Spin-down Wannierization issues (disentanglement window and TB2J exchange parameters)
17 views
Skip to first unread message
Nezar Adriano
May 6, 2026, 5:40:54 AMMay 6
to TB2J
Hello,
I’m new to Wannier–TB2J workflows and I’m currently facing an issue with my calculations.
My system is LaBaMnO₃, containing 4 La, 2 Ba, 6 Mn, and 18 O atoms. I’m using Quantum ESPRESSO, and I performed wannier.x with:
num_wann = 84, derived from:
Mn–d: 6×5=30
O–p: 18×3=54
Initial projections: Mn–d and O–p
I followed a setup similar to the TB2J tutorial:
https://tb2j.readthedocs.io/en/latest/src/wannier.html
For the spin-up channel, the Wannierization seems well-behaved. The spreads reported in the .wout file are reasonably small
Final State
WF centre and spread 1 ( -0.000247, 3.226525, 11.253949 ) 0.69429055
WF centre and spread 2 ( -0.001291, 3.226577, 11.260101 ) 0.57827088
WF centre and spread 3 ( 0.001130, 3.226899, 11.260021 ) 0.59500688
.
WF centre and spread 83 ( 4.333130, 1.136687, 1.036874 ) 1.10594447
WF centre and spread 84 ( 4.375231, 1.183547, 1.049835 ) 1.44185816
Sum of centres and spreads ( 95.019783,164.566436,533.581487 ) 85.75676674
Spreads (Ang^2) Omega I = 81.626964895
================ Omega D = 0.002634163
Omega OD = 4.127167944
Final Spread (Ang^2) Omega Total = 85.756767003
However, I encounter difficulties with the spin-down channel.
I have performed multiple trial-and-error adjustments, particularly varying:
dis_win_min/max
dis_froz_min/max
From my tests, it seems that dis_win_max plays a critical role (I may be misunderstanding this).
Using dis_win_max as low as possible (dis_win_max = 17.1943) leads to relatively large spreads (much worse than spin-up) and TB2J exchange parameters are a bit overestimated
.wout
Final State
WF centre and spread 1 ( -0.000335, 3.227717, 11.260447 ) 1.46529551
WF centre and spread 2 ( 0.042076, 3.345488, 11.716219 ) 16.08711979
WF centre and spread 3 ( 0.153398, 3.314790, 11.165702 ) 10.74825425
.
WF centre and spread 83 ( 4.335621, 1.162613, 1.065786 ) 1.43474333
WF centre and spread 84 ( 4.374192, 1.181239, 1.046380 ) 1.32424756
Sum of centres and spreads ( 94.473945,165.999905,531.561989 ) 250.12129651
Spreads (Ang^2) Omega I = 173.906245146
================ Omega D = 8.184441997
Omega OD = 68.030609629
Final Spread (Ang^2) Omega Total = 250.121296772
tb2j
Exchange:
i j R J_iso(meV) vector distance(A)
----------------------------------------------------------------------------------------
Mn2 Mn5 ( 0, -1, 0) 7.6236 (-0.000, -3.227, 2.251) 3.934
J_iso: 7.6236
----------------------------------------------------------------------------------------
Mn1 Mn6 ( 0, 1, 0) 9.4849 ( 0.000, 3.227, -2.251) 3.934
J_iso: 9.4849
----------------------------------------------------------------------------------------
Mn5 Mn2 ( 0, 1, 0) 7.6236 ( 0.000, 3.227, -2.251) 3.934
J_iso: 7.6236
----------------------------------------------------------------------------------------
Mn6 Mn1 ( 0, -1, 0) 9.4849 (-0.000, -3.227, 2.251) 3.934
J_iso: 9.4849
----------------------------------------------------------------------------------------
Mn5 Mn2 ( -1, 0, 0) 6.8700 (-2.795, -1.613, -2.251) 3.934
J_iso: 6.8700
.
.
I'm using python to calculate the mean J and Sum J for Nearest Neighbour
Mean J: 8.5495
Sum J : 307.7824
Then I gradually increasing dis_win_max (dis_win_max = 17.2636) and it improves TB2J results (more reasonable exchange parameters). However, Wannier spreads remain significantly larger than spin-up, though this time is better than the first one
.wout
Final State
WF centre and spread 1 ( 0.001459, 3.227197, 11.250960 ) 1.38442166
WF centre and spread 2 ( -0.770473, 3.030198, 10.985088 ) 12.38055445
WF centre and spread 3 ( 0.615704, 3.204737, 11.019994 ) 12.75318612
.
WF centre and spread 83 ( 4.305172, 1.149287, 1.057095 ) 1.41044494
WF centre and spread 84 ( 4.379539, 1.193527, 1.045992 ) 1.29661989
Sum of centres and spreads ( 94.990732,164.415678,533.389254 ) 200.13562892
Spreads (Ang^2) Omega I = 168.112795815
================ Omega D = 2.952998643
Omega OD = 29.069834720
tb2j
Exchange:
i j R J_iso(meV) vector distance(A)
----------------------------------------------------------------------------------------
Mn2 Mn5 ( 0, -1, 0) 11.6162 (-0.000, -3.227, 2.251) 3.934
J_iso: 11.6162
----------------------------------------------------------------------------------------
Mn1 Mn6 ( 0, 1, 0) 11.5253 ( 0.000, 3.227, -2.251) 3.934
J_iso: 11.5253
----------------------------------------------------------------------------------------
Mn5 Mn2 ( 0, 1, 0) 11.6162 ( 0.000, 3.227, -2.251) 3.934
J_iso: 11.6162
----------------------------------------------------------------------------------------
.
.
Mean J: 7.0005
Sum J : 252.0194
I even tried without disentanglement window (dis_win/froz_win not set) and it produces spreads comparable to the spin-up case ie quite good. But TB2J exchange parameters become strongly overestimated
.wout
WF centre and spread 80 ( -1.331384, 2.307697, 10.129439 ) 0.55913450
WF centre and spread 81 ( -1.330266, 2.302123, 10.133249 ) 0.57056179
WF centre and spread 82 ( 4.328880, 1.106559, 1.041711 ) 0.56156776
WF centre and spread 83 ( 4.324885, 1.107147, 1.042591 ) 0.55062756
WF centre and spread 84 ( 4.325797, 1.112102, 1.041094 ) 0.54865900
Sum of centres and spreads ( 95.015702,164.564629,533.582186 ) 45.05457457
Spreads (Ang^2) Omega I = 44.626682974
================ Omega D = 0.000082356
Omega OD = 0.427809500
Final Spread (Ang^2) Omega Total = 45.054574830
------------------------------------------------------------------------------
tb2j
Exchange:
i j R J_iso(meV) vector distance(A)
----------------------------------------------------------------------------------------
Mn2 Mn5 ( 0, -1, 0) 50.2076 (-0.000, -3.227, 2.251) 3.934
J_iso: 50.2076
----------------------------------------------------------------------------------------
Mn1 Mn6 ( 0, 1, 0) 50.2263 ( 0.000, 3.227, -2.251) 3.934
J_iso: 50.2263
----------------------------------------------------------------------------------------
Mn5 Mn2 ( 0, 1, 0) 50.2076 ( 0.000, 3.227, -2.251) 3.934
J_iso: 50.2076
----------------------------------------------------------------------------------------
Mn6 Mn1 ( 0, -1, 0) 50.2263 (-0.000, -3.227, 2.251) 3.934
J_iso: 50.2263
.
.
Mean J: 35.2662
Sum J : 1269.5838
I do not know the exact expected values of Jij for this system. However, I attempted to validate the results by running VAMPIRE simulations to estimate the Curie temperature Tc. The second case earlier gives a Tc that is consistent with literature while the no-window case clearly overestimates Tc
So I'm wondering where could be possibly the problem here?
Is it expected that spin-down channels require significantly different disentanglement windows?
Could this indicate that my projection choice (Mn–d and O–p only) is insufficient for the spin-down channel?
More generally what is the recommended way to determine dis_win and froz_win in such systems?
For reference, I also attach the bands.x results from QE for spin-up and spin-down channels as well as PDOS
Any guidance or suggestions would be greatly appreciated
Thank you!
I’m new to Wannier–TB2J workflows and I’m currently facing an issue with my calculations.
My system is LaBaMnO₃, containing 4 La, 2 Ba, 6 Mn, and 18 O atoms. I’m using Quantum ESPRESSO, and I performed wannier.x with:
num_wann = 84, derived from:
Mn–d: 6×5=30
O–p: 18×3=54
Initial projections: Mn–d and O–p
I followed a setup similar to the TB2J tutorial:
https://tb2j.readthedocs.io/en/latest/src/wannier.html
For the spin-up channel, the Wannierization seems well-behaved. The spreads reported in the .wout file are reasonably small
Final State
WF centre and spread 1 ( -0.000247, 3.226525, 11.253949 ) 0.69429055
WF centre and spread 2 ( -0.001291, 3.226577, 11.260101 ) 0.57827088
WF centre and spread 3 ( 0.001130, 3.226899, 11.260021 ) 0.59500688
.
WF centre and spread 83 ( 4.333130, 1.136687, 1.036874 ) 1.10594447
WF centre and spread 84 ( 4.375231, 1.183547, 1.049835 ) 1.44185816
Sum of centres and spreads ( 95.019783,164.566436,533.581487 ) 85.75676674
Spreads (Ang^2) Omega I = 81.626964895
================ Omega D = 0.002634163
Omega OD = 4.127167944
Final Spread (Ang^2) Omega Total = 85.756767003
However, I encounter difficulties with the spin-down channel.
I have performed multiple trial-and-error adjustments, particularly varying:
dis_win_min/max
dis_froz_min/max
From my tests, it seems that dis_win_max plays a critical role (I may be misunderstanding this).
Using dis_win_max as low as possible (dis_win_max = 17.1943) leads to relatively large spreads (much worse than spin-up) and TB2J exchange parameters are a bit overestimated
.wout
Final State
WF centre and spread 1 ( -0.000335, 3.227717, 11.260447 ) 1.46529551
WF centre and spread 2 ( 0.042076, 3.345488, 11.716219 ) 16.08711979
WF centre and spread 3 ( 0.153398, 3.314790, 11.165702 ) 10.74825425
.
WF centre and spread 83 ( 4.335621, 1.162613, 1.065786 ) 1.43474333
WF centre and spread 84 ( 4.374192, 1.181239, 1.046380 ) 1.32424756
Sum of centres and spreads ( 94.473945,165.999905,531.561989 ) 250.12129651
Spreads (Ang^2) Omega I = 173.906245146
================ Omega D = 8.184441997
Omega OD = 68.030609629
Final Spread (Ang^2) Omega Total = 250.121296772
tb2j
Exchange:
i j R J_iso(meV) vector distance(A)
----------------------------------------------------------------------------------------
Mn2 Mn5 ( 0, -1, 0) 7.6236 (-0.000, -3.227, 2.251) 3.934
J_iso: 7.6236
----------------------------------------------------------------------------------------
Mn1 Mn6 ( 0, 1, 0) 9.4849 ( 0.000, 3.227, -2.251) 3.934
J_iso: 9.4849
----------------------------------------------------------------------------------------
Mn5 Mn2 ( 0, 1, 0) 7.6236 ( 0.000, 3.227, -2.251) 3.934
J_iso: 7.6236
----------------------------------------------------------------------------------------
Mn6 Mn1 ( 0, -1, 0) 9.4849 (-0.000, -3.227, 2.251) 3.934
J_iso: 9.4849
----------------------------------------------------------------------------------------
Mn5 Mn2 ( -1, 0, 0) 6.8700 (-2.795, -1.613, -2.251) 3.934
J_iso: 6.8700
.
.
I'm using python to calculate the mean J and Sum J for Nearest Neighbour
Mean J: 8.5495
Sum J : 307.7824
Then I gradually increasing dis_win_max (dis_win_max = 17.2636) and it improves TB2J results (more reasonable exchange parameters). However, Wannier spreads remain significantly larger than spin-up, though this time is better than the first one
.wout
Final State
WF centre and spread 1 ( 0.001459, 3.227197, 11.250960 ) 1.38442166
WF centre and spread 2 ( -0.770473, 3.030198, 10.985088 ) 12.38055445
WF centre and spread 3 ( 0.615704, 3.204737, 11.019994 ) 12.75318612
.
WF centre and spread 83 ( 4.305172, 1.149287, 1.057095 ) 1.41044494
WF centre and spread 84 ( 4.379539, 1.193527, 1.045992 ) 1.29661989
Sum of centres and spreads ( 94.990732,164.415678,533.389254 ) 200.13562892
Spreads (Ang^2) Omega I = 168.112795815
================ Omega D = 2.952998643
Omega OD = 29.069834720
tb2j
Exchange:
i j R J_iso(meV) vector distance(A)
----------------------------------------------------------------------------------------
Mn2 Mn5 ( 0, -1, 0) 11.6162 (-0.000, -3.227, 2.251) 3.934
J_iso: 11.6162
----------------------------------------------------------------------------------------
Mn1 Mn6 ( 0, 1, 0) 11.5253 ( 0.000, 3.227, -2.251) 3.934
J_iso: 11.5253
----------------------------------------------------------------------------------------
Mn5 Mn2 ( 0, 1, 0) 11.6162 ( 0.000, 3.227, -2.251) 3.934
J_iso: 11.6162
----------------------------------------------------------------------------------------
.
.
Mean J: 7.0005
Sum J : 252.0194
I even tried without disentanglement window (dis_win/froz_win not set) and it produces spreads comparable to the spin-up case ie quite good. But TB2J exchange parameters become strongly overestimated
.wout
WF centre and spread 80 ( -1.331384, 2.307697, 10.129439 ) 0.55913450
WF centre and spread 81 ( -1.330266, 2.302123, 10.133249 ) 0.57056179
WF centre and spread 82 ( 4.328880, 1.106559, 1.041711 ) 0.56156776
WF centre and spread 83 ( 4.324885, 1.107147, 1.042591 ) 0.55062756
WF centre and spread 84 ( 4.325797, 1.112102, 1.041094 ) 0.54865900
Sum of centres and spreads ( 95.015702,164.564629,533.582186 ) 45.05457457
Spreads (Ang^2) Omega I = 44.626682974
================ Omega D = 0.000082356
Omega OD = 0.427809500
Final Spread (Ang^2) Omega Total = 45.054574830
------------------------------------------------------------------------------
tb2j
Exchange:
i j R J_iso(meV) vector distance(A)
----------------------------------------------------------------------------------------
Mn2 Mn5 ( 0, -1, 0) 50.2076 (-0.000, -3.227, 2.251) 3.934
J_iso: 50.2076
----------------------------------------------------------------------------------------
Mn1 Mn6 ( 0, 1, 0) 50.2263 ( 0.000, 3.227, -2.251) 3.934
J_iso: 50.2263
----------------------------------------------------------------------------------------
Mn5 Mn2 ( 0, 1, 0) 50.2076 ( 0.000, 3.227, -2.251) 3.934
J_iso: 50.2076
----------------------------------------------------------------------------------------
Mn6 Mn1 ( 0, -1, 0) 50.2263 (-0.000, -3.227, 2.251) 3.934
J_iso: 50.2263
.
.
Mean J: 35.2662
Sum J : 1269.5838
I do not know the exact expected values of Jij for this system. However, I attempted to validate the results by running VAMPIRE simulations to estimate the Curie temperature Tc. The second case earlier gives a Tc that is consistent with literature while the no-window case clearly overestimates Tc
So I'm wondering where could be possibly the problem here?
Is it expected that spin-down channels require significantly different disentanglement windows?
Could this indicate that my projection choice (Mn–d and O–p only) is insufficient for the spin-down channel?
More generally what is the recommended way to determine dis_win and froz_win in such systems?
For reference, I also attach the bands.x results from QE for spin-up and spin-down channels as well as PDOS
Any guidance or suggestions would be greatly appreciated
Thank you!
Reply all
Reply to author
Forward
0 new messages