[QE-users] Improve convergence with noncollinear calculation

[Christoph Wolf]

Dear all,

I am having some issues with the energy convergence of a noncollinear
calculation to determine energy difference for different spin alignments.
The expected energies are ~1...5 meV so energy convergence should be rather
tight.

However I find that with high degauss values I cannot reach energy values
with the necessary precision and with low degauss and low conv_thr I never
reach convergence. So I do a restart loop with gradually smaller degauss,
mixing_beta and conv_thr but the resulting "algorithm" is very slow and
despite going to mxiing_beta ~0.01, degauss=0.001 I cannot reach
conv_thr~1D-10

Does anyone have a hint on how to reach the desired conv_thr with
reasonable "speed"? I attach the convergence behavior for decreasing
degauss and mixing_beta (for values where convergence gets tricky: degauss
from 0.005, 0.001, 0.0005, conv_thr=5D-10, beta=10*degauss)

[image: image.png]

electron_maxstep is 60 for all iterations over degauss; I have to mention
that the magnetic momenta do not fluctuate anymore, it seems that there
simply is a lower limit to which convergence I can achieve (?)

Any help is appreciated!

Thanks in advance,
Chris

-- 
Postdoctoral Researcher
Center for Quantum Nanoscience, Institute for Basic Science
Ewha Womans University, Seoul, South Korea
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20191127/a9732efe/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 57630 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20191127/a9732efe/attachment.png>