Yambo Community Forum

Dear all,

My aim is to calculate the electron effective masses of germanium. For my calculations I am using LDA HGH pseudopotentials with spin orbit interaction, and Abinit to generate the input files for Yambo. I have generated the electron band structure of Ge from DFT for 16x16x16 k grids using Ypp and obtained good agreement with Abinit.

Using Ypp I obtained the conduction band energies outward from L in the parallel direction for DFT, these are the blue stars in the figure below, I have attached the input file used. I also obtained the energies at these same points using Abinit, the red pluses. The cyan squares are from a series of shifted 16x16x16 k grid DFT calculations from Abinit, where the shifts applied were chosen to align with the k points I requested from Ypp. The x axis is the displacement from L, epsilon, and the y axis is the band energy.



The effective mass obtained from Abinit's points agrees with experimental data, but there's some discrepancy between the energies calculated from Abinit and Yambo. There's a similar difference as well for the perpendicular direction from L. I was wondering if I was doing something incorrect or if there was an explanation somewhere that detailed the origin of this discrepancy and a possible method to use to overcome it? Outside of using a denser k grid, is there any way to give Yambo a specific line of k-points along which to calculate the energy more accurately?

I also attempted to calculate the effective mass under COHSEX, done on a 16x16x16 k grid, as seen in the figure below. The electronic band structure looks reasonable, but looking close to L there seems to be some wiggling in the conduction band energies in contrast to the expected parabola:



I obtained similar for PPA. I have also attached the input file for my COHSEX calculation. I was wondering if it's possible to do GW corrections on a denser line of k-points along a given direction without having to increase the k grid or if there is some way to correct for this?

Kind regards,
Ronan

Dear Ronan,
let me resume your question just to be sure I well understood.
You calculate your ground state with abinit with a uniform k-point grid. Next you used ypp to interpolate the band structure along a path that was not contained in your k-point sampling. You compared the results of the interpolation with the energies calculated with abinit with an nscf (or scf) calculation on that path.
The results are similar but not exactly the same.
Here you need to have in mind that the abinit calculations come from a diagonalization of your KS hamiltonian while the yambo results is an interpolation, so it is reasonable that some difference arise.
You can have a look to this post in order to try to have a better interpolation, but it is not guarantee that it will work.
Other solution would be to have the k points you are interested in contained in an uniform grid, in this case the yambo and abinit results would be exactly the same as yambo read that energies when using the interface a2y, and they are contained in the r_setup file.


About the possibility to calculate the correction along a specific path, you need to produce an uniform grid containing that path, you can have a look to this feature of ypp, but you need to define the points on the path one by one, this is useful for a specific k-point, when defining a path I suspect that it is equivalent to have a much denser grid. I'm not sure again if you already used this tools to produce the cyan points, so performing a series of calculations matching each points you were interested in. If this is the case, you would not need to use the interpolator, as that energies should be present in the r_setup files, and are essentially the same of abinit, as they are read from the database produced by abinit, so it looks me very strange they do not coincide.

About the cohsex energies, it is not clear to me if this is again the result of an interpolation, or it is done in the k point present in the ground states, anyway in both case the fact you do not have the expected parabola could be due to convergences issue? It is not guarantee that all the points converge exactly with the same parameter. If this is an interpolation, anyway, the problem also here could be due to the quality of the interpolation.

Please also note that the interpolator could have some problem if the time-reversal symmetry is present (or better if the inversion symmetry is not present).
In the next release of yambo (coming very soon) there will be tools to overcome this problem.


But again, I'm not sure I fully understood the steps you did. In this case do not hesitate to post here again your points.

Hope it helps,

Daniele

Dear Daniele,

Thank you very much for the reply. Yes, you have that correct for what I did. And for the COHSEX calculation it was an interpolation. In regards to time symmetry, I did remove that from these calculations.

Just so I have it clear, to avoid any differences with interpolation I should make the grid dense enough to contain the points I'm interested in or use the ypp tool you linked too?

Thank you again,
Ronan

Dear Ronan,
once you have a grid with the points you are interested in, then you do not need to interpolate them anymore.
The tool I indicated it needs to generate a grid containing the points you indicate.
Moreover, having even some points (not all) along the path in your grid, should help the interpolation to be more precise.
Best,
Daniele