[ase-users] Jacapo convergence issues

[Niels Theis Bendtsen]

Thank you for your commnts.

I will check for other psp.

There is indeed a need for optimization run as the total forces are around 2 according to the logfile of the BFGSlinesearch optimization run.

Now just comes the question? Which pseudopotential is more correct? I mean the adsorption energies for RuO2 during OER are repoted with Dacapo and they agree reasonable well eperiments.

Now using GPAW RuO2 is worse catalyst and very a bit more from the experimental results.

Oh well guess I have to worry a bit about this. 

Niels Theis Bendtsen
Ph.d.-studerende
FYS-CAMD
DTU Fysik
Bygning 307
2800 Kgs. Lyngby
Direkte telefon 45253177
ntben at fysik.dtu.dk
www.fysik.dtu.dk



________________________________________
Fra: johnrkitchin at gmail.com [johnrkitchin at gmail.com] på vegne af John Kitchin [jkitchin at andrew.cmu.edu]
Sendt: 24. januar 2013 22:47
Til: Niels Theis Bendtsen
Cc: ase-users at listserv.fysik.dtu.dk
Emne: Re: Jacapo convergence issues

This sounds like there could be a problem with the pseudopotential of Ru. I vaguely recall Ru calculations going slowly. You may see if there are other psp available for Ru. maybe one with fewer valence electrons. An indicator something is wrong is if there are large forces on your atoms with a GPAW optimized geometry. That sounds incorrect. I would expect some forces, but not large ones.

I am not too surprised the files are large; you have a big system and a large unit cell, and a whole lot of bands. Dacapo does not parallelize well. It is pretty good over kpoints, and not very good over planewaves, so parallelization is unlikely to speed up things as much as you would hope if you are using small kpt grids.

Calculations in Jacapo should automatically start at the last position in the ncfile.


On Thu, Jan 24, 2013 at 6:33 AM, Niels Theis Bendtsen <ntben at fysik.dtu.dk<mailto:ntben at fysik.dtu.dk>> wrote:
Hi ASE user and John Kitchin (I was told to cc to you by Marcin Dulak as you were the most experienced Jacapo user)

I hope at least one you can help me. I am normally using GPAW but I have been trying to verify my results with Jacapo in which I am very new. The issue is that the calculations are very hard to converge and so far I only managed to converge a few. On top that the are very slow at least compared to GPAW.

So the system is a four layer slab of 3x2 super cell of RuO2 with doped Ni and different adsorbates on top. And they are spin polarized. I use the optimized structures from GPAW using the .traj file to get the atomic positions and magnetic moments etc. I have attached a script and the submission shell script and the associated txt output.

The pw and dw values I have taken from similar (older) calculation on a 2x2 RuO2 slab.


In the time allocated it only manages to do two geometry optimization step in 45 hours on 48 cores on niflheim.

Any tips on improving convergence or speed? Is it possible to restart the calculation from where it has ended?

I do know that my kpoint sampling might be a bit odd but it is converged at least with GPAW.

Also the associated .nc files are huge...

I hope I am doing some awfully wrong so that it can be fixed.

Sincerely
Niels Theis Bendtsen
Ph.d.-studerende
FYS-CAMD
DTU Fysik
Bygning 307
2800 Kgs. Lyngby
Direkte telefon 45253177
ntben at fysik.dtu.dk<mailto:ntben at fysik.dtu.dk>
www.fysik.dtu.dk<http://www.fysik.dtu.dk>