Inquiry Regarding Use of GW Output for Optical Property Calculation in Yambo

[y.mostafa]

Dear Yambo Developers

I have effectively determined the band gap of certain silicon crystal structures. Presently, my objective is to explore the optical properties of these structures utilizing Yambo through the Bethe-Salpeter method.

I have a query: "Is it possible to use the outputs of the GW band structures to reduce the computation time and avoid Yambo from recomputing the quasiparticle approximate distribution from the scratch?"

Could you please guide me on whether Yambo offers a specific method to utilize the existing GW output for this purpose? Your insights into this matter would be highly valuable, as it holds the potential to significantly enhance the efficiency of my research.

Thank you very much for your time and consideration.

Best regards,

[Daniele Varsano]

Dear Mostafa,

once you have calculated the GW band structure, you can use the QP energies to plug them in the BSE step.
This can be done in two ways:
1) reading the QP database
2) approximate it with a scissor/stretching approximation.

You can find here how to do that:
https://www.yambo-code.eu/wiki/index.ph ... nalization
https://www.yambo-code.eu/wiki/index.ph ... e_energies

the scissor/stretching parameter can be calculated as shown here (step 3):
https://www.yambo-code.eu/wiki/index.ph ... _Rome_2023

Note, you can also use the screening calculated in GW (ndb.pp) avoiding to recalculate the static screening (use -X p instead of -X s when generating the BSE input).

Best,
Daniele

[y.mostafa]

Dear Daniele,

Thank you for your reply and the valuable information you provided. I followed the steps mentioned in your solution for the silicon structure. However, despite implementing each variation, I encountered a discrepancy in the imaginary part of the dielectric that did not match the experimental results. The imaginary part of the dielectric is illustrated in the attached image, and the input files are also included.
I am asking for your guidance to deal with this issue. Your help in solving this issue would be greatly appreciated.

Thank you very much.

Best regards,

[Daniele Varsano]

Dear Mostafa,

In general, discrepancies with experiments happens for many reasons: lack of convergences, approximations, experimental conditions etc...
I suggest you to go step by step, comparing your calculation with the existing literatures (e.g. calculations done at the same level of theory), comparing first the GW gap and next the optical absorption.
When posting, it is always useful to attach also the report file to understand all the step of the calculation.
Looking at your input, please note that there is no need to include deep bands in the BSE (BSEBands), usually few bands across the gap are enough to get converged results. Including many bands results in very large excitonic matrix, making the calculation cumbersome, more than needed.
When checking convergence, is very important to check convergences with respect the k points grid.

Best,
Daniele

[y.mostafa]

Dear Daniele,

I appreciate your prompt response to my inquiries.
Currently, I am trying to replicate the findings presented in the following article on page 18.
https://arxiv.org/pdf/0810.3118.pdf
Specifically, I am focusing on the atomic structure of diamond silicon, and I have attached the Quantum Espresso input file corresponding to this structure for your reference.

In alignment with the procedures outlined in the aforementioned article, my objective is to solve the BSE equation in order to reproduce figure 8. I initially attempted this with 4x4x4 k-points, but the obtained results did not align with the expected outcome.

In a subsequent attempt, I increased the k-points to 14x14x14; however, I encountered a persistent error that I am unable to resolve. I would greatly appreciate your assistance in resolving this issue.

Code: Select all

P1-lrdn0530.leonardo.local: [ERROR] STOP signal received while in[06.03] Diago Solver @q1
P1-lrdn0530.leonardo.local: [ERROR] LINEAR ALGEBRA driver [PARALLEL_HERMITIAN_diagonalization]performing P(Z/C)HEEV

Report files are also attached.

Thank you for your continued support.

Best regards,
Mostafa Yaghoobi

[Daniele Varsano]

Dear Mostafa,

as mentioned in the previous post, try to reduce the number of bands in the BSE: BSEBands.
You included deep states that are not needed, resulting in a very large excitonic matrix.
Few bands across the gap are enough, e.g.

Code: Select all

% BSEBands                                      
   2 | 7 |                           # [BSK] Bands range
%

Next, please note that in input you indicated (0,20). Yambo is a Fortran and index 0 should not be used, as counting starts from 1.
Actually, report files seem to come from different inputs with respect to the ones you posted.

Best,
Daniele

[y.mostafa]

Dear Daniele

Thanks for your quick feedback and guidance, I appreciate your help in optimizing the BSE calculation.
I had sent the wrong files.
But according to your advice, my problem was solved.
Thank you very much.

Mostafa yaghoobi