Significant variations in putput across different versions

[Raito Shin]

Dear TB2J Developers,

Recently, while testing the magnetic exchange parameters of MnTe and MnF2 using ABACUS+TB2J, I noticed significant differences in the output across different TB2J versions. Interestingly, the results from older versions (before 0.9.0) appear to align better with experimental data.

As shown in the attached image, the first and third columns respectively represent the J1, J2, J3 values obtained from the mapping method and experimental measurements [Phys. Rev. B 111, 104416].

For MnTe, the following results were obtained with different TB2J versions:
Versions before 0.9.0: J1 = -11 meV, J2 = -0.4 meV, J3 = -3.76 meV
Version 0.9.12.23: J1 = -9.25 meV, J2 = 1.9 meV, J3 = -2.4 meV

Similarly, for MnF2, earlier versions (before 0.9.0) also yield more reasonable results.

Version 0.8.0: J1=-0.2 meV J2=-2.5 meV    (mapping J1=0.6 meV)

Version 0.9.12.13 : J1=-2.2 meV J2=-3.1 meV 

This issue has been raised by other users as well, as discussed in the following thread:
https://groups.google.com/g/tb2j/c/8VFaqwMNUwo/m/sDGsbgzIAwAJ

Given that older versions do not support DMI calculations, while newer versions produce exchange parameters that seem less physically consistent, could you please provide guidance on how to resolve this discrepancy?

Thank you for your attention and support.

[Xu He]

Hi, 

There is a algorithm to find the integration range by searching for a gap below the emin, which reuires a huge gap before v0.9. Thus the integration is from a very deep states to the Fermi energy, which makes the integration less accurate with the same number of points on the integration path.

So mathematically, this should improve the result. The change of the agreement with experimental result was a coincidence, not because the previous method is systematically better.

I am aware of the issue in MnF2, this is because the Mn-F-Mn is a 90 degree bond superexchange, in which the 3d orbital of one Mn atom hybridize with one F 2p orbital, and the other Mn 3d hybridize with onther F 2p orbital. And the intra-site exchange between these two F 2p orbitals are not counted in the J(Mn-Mn). Thus the 1NN J is not ferromagnetic. This can be corrected by the ligand correction method, by first treating the F atom as magnetic, computing the Mn-F exchange, and then do a downfolding. 

Best regards, 

HeXu

[Raito Shin]

Dear He Xu,

Thank you very much for your prompt reply and the detailed explanation. I will proceed to test the downfolding functionality.

I have one technical question regarding this: does the downfolding procedure require the DFT calculations to be performed in a FM configuration? According to the TB2J manual, it seems that the downfolding scheme does not explicitly support AFM states, but I am not sure if I have interpreted this correctly.

I would appreciate your clarification.
Thank you again for your help.

Best regards,
Raito