evaluate the wannierization to get J

[chao zhou]

Dear Dr. Xu,

I hope this message finds you well.

I’d like to revisit a recurring issue related to Wannierization. Specifically, I’m trying to understand which part of the band structure needs to be fitted most accurately in order to obtain reliable exchange parameters J when constructing the Wannier functions.

By varying the frozen window, I am able to obtain a good fit for the bands just below the Fermi level. However, I’ve noticed that this sometimes introduces inconsistencies. The wannier90.wout files look quite similar in all cases, and the main difference seems to lie in how well the target window is fitted.

When I ensure that the bands below the Fermi level are well-reproduced and proceed to calculate the exchange coupling J, the resulting magnetic moment tends to be underestimated compared to VASP. For example, using ABACUS gives a nearest-neighbor J \approx -10, while the Wannier-based approach only yields around -4.

I believe the results shouldn’t differ too much across different codes, so I suspect the issue lies elsewhere in the Wannierization process. I’d greatly appreciate any insights you could provide.

I will also share the relevant result files with you for reference.

Best regards,

Chao Zhou

[matthieu verstraete]

Hello, 

Indeed,  the J are very sensitive to the wannierization. The bands can be equally good with different optimizations and localization but the wf are not centered or symmetric and the J will suffer. We are working on corrections for this problem. Abacus and siesta will be more robust. 

When you say VASP you mean energy differences or the 4 state method?

Comparing to kkr, He Xu has also observed that the wf J are usually smaller, but not 40%.

Best

Matthieu

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Professor Matthieu J Verstraete

Fellow, American Physical Society

Chair, Steering Committee, European Theoretical Spectroscopy Facility

Alumnus Fellow, Young Academy of Europe  

Institute for Theoretical Physics (ITP) Department of Physics

Buys Ballot Gebouw/Building, Princetonplein 5, office 

University of Utrecht,  3584 CC Utrecht

Phone: +31 30 253 5926 

Secretariat: +31 30 253 5929 and science....@uu.nl

Group web page (Liège): http://www.nanomat.ulg.ac.be/

Nanomat lab, Q-Mat center, Université de Liège
Département de Physique, Bat. B5a, 4/50
Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium
Phone : +32 4 366 90 17

European Theoretical Spectroscopy Facility (ETSF)

Mail : matthieu....@uliege.be
          matthieu.jea...@gmail.com

          m.j.ver...@uu.nl

--
You received this message because you are subscribed to the Google Groups "TB2J" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tb2j+uns...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/tb2j/375f5e9b-a362-42f4-963c-0c610e3acc8cn%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[chao zhou]

Dear Matthieu,

Thank you very much for your helpful response.

It seems that the discrepancy indeed originates from the Wannier fitting using VASP. This is quite puzzling. When using either energy differences or the four-state method, the nearest-neighbor J values are around 17 meV. However, the values obtained from TB2J are only around 9 meV (after applying the factor of 2), which is clearly inconsistent.

Interestingly, when I reduce the frozen window of the Wannierization to a narrow region near the Fermi level, the band structure below the window becomes poorly fitted—but the resulting J values get closer to the expected ones. This makes it quite difficult to evaluate which approach is more trustworthy.

Thanks again for your insights!

Chao Zhou

[matthieu verstraete]

Reducing the frozen window makes things worse - the impact on the J is not "improvement" it's just a monotonic change, which happens to be in the direction you expect.

You can try to avoid the w90 optimization, and just take the first iteration of projection on atomic orbital characters. This tends to give more centered and eventually symmetric WF (and usually gives decent band interpolation though more spread WF) - let's see if it affects the Js as well.

M.

To view this discussion visit https://groups.google.com/d/msgid/tb2j/5a9294a4-55c8-48aa-b6fc-63db74ae67b1n%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Professor Matthieu J Verstraete

Fellow, American Physical Society

Chair, Steering Committee, European Theoretical Spectroscopy Facility www.etsf.eu

Alumnus Fellow, Young Academy of Europe  yacadeuro.org

Institute for Theoretical Physics (ITP) Department of Physics

Buys Ballot Gebouw/Building, Princetonplein 5, office 

University of Utrecht,  3584 CC Utrecht

ITP Secretariat: +31 30 253 5928 E-mail: science....@uu.nl

[chao zhou]

Thanks, Matthieu. I’ve double-checked: when I also include the bands above the Fermi level in the fit, the exchange J looks much better—Wannier now gives −7.5 meV vs. Abacus −10 meV, which is acceptable.

The DMI, however, is still puzzling: the Wannier result is far off, matching neither the energy-mapping method nor Abacus. I generated the Wannier functions with the “structure-rotation → merge” workflow. Is there something in that procedure I should watch out for? I’ve attached the SOC-fitted band structure for one direction. How can I further improve the DMI obtained from Wannier?