Problem in convergence

Vasp transition state theory tools

Moderator: moderators

Post Reply
zhangj
Posts: 12
Joined: Mon May 21, 2007 1:54 am

Problem in convergence

Post by zhangj »

Dear Sir,

I run my job with POTIM=0.01 unitil the forces drop to about 0.5eV/Ang, then increase POTIM =0.1; also I run the job with POTIM= 0.01 all the time. However, after 5000 steps both jobs have not converge yet. I find it is awkward to visualize the OUTCAR and the structure goes crazy. The adsorbed oxygen atom move into the suface and for the surface the relaxed top three layers are in a mess.
I read the VASP manual, it seems that the problem in reading the final results is caused by the fake forces and the fdake forces make the images drift against each other.
I wonder if my problem is due to fake forces.
In VASP manual, it is mentioned that making sure the sum of all positions is the same for each cell will avoid fake forces. I don't quite understand it. Now I don't know how to continue. Should I fix all surface atoms and only relax adsorbates?


Thank you for your help
Best Regards
Zhang Jia
graeme
Site Admin
Posts: 2256
Joined: Tue Apr 26, 2005 4:25 am
Contact:

Post by graeme »

This sounds like a more basic problem than any fictitious forces. Have you visualized a movie of your initial band, and the one with forces of 0.5 eV/Ang? You could use the neb_movie.pl script to generate an xyz movie, and see if there are problems with the geometry.

There is not enough information to tell what is happening, but start by making sure the endpoints are relaxed, and that the initial band looks reasonable. If you still run into problems, send a .tar.gz file of the calculation.
zhangj
Posts: 12
Joined: Mon May 21, 2007 1:54 am

Post by zhangj »

Dear Sir,

Thank you for your reply.

I tried to interpolate three images between start and final state. It seems that the energy and force are reasonable.
First, I run the job with POTIM = 0.01 for 300 ionic iterations
The "nebef.pl" script give
0 0.03601800 -121.36430300 0.00000000
1 0.04227000 -120.69557800 0.66872500
2 0.01866100 -121.71952400 -0.35522100

Then I continue this job with POTIM= 0.1 for another 300 ionic iterations, however, the job has not converge yet. Also, I find the forces go up when POTIM increases to 0.1. Is it normal?
The "nebef.pl" script give
0 0.09458200 -121.39132500 0.00000000
1 0.07651500 -120.70300700 0.68831800
2 0.02153700 -121.72516900 -0.33384400

I've sent the related .tar.gz files of my run to you by email.
Thanks a lot!
Zhang Jia
graeme
Site Admin
Posts: 2256
Joined: Tue Apr 26, 2005 4:25 am
Contact:

Post by graeme »

These calculations look reasonable to me. I don't see any craziness in the surface atoms.

The forces are fairly low, 0.04 eV/Ang at the saddle. Since the band is reasonable, and fairly well converged, you could get faster convergence using vasp's quasi-Newton (POTIM=1), or our second order, force-based optimizers: http://theory.cm.utexas.edu/vtsttools/optimizers/ , but I would not expect the barrier to change significantly.
zhangj
Posts: 12
Joined: Mon May 21, 2007 1:54 am

Post by zhangj »

Thank you very much !
I'm using CI-NEB method, so I don't need run a dimmer calculation to confirm saddle point, right? I read your previous post, you mentioned " The climbing image takes away the need to increase the density of images near the saddle -- there only needs to be enough points in the path to resolve the reation coordinate at the saddle." ( The topic is "Several maxima and minima with CI-NEB").
I wonder how many images in the path are enough to get saddle point.

Thanks again.

Zhang Jia
Post Reply