NaN error when computing electron self-energy

[anhhv]

Hi all,

I am computing electron self-energy for a narrow gap semiconductor (~0.25 eV) and get NaN error. I checked Wannier interpolation and phonon running and they're converged. I also tried to set eptemp very low at 10 K but seeing the same problem. One thing I couldn't understand is that even I set eptemp = 10 K and degaussw = 0.005 eV and the Fermi level at the mid-gap but I still got the DOS > 0 at Fermi level. Who has some comments? Thank you guys so much.

Best,
Anh.

[sponce]

Dear anhhv,

The fact that it is a narrow gap should not be too much of a problem.

You can try adding a scissor with the input variable "scissor".

You can also try to impose by hand the Fermi level inside the bandgap using " efermi_read" and " fermi_energy".

Finally, you should try to plot the electronic and phononic bandstructure using "band_plot" and filkf and filqf with two high symmetry path.

If both the resulting BS look ok, and the matrix element decays is also good, it should be fine.

Best wishes,
Samuel

[anhhv]

Hi Samuel,

Factually, the scissor shift is just working for indirect optical absorption computation. But I modified the source code of EPW and it's active for electron-phonon self-energy as well. However, I still got the same problem: NaN for both electron & phonon self-energy. I plotted the electronic and phonon band structures. They look fine. I think the problem might be at electron-phonon scattering matrix g.

Best,
Anh.

[sponce]

Dear Anh,

You can try looking at the el-ph matrix element directly by using the input variable "prtgkk = .true."

http://epw.org.uk/Documentation/Inputs#prtgkk

If you still got NaN, then I would try debugging the code directly with a simplify version of your system (i.e. do a restart with very small fine grids).

You can then add write statement directly inside the code

You will want to look into the ephwann_shuffle.f90 a bit before the CALL print_gkk( iq )

For example you can try printing the "epmatf" variable. If this is still NaN, then try working your way up to understand which variable is causing the Nan.

Hope this helps,
Best wishes,
Samuel

[anhhv]

Hi Samuel,

The problem might be at EPW version. I used EPW copied from School on Electron-Phonon Physics from First Principles in Trieste, Italy (2018). I tried to print |g| and epmatf and they were NaN. But when I rerun everything with qe-6.2 and EPW integrated there, the NaN errors went away. Thank you so much.

Best,
Anh.

[sponce]

Hello Anh,

Oh I see. Sorry about that. We indeed did some last minute implementation for the school and might have introduced some bugs :/ (although it seemed to work during the school).

Any happy it works for you now.

Best,
Samuel