External optional components of Abinit (BigDFT, Wannier90...) and codes using/providing data from/to Abinit (AtomPAW, ONCVPSP, EXC, DP, Yambo...)
Moderators: ebousquet, bxu
-
crischawo
- Posts: 7
- Joined: Thu Jan 19, 2012 10:35 am
Post
by crischawo » Sat Mar 03, 2012 5:38 pm
Dear all,
as an unexperienced user of abinit/atompaw I try to generate PAW-Potentials for La and S to do geometry-optimizations.
Using the given atompaw.input files for La and S I chose "loggrid" and then lowered the number of gridpoints until there was a difference in energy of 1mHa between all-e and pseudo. Is it necessary to stick to the same convergence-criteria (1mHa) as used in later calculations here, or what does one usually stick to?
For La (+scalarrelativistic) that was the case for "800", for S it seems to be 1500. Is there any way to get that value lower, as I want to do geometry optimizations and the calculations using that S-paw are really slow (ecut 28Ha to have no energydif. from that)?
Code: Select all
S 16
GGA-PBE loggrid 1500
3 3 0 0 0 0
3 1 4
0 0 0
c
v
v
v
v
1
1.51
n
n
VANDERBILT
2 0
1.51
1.51
1.51
1.51
2
default
0
Sincerely Yours,
Christian
-
Alain_Jacques
- Posts: 279
- Joined: Sat Aug 15, 2009 9:34 pm
- Location: Université catholique de Louvain - Belgium
Post
by Alain_Jacques » Tue Mar 06, 2012 11:18 pm
Hi Christian,
1mHa is a rule of thumb ... the important is to know if this is significant for your problem. Ecuts are generally reported at 1mHa or 0.1mHa.
Reducing the grid size will improve speed but ecut optimization is primordial. Your S PAW pseudo is weird, almost everything is in valence and there is no extra projector. As it comes from a reliable source, I wouldn't pretend it isn't a good one.
If you have time to try another version, please find mine hereunder...
Code: Select all
S 16
GGA-PBE scalarrelativistic loggrid 800
3 3 0 0 0
3 1 4
0 0 0
c
c
v
c
v
1
1.8
y
0.7
n
y
2.0
n
vanderbilt
2 0 ultrasoft
1.8
1.8
1.8
1.8
2
prtcorewf noxcnhat nospline noptim
0
I would be delighted to have a feedback about its results.
Kind regards,
Alain
-
crischawo
- Posts: 7
- Joined: Thu Jan 19, 2012 10:35 am
Post
by crischawo » Fri Mar 16, 2012 6:16 pm
Dear Alain,
thank You very much for help! Meanwhile it makes sense to me that a smaller value of r_paw leads to fewer projectors - at the cost of more planewaves/valence declarations needed to describe the excluded regions. I will report my experiences of the running geometry-optimizations.
Christian
