.
Can you post just the input/report of the BSE calculation you are referring at?
For instance from the ouptut contained in report_aa I can see you applied a scissor to the calculation of the screening ( XfnQP_E) and not to the BSE kernel KfnQP_E. Also note that including a scissor in X it will reduce the screening increaseing the binding energy.

Best,
Daniele

[AmberLee123]

Dear Daniele Varsano,

1. The password is 5vlo

2. I redo the calculation with
| % KfnQP_E
| 3.591969 | 1.000000 | 1.000000 | # [EXTQP BSK BSS] E parameters (c/v) eV|adim|adim
| %
and
| % XfnQP_E
| 3.591969 | 1.000000 | 1.000000 | # [EXTQP Xd] E parameters (c/v) eV|adim|adim
| %

The problem remains.

The files are attached as report_02.tar.gz.

Best,

[AmberLee123]

The input, o.qp and report file of G0W0 calculation.

[Daniele Varsano]

Dear Guoyu,

I had a look at your report/output here are some considerations:

1) As I wrote in the previous post XfnQP_E should not be used in the BSE calculation, anyway it did not take effect as you are reading the screening of a previous calculation
2) It seems you are dealing with a 2D system, the vacuum you are using to isolate the system seems a bit low, I would add vacuum in your supercell
3) In the BSE matrix it seems to me you are using a quite low value for BSENGexx and BSENGBlk, in particular BSENGexx, and you should verify it.
4) Check the convergence wrt k points

Having said that, if I'm not wrong your system has very flat bands, this means that it is possible you have an extremely localized exciton with enormous biding energy.
As you said, usually typically, IP(Independent Particle) <= GW+BSE but this is not always true when you have a very large electron-hole interaction.

Best,
Daniele

[AmberLee123]

Dear Daniele Varsano,

Many thanks for your dedication. Here are some related results I have at hand. Would you please give me some more comments before I continue to the tests in your suggestion?

1) As I wrote in the previous post XfnQP_E should not be used in the BSE calculation, anyway it did not take effect as you are reading the screening of a previous calculation

Now I understand, the parameters should be like:

| % KfnQP_E
| 3.591969 | 1.000000 | 1.000000 | # [EXTQP BSK BSS] E parameters (c/v) eV|adim|adim
| %
and
| % XfnQP_E
| 0.0 | 1.000000 | 1.000000 | # [EXTQP Xd] E parameters (c/v) eV|adim|adim
| %

2) It seems you are dealing with a 2D system, the vacuum you are using to isolate the system seems a bit low, I would add vacuum in your supercell

Yes, I am dealing with a 2D system. I will do the test of BSE absorption wrt vacuum thickness.

3) In the BSE matrix it seems to me you are using a quite low value for BSENGexx and BSENGBlk, in particular BSENGexx, and you should verify it.

I make the convergence of BSENGexx(

03010301_Conv_BSENGexx_1.png

03010301_Conv_BSENGexx_1.png) and BSENGBlk(

03010203_Conv_BSENGBlk_1.png

03010203_Conv_BSENGBlk_1.png) on kpoints 6x6x1. And use the parameters in kpoints 9x9x1, 12x12x1 and so on... Do I need to converge them for larger kpoints mesh... http://www.yambo-code.org/wiki/index.ph ... parameters

4) Check the convergence wrt k points

The kpoints converged in 12x12x1.

03020108_alpha_3.png

03020108_alpha_5.pdf It seem that happens to all k-points parameters.

Having said that, if I'm not wrong your system has very flat bands, this means that it is possible you have an extremely localized exciton with enormous biding energy.

Yes, the band is flat... Also, is it possible to calculate band of 2D systems with rim cut... I found only the one for 3D hBN in tutorial.

As you said, usually typically, IP(Independent Particle) <= GW+BSE but this is not always true when you have a very large electron-hole interaction.

I found a paper showing the situation J.Mater. Chem. C,2019, 7, 14284 (https://arxiv.org/pdf/1905.10569.pdf). The procedure there is to manually shift the band with Delta E = bandgap(HSE) - bandgap(PBE)

The results obtained from the PBE functional
strongly require the shifting of peaks manually in order to
match the data from HSE06, as shown in Fig. 4.

I just want to verify it means parameters are:

in G0W0 calculation with rim cut,
| % XfnQP_E
| Delta E | 1.000000 | 1.000000 | # [EXTQP Xd] E parameters (c/v) eV|adim|adim
| %

and in BSE calculation with rimcut,
| % KfnQP_E
| 3.591969+Delta E | 1.000000 | 1.000000 | # [EXTQP BSK BSS] E parameters (c/v) eV|adim|adim
| %
and
| % XfnQP_E
| Delta E | 1.000000 | 1.000000 | # [EXTQP Xd] E parameters (c/v) eV|adim|adim
| %

Best,

[AmberLee123]

03020108_alpha_5.pdf

[AmberLee123]

flat band with kpoint 9x9x1 (the GW gap converges at kpoint 9x9x1) 02020102_plot_GW_band.pdf

[Daniele Varsano]

Dear Guoyu,

1)OK your calculation seems reasonably converged
2)

Also, is it possible to calculate band of 2D systems with rim cut... I found only the one for 3D hBN in tutorial.

rim_cut activate the treatment of the coulomb interaction, i.e. average of Coulomb integral and truncated potential. In 3D you want just average the coulomb integrals CUTGeo='none', in 2D you also activate the truncated potential.

3) You can recalculate the GW band structure using a scissor in the screening, but this is rather arbitrary, and also if PBE is not a good starting point it means that also the wfs will be very different from the HSE calculation.

you can think about doing an eigenvalue self-consistent calculation:
see:
http://www.yambo-code.org/wiki/index.ph ... alues_only

4) In BSE you will then use a scissor obtained from your GW calculation. It does not make much sense, to sum up a scissor for the starting point and a scissor calculated using PBE starting point.

Also, note that once you have calculated the screening you do not need to set the XfnQP_E anymore.

Best,

Daniele