Dear All
Finally the problem is not completely solved.
More precisely, when we are doing GGA+SO calculations and using a
correct kmesh (no temporal symmetry), we obtain a symmetric
magnetocrystalline anisotropy, namely same MAE along [0 1 0] and [0 -1 0].
In contrast, when we are doing
Dear Profs. Blaha and Marks,
Thank you for your advices. I have a question regarding your comment on
the EECE case.
The resulting effective potential of LDA+U and EECE corrections is
decomposed in vorb files, but I don't understand why the spin-coupling
file case.vorbud shouldn't appear in
Dear Peter and Laurence
Thank you for your replies. Using the following strategy it seems to
work nicely (preliminary results):
runsp_lapw -eece -p -ec 0.1 -NI
Then for larger EMAX and kmesh without SO we do :
x lapw1 -up -c -orb
x lapw1 -dn -c -orb
And for the different magnetization
Dear Xavier,
The method you just used avoids time-reversal symmetry for -orb. My
intuition is that this is right, but I don't think I ever convinced
Peter of this. An alternative (that might be wrong) and avoids double
counting is:
Setup case.inso with NO spin-orbit on any atom, then do
Here is a document showing the results graphically.
https://filesender.renater.fr/?s=download=8ac3a214-edfa-4894-fa1f-27aba5a5522f
It really looks like the problem we had before (using bad kmesh).
We test it on two different compounds and in both cases WIEN2k_16 gives
a correct picture and
Dear Laurence
Here is the point. Our results show that in version 17 a problem occurs
related to the time-reversal, which appears only if we do GGA+U or EECE.
a) The calculations we showed are already done in P1. Indeed, I wanted
to avoid any problems related to symmetry.
b) we (me and my
Hallo Xavier,
Looks rather strange.
Eventually I would have expected problems both, in 16.1 and 17.1 (but
not 14.2) due to the off-diagonal density matrices. But this should
concern ONLY LDA+U, not -eece.
Just to be sure:
I expect you do SO non-selfconsistent, so vorbup/dn(du) files are
Dear Peter
You are totally correct. We are doing SO non-selfconsistent by using a
standard procedure for EECE calculations:
runsp_lapw -eece -p -ec 0.1 -NI
and then we estimate the MAE using this non-SCF procedure :
Increase EMAX in case.in1c - increase kmesh if needed
x lapw1 -up
OK we will check it. In EECE we have no case.vorbud file (empty file).
Le 16/01/2018 à 16:58, Peter Blaha a écrit :
Hups: If this is true, you are counting the orbital potential twice !
-orb should only be present in the lapwso step.
(And in fact, the lapw1 steps need to be done just once for
Hups: If this is true, you are counting the orbital potential twice !
-orb should only be present in the lapwso step.
(And in fact, the lapw1 steps need to be done just once for the
increased k-mesh; but not when changing the M-direction in case.inso)
Please check the presence of case.vorbud.
I was not clear Peter. I clarify the way we proceed.
We do runsp_lapw -eece -p -ec 0.1 -NI
Then for larger EMAX and kmesh without SO we do :
x lapw1 -up -c -orb
x lapw1 -dn -c -orb
And for the different magnetization directions we do:
x lapwso -c -up -orb
x lapw2 -so -up -c
x lapw2 -so
Just a few comments:
In many posts to this topic I saw the -c switch. This is not
necessary and prone to errors. Don't use -c . It will be done
automatically.
I think, the orb program reads case.dmat* files, and produces
case.vorb* files. It is not reading vorb files.
And it does
Dear Xavier,
One other suggestion, with a caveat. In general if the physical model
is weak/wrong or there are coding errors, the convergence is poor.
This has to do with what I call "noise" in the calculation (for want
of a better term). Since the mixing is implicitly a Simplex derivative
noise
I may annoy Peter with the comment below, so it goes
Personally I have some reservations about the consistency between the
orbital potential and time-reversal operations in -so. Three possible
tests:
a) Reduce the symmetry to P1 and run. While this is slower, I believe
your test case is
14 matches
Mail list logo
