I'm not quite sure I understand what you are saying, so correct me if
I'm wrong :
- You tried my input file, and found a very very slow convergence (the
same as me).
- To see the difference with ecutsm, you tried to remove ecutsm from my
input, and launched it. You also got a very very slow convergence
(=> that's why you say there is no big difference with ecutsm). Then
I don't understand why we get different results.
I tried to inprouve my calculations as you told, and as soon as these
calculations with a denser kpt coverage and more bands are finished, I
will let you know. Still, I just need to remove ecutsm to get a good
convergence...
As for the primitive cell, anyhow, the calculations I'm interested in
don't have symetries, but I agree with you that it is much faster.
Thanks
Emmanuel ARRAS
Josef W. Zwanziger a écrit :
461275.54164.qm@web1111.biz.mail.sk1.yahoo.com"
type="cite">
Hi,
I still don't get the big difference with ecutsm, but I DO get
very very slow convergence if I run the same paramters as you. Try this:
1) Much denser kpt coverage (your choice of ngkpt leads to
kptrlen of only about 32, which is not much, try it with kptrlen of 50
or 60 and let abinit find the kptrlatt for you)
2) This helps even more: add more bands. Your default band
number is only slightly over the set of (nominally) full bands, I found
that if I increased nband to say 14 (so there are 6 empty bands) then
convergence is MUCH faster, everthing was great in 16 steps with ecutsm
= 0.0 or 0.5.
3) Why are you not running the primitive cell? It's so much
easier and faster....
Joe
Josef W. Zwanziger
Professor of Chemistry
Canada Research Chair in NMR Studies of Materials
Director, Atlantic Region Magnetic Resonance Centre
Department of Chemistry
Dalhousie University
6274 Coburg Road
Halifax, Nova Scotia B3H 4J3 Canada
tel: +1.902.494.1960
fax: +1.902.494.1310
web: http://jwz.chem.dal.ca
jzwanzig@jzwanzig.org,jzwanzig@dal.ca
From:
Emmanuel Arras <emmanuel.arras@cea.fr>
To: forum@abinit.org
Sent: Monday, December
15, 2008 6:52:34 AM
Subject: {Spam?} Re:
[abinit-forum] ecutsm and electronic convergence
I really don't get the same results as you. Even if I use ecut 15 Ha
and pawecutdg 40 Ha, I still get the same big difference between with
an without ecutsm. I give you my input file. I'm using the pseudo given
on the abinit site
( http://www.wfu.edu/~natalie/papers/pwpaw/periodictable/atoms/Fe/index.html),
abinit 5.6.4, and the band-fft parallelisation. Also, my BCC box is not
primitive.
Don't you get the same results with my input?
Also, if I use nsym 1 (on the cubic box), the convergence is even worse.
Thanks for your help.
Emmanuel ARRAS
Josef W. Zwanziger a écrit :
I
used ecut 15, pawecutdg 40, ecutsm 0.0 and 0.5. And, in accord with
Matthieu's remarks, I find magnetic BCC iron to converge slower than
nonmagnetic insulators and metals. What often happens for me is that
convergence seems to proceed normally for some steps (say 20) then get
worse again for awhile, then abrubtly get much better and complete.
However, I don't find any dependence on ecutsm to the convergence rate.
Joe
Josef W. Zwanziger
Professor of Chemistry
Canada Research Chair in NMR Studies of Materials
Director, Atlantic Region Magnetic Resonance Centre
Department of Chemistry
Dalhousie University
6274 Coburg Road
Halifax, Nova Scotia B3H 4J3 Canada
tel: +1.902.494.1960
fax: +1.902.494.1310
web: http://jwz.chem.dal.ca
jzwanzig@jzwanzig.org,jzwanzig@dal.ca
From:
matthieu verstraete <matthieu.jean.verstraete@gmail.com>
To: forum@abinit.org
Sent: Friday,
December
12, 2008 8:27:24 PM
Subject: Re:
[abinit-forum] ecutsm and electronic convergence
Hello,
I notice Emmanuel's pawecutdg is very low (22.5). Are you sure it is
converged, and in particular that it is still converged once you add
ecutsm (I think it is affected by ecutsm the same way normal ecut is -
the contribution of those plane waves with energies in the last ecutsm
Hartrees are smoothly cut off to 0). As it should be around, and
sometimes higher than, 2*ecut, subtracting ecutsm=0.5 from it may
change convergence. Try with pawecutdg 40 to see. Also, do all your
runs converge to the same energy? You need a reference too, with higher
ecut and pawecutdg=2*ecut. Joe - which values did you use?
On a side note, magnetic systems are often hard to converge, and can
have multiple local minima, unlike normal DFT. Non-collinear magnetism
is even worse.
Matthieu
On Fri, Dec 12, 2008 at 11:03 PM, Josef W.
Zwanziger <jzwanzig@jzwanzig.org>
wrote:
Hi,
I can't repeat what you are finding. If I explicitly break the BCC
symmetry of iron by straining the sample, it doesn't change the
convergence with respect to the value of ecutsm at all. I have also
used ecutsm to optimize cell geometries in lots of non-cubic cases, and
never had a problem like what you are describing. I'm stuck. Maybe
someone else can be more helpful?
Joe
Josef W. Zwanziger
Professor of Chemistry
Canada Research Chair in NMR Studies of Materials
Director, Atlantic Region Magnetic Resonance Centre
Department of Chemistry
Dalhousie University
6274 Coburg Road
Halifax, Nova Scotia B3H 4J3 Canada
tel: +1.902.494.1960
fax: +1.902.494.1310
web: http://jwz.chem.dal.ca
jzwanzig@jzwanzig.org, jzwanzig@dal.ca
Sent: Friday,
December 12, 2008 2:40:23 PM
Subject: Re:
[abinit-forum] ecutsm and electronic convergence
Ok, I think found the problem. In fact, it happens when the box is
non-isotropic :
Here are the plots of the electronic convergences ( deltaE(h) ) vs SCF
steps for 4 calculations :
Fe-BCC_ecutsm_iso : smearing = 0 and 1, on a cubic box (acell 3*2.9
angstrom)
Fe-BCC_ecutsm_aniso : smearing = 0 and 1, on a non-cubi box (acell
2*2.9 3.1 angstrom)
the output given is the one for non-cubic box and smearing = 1.
What do you think?
Emmanuel Arras a écrit :
I will try this on Fe BCC. Maybe it is
linked with my pseudo, or my
system (which is quite specific).
Thanks for your response. I'll keep you posted.
Emmanuel ARRAS
Josef W. Zwanziger a écrit :
This is odd. I've done lots and lots of PAW calc's, with ecutsm and without, on magnetic and non-magnetic systems, and have never observed anything like what you're describing. Can you post your input file?
Josef W. Zwanziger
Professor of Chemistry
Canada Research Chair in NMR Studies of Materials
Director, Atlantic Region Magnetic Resonance Centre
Department of Chemistry
Dalhousie University
6274 Coburg Road
Halifax, Nova Scotia B3H 4J3 Canada
tel: +1.902.494.1960
fax: +1.902.494.1310
web: http://jwz.chem.dal.ca
jzwanzig@jzwanzig.org,jzwanzig@dal.ca
----- Original Message ----
From: Emmanuel Arras <emmanuel.arras@cea.fr>
To: forum@abinit.org
Sent: Thursday, December 11, 2008 1:27:40 PM
Subject: [abinit-forum] ecutsm and electronic convergence
Dear abinit users,
I have observed that the use of ecutsm (typically 0.5 Ha for a cutoff
of 10 Ha, ecut CV around e-2 eV/atom) makes it very difficult for abinit
to achieve convergence : i.e. over 200 iterations to reach e-8 in toldfe
(despite a steady slope), as compared to the 30 usually needed
iterations. I am wondering where this comes from, and how to solve this
issue.
Note that I am using paw, and that my system is magnetic. It seems in
fact that magnetism is hugely affected by this ecutsm : it takes much
more time to reach final value (whereas magnetism is usualy very quick
to converge). Also note that whatever the cutoff (up to 20 Ha, which is
huge in paw), this problem appears.
Thanks for any advices.
Emmanuel ARRAS
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr. Matthieu Verstraete
European Theoretical Spectroscopy Facility (ETSF)
Dpto. Fisica de Materiales,
U. del Pais Vasco,
Centro Joxe Mari Korta, Av. de Tolosa, 72, Phone: +34-943018393
E-20018 Donostia-San Sebastian, Spain Fax : +34-943018390
Mail : matthieu.jean.verstraete@gmail.com
http://www-users.york.ac.uk/~mjv500
|
