EPW Forum

Hello,

I have a question on calculations using SOC. I have made an input file for the standalone wannier90.x which correctly reproduces my bandstructure in the selected window. But when I run EPW (version 4.0.0) with the same parameters the wannierization differs and results in a substantially larger spread (not catastrophic, but resulting already in a bandstructure with noticeable differences). The only major difference I can see when comparing the *.wout produced by both runs is that in the EPW-library mode there is this warning:

Ignoring <spinors> in input file

And I am wondering if that might be the cause for the discrepancies. Maybe, I am lacking knowledge of wannier90, but my impression was that I needed this keyword to run it with spin-orbit coupling.

Is there another possible reason, why the library mode differs from the standalone (both are from the same QE installation).

Thank you very much for your help.

Best wishes,
Hannes

Dear Hannes,

There was a bug in the Wannier centre initializations.

This has been corrected in EPW v4.1 and is reported in http://epw.org.uk/Main/About (if you scroll down).

Please let us know if it solves your problem.

Best,

Samuel

Dear Samuel,

Thank you for the answer. I installed QE 6.0 and reran the exact same input files (removing the obsolete keywords for epw of course), but now the wannierisation fails completely. That is to say the disentanglement stage is successful (but requires more steps than in stand-alone or epw 4.0) but the subsequent minimisation does not give localised orbitals. I was wondering if there was a way to use the results of the stand-alone setup (*.amn, *.mnn etc.) also in the library mode, but wannier90 doesn't seem to provide this as a feature. I did notice that now EPW is writing the *.nnkp file and I do find a significant difference here, in the spinor block

wannier90.x stand alone:

begin spinor_projections
44
0.00000 0.57735 2.47586 1 1 1
0.0000000 0.0000000 1.0000000 1.0000000 0.0000000 0.0000000 1.00
1 0.0000000 0.0000000 1.0000000
0.00000 0.57735 2.47586 1 1 1
0.0000000 0.0000000 1.0000000 1.0000000 0.0000000 0.0000000 1.00
-1 0.0000000 0.0000000 1.0000000
0.00000 0.57735 2.47586 1 2 1
0.0000000 0.0000000 1.0000000 1.0000000 0.0000000 0.0000000 1.00
1 0.0000000 0.0000000 1.0000000

etc.

EPW v4.1

begin spinor_projections
44
0.33333 0.66667 0.62602 1 1 1
0.0000000 0.0000000 1.0000000 1.0000000 0.0000000 0.0000000 1.00
1 0.0000000 0.0000000 1.0000000
0.33333 0.66667 0.62602 1 1 1
0.0000000 0.0000000 1.0000000 1.0000000 0.0000000 0.0000000 1.00
-1 0.0000000 0.0000000 1.0000000
0.33333 0.66667 0.62602 1 2 1
0.0000000 0.0000000 1.0000000 1.0000000 0.0000000 0.0000000 1.00
1 0.0000000 0.0000000 1.0000000

The difference in the center coordinates looks to me like a lattice vs. cartesian unit problem.

If you don't have any idea what might be causing this, would you think it could be a viable strategy for me, to change the code such that it reads these files from a separate wannier90.x -pp (and pw2wann.x) run? Or do you think the source of the problem is not in the setup but in the main wannierisation call?

Thank you very much for your help.

Best wishes,
Hannes

PS: I see that my original question regarding "ignoring <spinors> in input file " is not relevant, as the code outputs "Spin CASE ( non-collinear )" correctly.

Dear Hannes,

Could you put here or send me via email a reduced case of your problem (i.e. preferably something fast to run, also including the psps)?

I'm happy to look at this in detail since it should be working correctly now.

Best,

Samuel

The problem has been resolved by Samuel. The wannierization of EPW did not fail, it was simply a difference between the input file used for the standalone wannier90 and the internal EPW calls (the used input file for wannier90 was wrong).

Thanks for the help again.
Hannes