[Wannier] Spread convergence in a spin-oribit calculation
ppadmana at ph2.uni-koeln.de
ppadmana at ph2.uni-koeln.de
Tue Mar 6 17:43:12 CET 2018
Hello Everyone,
I am attempting to perform a Wannierization of the valence bands of PbTe. I have performed a spin orbit calculation in QE (with lspinorb = true and noncolin = true). I did this using a 16x16x16 shifted grid for the scf calculation and an 8x8x8 full grid for the nscf calculation. Based on the results of a non-SO calculation, I know that the bands are primarily of Te-p and Pb-s character.
In the Wannierization step, I use the following input:
num_bands = 10
num_wann = 10
num_iter = 1000
conv_window = 10
conv_tol = 1e-5
spinors = true
bands_plot = true
begin kpoint_path
L 0.500 0.500 0.500 G 0.000 0.000 0.000
G 0.000 0.000 0.000 X 0.000 0.500 0.500
X 0.000 0.500 0.500 W 0.250 0.750 0.500
W 0.250 0.750 0.500 L 0.500 0.500 0.500
L 0.500 0.500 0.500 G 0.000 0.000 0.000
G 0.000 0.000 0.000 K 0.375 0.750 0.375
end kpoint_path
begin projections
c=-3.046778,3.731389,3.046598:pz
c=-2.754606,2.754517,2.755727:s
c=-3.731378,3.046842,3.046648:pz
c=-3.047456,3.047471,3.731244:pz
c=-3.145233,3.145233,6.290419:s
end projections
mp_grid = 8 8 8
begin kpoints
**list of 512 points identical to nscf calculation**
end kpoints
begin atoms_frac
Pb 0.000 0.000 0.000
Te 0.500 0.500 0.500
end atoms_frac
begin unit_cell_cart
Ang
-3.145186 0.000000 3.145186
0.000000 3.145186 3.145186
-3.145186 3.145186 0.000000
end unit_cell_cart
Unfortunately, even after 1000+ iterations, though my spreads are in the range of 4-5 Ang^2, they aren't converged, and I can only get to a DLTA of approximately 1E-5. If I look at the band structure, I see a difference between the Kramer’s degenerate states and wiggles at the high symmetry points. Moreover, unless I explicitly set the conv_tol to 1e-5 and convert_window to 10, it will basically iterate to the end and give me results that are even worse. It seems like gradient descent can’t find the global minima. Here’s a grep of the output and final state:
+--------------------------------------------------------------------+<-- CONV
| Iter Delta Spread RMS Gradient Spread (Ang^2) Time |<-- CONV
+--------------------------------------------------------------------+<-- CONV
0 0.886E+02 0.0000000000 88.5961756930 2.02 <-- CONV
1 -0.193E+02 3.6519324709 69.2719457144 2.09 <-- CONV
2 -0.119E+02 2.9217065446 57.4196136819 2.16 <-- CONV
3 -0.789E+01 2.2523091155 49.5285805341 2.23 <-- CONV
4 -0.405E+01 1.8769695123 45.4786941630 2.29 <-- CONV
5 -0.117E+01 1.4810561663 44.3102929602 2.36 <-- CONV
...
732 -0.954E-05 0.0812066914 41.5149635930 50.03 <-- CONV
733 -0.485E-05 0.2124621933 41.5149587471 50.09 <-- CONV
734 -0.587E-05 0.0551927135 41.5149528784 50.16 <-- CONV
735 -0.479E-05 0.2156962496 41.5149480855 50.22 <-- CONV
736 -0.461E-05 0.0560167032 41.5149434756 50.28 <-- CONV
737 -0.365E-05 0.1892001198 41.5149398256 50.35 <— CONV
Final State
WF centre and spread 1 ( -2.924839, 3.504327, 3.469653 ) 4.69435114
WF centre and spread 2 ( -2.925054, 3.503452, 3.470009 ) 4.71351191
WF centre and spread 3 ( -2.673238, 2.673025, 2.971221 ) 3.59977954
WF centre and spread 4 ( -2.672922, 2.673446, 2.971132 ) 3.60028628
WF centre and spread 5 ( -3.503953, 2.924918, 3.470236 ) 4.70571734
WF centre and spread 6 ( -3.505087, 2.924613, 3.471128 ) 4.70069921
WF centre and spread 7 ( -3.482302, 3.482023, 2.682537 ) 3.93809935
WF centre and spread 8 ( -3.481404, 3.482794, 2.682584 ) 3.93941997
WF centre and spread 9 ( -3.145268, 3.145207, 6.293045 ) 3.82072386
WF centre and spread 10 ( -3.145260, 3.145161, 6.293306 ) 3.80235121
Sum of centres and spreads (-31.459326, 31.458966, 37.774852 ) 41.51493983
I should point out that when doing the non spin-orbit case, I didn't have a problem achieving convergence with spreads in the 2-3 Ang^2 range and DLTAs of 1E-11.
I tried various initial guesses for the projections, going so far as to using the centers from the non-SO MLWFs, but no luck. I would greatly appreciate any guidance. Thank you in advance.
Sincerely,
Prashant Padmanabhan
---
Post-doctoral Researcher
II. Physics Institute
University of Cologne
50823 Cologne
Germany
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/wannier/attachments/20180306/841c6083/attachment.html>
-------------- next part --------------
_______________________________________________
Wannier mailing list
Wannier at lists.quantum-espresso.org
https://lists.quantum-espresso.org/mailman/listinfo/wannier
More information about the Wannier
mailing list