EPW Forum

Dear experts,

I am calculating a 3D structure with 12 atoms in the unit cell. In order to successfully wannierize it I had to consider 40 wannier bands (bc they are all entangled).
From these 40 states I really need maybe 10 around the Fermi level, and all the rest I want to ignore. Is there a way to do this?
The problem is that with my 40 bands the resulting .epb files are extremely heavy (more than my storage quota), so I want to exclude bands that are not close to the Fermi level.

I believe that variables 'nbndskip', 'bands_skipped' etc will actually influence the wannierisation and that is not what i want. I want to keep wannierisation with 40 bands, and then ignore most of them in the actual epw calculation.

Thank you.
Mikhail

I've got the same question. I've got a huge bandwidth of entangled bands but only want the matrix elements between the bands near the Fermi energy calculated. I only seem to get good localization in the wannierization by including ~ 65 or more bands, but then EPW takes a long time, runs into memory issues, etc.

Dear adenchfi:

In case that you excluded some bands in the Wannierization step, the recent development version of EPW (downloaded at https://gitlab.com/QEF/q-e) excludes these bands also in the step of evaluation of electron-phonon matrix elements on coarse grids, leading to the shorter computing time, the smaller epb files, etc.

The EPW included in the official release of Quantum ESPRESSO (QE) v6.5 (downloaded at https://github.com/QEF/q-e/releases or https://gitlab.com/QEF/q-e/-/tags) evaluates electron-phonon matrix elements for all bands (nbnd) on coarse grids even if we excluded bands in the Wannierization step.

However, now you want to exclude additional bands (consequently, the number of bands is small than that in the band manifold determined by the Wannierization step) and this requires direct modification of codes. I also would like to point out that in this case there might be some issue which we have to be careful at. Especially, if your reduced manifold near the Fermi level is not isolated from the bands above or below, the situation becomes more complex.

Therefore, I would suggest you to find the solution without changing the band manifold determined by the Wannierization step.

If you upload all inputs and outputs with the information on pseudopotentials, it helps for me to suggest a possible solution.

For instance,
1. you can try the SCDM method recently implemented in Wannier90 v3.0 (https://epubs.siam.org/doi/abs/10.1137/17M1129696) for the smallest isolated band manifold including your reduced one.
2. you can try band parallelization in EPW using the image communicator in QE.
3. you can try to change the value of etf_mem (http://epw.org.uk/Documentation/Inputs#etf_mem).

However, the usefulness of the examples above depends on where you encounter problems. So I need to know in detail about your calculations.

Sincerely,

H. Lee

Hello H. Lee,

Thanks for the reply! I am installing the latest version of QE/EPW to see if using the SCDM method gives a better guess for finding a smaller reduced manifold, which would indirectly fix my problems.

I will reply further if that doesn't work!