Matrix elements in Si with FHI pseudo

[gbrunin]

Dear all,

I am working on a comparison of the matrix elements computed from QE and from EPW in Silicon.
First, I am using an old pseudo : https://www.quantum-espresso.org/upf_files/Si.pw-mt_fhi.UPF
Using this pseudo, I do not manage to get a good interpolation of the matrix elements.
When I plot the Wannier-interpolated phonon dispersion and band structure, they look fine.
When I plot the decays, however, I see that the Wannier representation of the matrix elements decay to 2e-4, the Hamiltonian to 2e-5, the dipole to 2e-4, |g(Re,0)| to 1e-4 but |g(0,Rp)| to 2e-2 only.

When I use a recent pseudo (from the pseudo dojo), then all the bands and decays are fine (~1e-4 - 1e-5), and in this case I can get a good correspondence between the DFPT and interpolated matrix elements.
All the other parameters are the same (adapted to the pseudo: the structure, energy cut-off, and energy windows for the wannierization are changed of course).

My question is: could this discrepancy with the old pseudo between the DFPT and interpolated matrix elements be due to the bad decay of the g(0,Rp) ?
If so, how could I change the wannierization to fix this issue ? Is it sufficient to try and decrease the spread of the wannier functions ? The spreads with the old pseudo are slightly larger than with the recent pseudo.

Do not hesitate to ask me for more data or plots if needed.

Thanks for your help,
Guillaume

[gbrunin]

Dear all,

Using only the 4 valence bands and s states localized on the centers of the bonds as initial projections, I managed to decrease the spread of my Wannier functions using the FHI pseudo.
All the decays except are not ~1e-5 - 1e-6. The decay of |g(0,Rp)| is still around ~1e-2, and the problem is still there when interpolating the matrix elements.

Would you have any suggestion to decrease the decay of g(0,Rp) ?

Thanks,
Guillaume