<html>
<head>
<meta content="text/html; charset=windows-1252"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">Dear prof. Blaha,<br>
<br>
thank you very much for the fixed symmetso; after running some
more tests I am happy to say that the jumps reduced tremendously.
Only a small jump now remains, probably connected with the CLM(R)
part of density files. Anyway, I believe that now with the new
symmetso it will be possible to continue after initso even for
more complex structures (e.g., 100+ atoms).<br>
<br>
If someone is interested the tests are described below.<br>
<br>
Best regards<br>
Vojtech<br>
<br>
<br>
As a test case I used rhombohedral antiferromagnetic FeF3 (SG
#148, R-3).<br>
<br>
Firstl, reproduced the jump with old symmetso using steps:<br>
1) start from converged hisym<br>
2) initso with (3 5 7), leave Emax at 2.5, copy all the *_so files<br>
3) runsp_lapw -p -orb -I -cc 0.00001<br>
As expected, large jump occured and subsequently the scf crashed
in 4th iteration:<br>
:DIS : CHARGE DISTANCE ( 5.5460638 for atom 2 spin
2) 3.5299069<br>
:DIS : CHARGE DISTANCE ( 5.4495877 for atom 1 spin
2) 5.3681200<br>
:DIS : CHARGE DISTANCE ( 5.3167277 for atom 1 spin
2) 5.2076606<br>
<br>
<br>
Now, using the fixed symmetso and following exactly the same steps
produced much smaller jump and converged within a few iterations:<br>
:DIS : CHARGE DISTANCE ( 0.0265515 for atom 2 spin
1) 0.0348338<br>
:DIS : CHARGE DISTANCE ( 0.0611098 for atom 3 spin
2) 0.1550994<br>
:DIS : CHARGE DISTANCE ( 0.0593840 for atom 3 spin
2) 0.1500394<br>
:DIS : CHARGE DISTANCE ( 0.1954766 for atom 3 spin
2) 0.4753107<br>
:DIS : CHARGE DISTANCE ( 0.2505027 for atom 3 spin
2) 0.5898240<br>
:DIS : CHARGE DISTANCE ( 0.1139920 for atom 4 spin
1) 0.2875368<br>
:DIS : CHARGE DISTANCE ( 0.0835918 for atom 4 spin
1) 0.2053963<br>
:DIS : CHARGE DISTANCE ( 0.0637785 for atom 3 spin
2) 0.1722495<br>
:DIS : CHARGE DISTANCE ( 0.0671636 for atom 3 spin
2) 0.1724862<br>
:DIS : CHARGE DISTANCE ( 0.0561695 for atom 3 spin
2) 0.1454023<br>
:DIS : CHARGE DISTANCE ( 0.0216516 for atom 1 spin
2) 0.0290190<br>
:DIS : CHARGE DISTANCE ( 0.0152166 for atom 1 spin
2) 0.0160854<br>
:DIS : CHARGE DISTANCE ( 0.0084897 for atom 1 spin
2) 0.0088889<br>
:DIS : CHARGE DISTANCE ( 0.0025667 for atom 1 spin
2) 0.0026985<br>
:DIS : CHARGE DISTANCE ( 0.0015687 for atom 1 spin
2) 0.0017820<br>
:DIS : CHARGE DISTANCE ( 0.0001951 for atom 1 spin
1) 0.0004541<br>
:DIS : CHARGE DISTANCE ( 0.0001265 for atom 1 spin
1) 0.0002052<br>
:DIS : CHARGE DISTANCE ( 0.0000507 for atom 1 spin
1) 0.0000872<br>
:DIS : CHARGE DISTANCE ( 0.0000190 for atom 4 spin
2) 0.0000436<br>
:DIS : CHARGE DISTANCE ( 0.0000113 for atom 2 spin
2) 0.0000173<br>
:DIS : CHARGE DISTANCE ( 0.0000069 for atom 1 spin
1) 0.0000123<br>
<br>
<br>
I am a Fortran illiterate but as far as I understand this fix in
symmetso corrects the plane-waves part of density files, while the
CLM(R) parts remain the same. So I tried a few more tests. I
converged the structure in low symmetry from scratch in order to
obtain CLMs in the final low-symmetry basis. Checked that it
reached the same energy as in the original high symmetry:<br>
:ENE : ********** TOTAL ENERGY IN Ry = -6289.92473283
(hisym)<br>
:ENE : ********** TOTAL ENERGY IN Ry = -6289.92473386
(lowsym)<br>
<br>
Then I used the fixed symmetso and on top of it I replaced the
CLM(R) of iron atoms (these are the only species where LMs were
changed, fluorines are already at lowest symmetry), i.e.:<br>
1) start from converged hisym<br>
2) initso with (3 5 7), leave Emax at 2.5, copy all the *_so files<br>
3) replace CLM(R) of irons with those from well converged "lowsym"
calculation.<br>
4) runsp_lapw -p -orb -I -cc 0.00001<br>
A further improvement, though small:<br>
:DIS : CHARGE DISTANCE ( 0.0265633 for atom 1 spin
2) 0.0348425<br>
:DIS : CHARGE DISTANCE ( 0.0579934 for atom 3 spin
2) 0.1431763<br>
:DIS : CHARGE DISTANCE ( 0.0583868 for atom 3 spin
2) 0.1447817<br>
:DIS : CHARGE DISTANCE ( 0.0700905 for atom 1 spin
2) 0.1198294<br>
:DIS : CHARGE DISTANCE ( 0.0697802 for atom 1 spin
2) 0.1168461<br>
:DIS : CHARGE DISTANCE ( 0.0299499 for atom 1 spin
2) 0.0588715<br>
:DIS : CHARGE DISTANCE ( 0.0243737 for atom 4 spin
1) 0.0569459<br>
:DIS : CHARGE DISTANCE ( 0.0147642 for atom 3 spin
2) 0.0347052<br>
:DIS : CHARGE DISTANCE ( 0.0146602 for atom 3 spin
2) 0.0352172<br>
:DIS : CHARGE DISTANCE ( 0.0149405 for atom 3 spin
2) 0.0360313<br>
:DIS : CHARGE DISTANCE ( 0.0044088 for atom 3 spin
2) 0.0101669<br>
:DIS : CHARGE DISTANCE ( 0.0023543 for atom 5 spin
1) 0.0052931<br>
:DIS : CHARGE DISTANCE ( 0.0012384 for atom 4 spin
2) 0.0031746<br>
:DIS : CHARGE DISTANCE ( 0.0003420 for atom 3 spin
1) 0.0009373<br>
:DIS : CHARGE DISTANCE ( 0.0002330 for atom 2 spin
2) 0.0005261<br>
:DIS : CHARGE DISTANCE ( 0.0001594 for atom 2 spin
2) 0.0004024<br>
:DIS : CHARGE DISTANCE ( 0.0000265 for atom 1 spin
1) 0.0000696<br>
:DIS : CHARGE DISTANCE ( 0.0000173 for atom 1 spin
1) 0.0000423<br>
:DIS : CHARGE DISTANCE ( 0.0000115 for atom 1 spin
1) 0.0000145<br>
:DIS : CHARGE DISTANCE ( 0.0000046 for atom 2 spin
2) 0.0000067<br>
<br>
<br>
I also tried playing with vorb and dmat files, but these do not
seem to make it better nor worse - probably expectable, as at
least dmat is represented in global coordinates system.<br>
<br>
<br>
<br>
<br>
<br>
On 18-May-15 13:47, Peter Blaha wrote:<br>
</div>
<blockquote cite="mid:5559D163.9010903@theochem.tuwien.ac.at"
type="cite">We recently fixed one problem in SRC_symmetso
connected with the reduction of symmetry and splitting of
equivalent atoms into no-equivalent ones.
<br>
<br>
Replace clmchange.f with the attached file and recompile.
<br>
<br>
Regards
<br>
<br>
On 05/18/2015 01:36 PM, Vojtech Chlan wrote:
<br>
<blockquote type="cite">Dear WIEN2k community,
<br>
<br>
I am facing a problem with disturbance of convergence when the
symmetry
<br>
of the structure is lowered (e.g., by initso_lapw).
<br>
<br>
It is well known in spin-polarized calculations that introducing
the
<br>
spin-orbit interaction may reduce symmetry - in dependence on
whether
<br>
the chosen direction of magnetization is compatible with present
<br>
elements of symetry or not. In spin-polarized cases where the
symmetry
<br>
is not reduced, e.g. uniaxial structure with magnetization
parallel to
<br>
the axis, the process is usually quite smooth: After well
converged
<br>
runsp_lapw and initso call, the subsequent "runsp_lapw -so"
calculation
<br>
starts almost converged and usually converges within a few
iterations
<br>
(at least when no really heavy elements are present so that the
<br>
spin-orbit coupling is rather weak).
<br>
<br>
However in cases where the symmetry is reduced during initso
(symmetso),
<br>
the continuation by runsp_lapw -so is not so smooth. According
to my
<br>
experience there is often a substantial jump in the charge
distance, the
<br>
magnetic moment may start to collapse etc. In fact, the jump in
<br>
convergence is present regardless the -so switch (or other
parameters
<br>
that are usually changed during initso, e.g., Emax).
<br>
<br>
To illustrate this behaviour I did a few tests on barium
hexagonal
<br>
ferrite (SG #194, P63/mmc):
<br>
Firstly, it was fully converged with quite standard parameters -
using
<br>
PBE-GGA+U (U=4.5eV, J=0), with RKM=6.0, 100 k-points; wien
version 14.2
<br>
was used.
<br>
Now, since there is the hexagonal axis (direction 001), starting
initso
<br>
and setting the magnetization in 001 naturally does not reduce
symmetry.
<br>
As expected, everything is nice and smooth when one starts
runsp_lapw
<br>
-so .... there is only :DIS = 0.015 in the first iteration and
the
<br>
calculation converges within a few iterations.
<br>
But when I set magnetization in direction 100, which kills half
of the
<br>
symmetry operations (and 11 non-equivalent atoms become 15), the
first
<br>
iteration starts with a jump in :DIS = 2.31. The -so switch is
<br>
irrelevant, the jump is there even for runsp_lapw without the
-so switch.
<br>
<br>
My (naive) understanding of the origin of this jump is that it
arises
<br>
from the change of basis for those atom sorts that became
nonequivalent.
<br>
The inclusion of magnetization reduces also the local point
group
<br>
symmetries of atoms (also often accompanied by change in
rotation
<br>
matrices), which sumsequently changes their lists of LM
expansions in
<br>
case.in2 file. The increase of LMs in expansion then manifests
after
<br>
first iteration also in CLM files and one gets a jump in
convergence.
<br>
<br>
When such changes concern only a few atoms in the structure or
the
<br>
change in basis is small, it seems the calculation can often be
<br>
converged despite the jump. However when more and more atoms
have their
<br>
expansions changed the jump becomes higher, eventually, for
large
<br>
structures and in cases where the magnetization strips the
structure of
<br>
almost all symmetries the jump becomes irrecoverable: the
calculations
<br>
(with or without -so) crash typically in second iteration (when
the new
<br>
LMs are first mixed) or even in first (in SELECT), in some cases
the
<br>
runs survive without a crash but the potentials go all crazy and
for
<br>
example magnetic moments collapse (anyway the convergence
fails). I have
<br>
partially learned to live with that and partially learned to
circumvent
<br>
this by reducing the symmetry not all at once, but in steps (and
<br>
re-converge in between) and by using other tricks to avoid crash
and
<br>
maintain convergence. However, currently I try to switch on the
<br>
spin-orbit in a system too large where this simply does not
help.
<br>
<br>
I can be all wrong, blaming the change in LMs, but I could not
find any
<br>
other cause for the jump. And I believe I cannot touch the LM
expansions
<br>
as they are given by the point group symmetry of the site.
<br>
<br>
I made a few naive attempts to fool some routines (mixer,
clmextrapol)
<br>
into translating the clm files into ones with a new set of LMs,
but
<br>
without proper knowledge of what I was doing, I naturally only
ended up
<br>
with nonsenses and segmentation faults.
<br>
<br>
Is there any possibility to "smoothly renormalize" the density
for a
<br>
changed set of LMs?
<br>
Well, perhaps I am blind to some completely different and
obvious
<br>
solution, so any help would be appreciated.
<br>
<br>
Best regards
<br>
Vojtech
<br>
<br>
<br>
<br>
<br>
<br>
_______________________________________________
<br>
Wien mailing list
<br>
<a class="moz-txt-link-abbreviated" href="mailto:Wien@zeus.theochem.tuwien.ac.at">Wien@zeus.theochem.tuwien.ac.at</a>
<br>
<a class="moz-txt-link-freetext" href="http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien">http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien</a>
<br>
SEARCH the MAILING-LIST at:
<br>
<a class="moz-txt-link-freetext" href="http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html">http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html</a>
<br>
</blockquote>
<br>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<br>
<pre wrap="">_______________________________________________
Wien mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Wien@zeus.theochem.tuwien.ac.at">Wien@zeus.theochem.tuwien.ac.at</a>
<a class="moz-txt-link-freetext" href="http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien">http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien</a>
SEARCH the MAILING-LIST at: <a class="moz-txt-link-freetext" href="http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html">http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html</a>
</pre>
</blockquote>
<br>
</body>
</html>