The negative absorption peak obtained by BSE calculate

[WEIW]

Dear all,

When I use the scissor operator (and consider GW correction) to calculate bse, we both obtain a negative absorption peak. Can you explain why this occurs?
Note that the system I am investigating experience a band inversion after introducing strain. Prior to the band inversion the optical absorption is above zero across the all range of energy.
Therefore, the band inversion will affect the optical absorption? is it true?

Thanks in advance.
Best,
WeiW

[Daniele Varsano]

Dear LIU,

it seems you have a negative excitation energy and you see negative peaks at positive energies because of the time ordering of the response function. We would need more information (input/report files that can be attached) to investigate the source of the problem.

Best,
Daniele

[WEIW]

Dear developers,

Thanks for your quick reply. I have attach the report file.

Best,
Liu

[Daniele Varsano]

Dear Liu,

your system has a very narrow gap (0.28 eV) and you are adding a very small (unusual) QP correction (0.01 eV). In this situation
it seems the binding energy is larger than the gap and you face an instability (negative excitation energy). Please check if such QP correction is meaningful. Next, check the usual convergences (K points etc.).

Out of the problem: I suggest you to update to a more recent release of the code and use the "slab z" cutoff instead of the "box" which is rather obsolete.

Best,
Daniele

[WEIW]

Dear Daniele,

Thanks for your explanation!

Thank you for your suggestions on our calculations. We will try to correct this later. Allow me to explain why I added a very small (unusual) QP correction. This choice was made in order to verify whether the negative absorption peak could be attributed to the QP correction. We find that larger "scissor operators" can indeed prevent negative absorption peaks. However, despite considering the true GW correction, our calculation still yields a negative absorption peak, which is quite perplexing.

One more thing I want to understand is that why there is a discrepancy between the exciton energy distribution (o-2D.exc_qpt1_E_sorted) and the peak value of the absorption spectrum (o-qp5.alpha_q1_diago_bse) that I obtained. For instance, while the light absorption spectrum displays a negative absorption peak at 0.629eV, it does not appear in o-2D.exc_qpt1_E_sorten. Why does this occur?

The related datas are attached.

Best,
Liu

[Daniele Varsano]

Dear Liu,

I cannot see discrepancy between the spectrum and the obtained eigenvalue (see plot attached). Please, note that negative peaks at positive energy correspond to negative excitation energies because of the time ordering of the response function.

For instance, while the light absorption spectrum displays a negative absorption peak at 0.629eV, it does not appear in o-2D.exc_qpt1_E_sorten

Indeed, you have negative eigenvalues at -0.633 with substantial oscillator strength.

However, despite considering the true GW correction, our calculation still yields a negative absorption peak, which is quite perplexing.

Check your GW and BSE convergence, it is also possible that you have an instability pointing to an excitonic insulator scenario (there is plenty of literature in that), but I do not know what's your system. It is also possible that your DFT ground state (Vxc functional) is not adequate.

Best,

Daniele

[WEIW]

Dear Daniele,

Recently I encountered some new problems when I calculate the absorption.

1)I used "BSEmod=resonant" when I calculate the Bethe-Salpeter kernel and bse, but the output file of bse computation showed that I did not read successfully and I still used "BSEmod=retarted". What is the reason for this? How to solve this problem?

2)When I used "BSEmod=coupling" in calculations, this step of the bse calculation always generates an error. How can I solve this problem?

3)The absorption spectrum I calculated has a small peak near "energy=0" , but it didn't show in the exciton calculation.

I have attached the necessary output and input files for the calculation I did. A part of file "o-2D.exc_qpt1_E_sorted" was deleted, and the useful part was kept.

Looking for your kindly reply!

Thanks in advance.
Best,
Liu

[Daniele Varsano]

Dear Liu,

1) The reason is shown in the report file:

Code: Select all

 [WARNING]Fractional e/h occupations. Retarded BSEmode forced.

2) You have a rather large matrix, do you have any error message in the log files or in submission output? Is it possible that you run out of wall time as direct diagonalization of such matrix can take long time?

3) Actually, it seems that the sorted exciton output is not the same of the spectrum. Please check that the ypp was run on top of the right directory containing the ndb.BSSdiago file that produced the spectrum. I could not compare the kernel parameters as it was deleted in the exc files.

Best,
Daniele