Geometry Convergence Problems

Message

aarondesk · #1 Post by **aarondesk** » Fri Jul 01, 2005 2:56 pm

Hello everyone,

I'm having a little problem with geometry convergence. I've seen this many times in my calculations. I'll typically use Ibrion = 2 (Conjugate Gradient) and my largest force typically goes to around 0.05 eV/A.

Further convergence seems difficult. The largest force will seem to oscillate around 0.05 eV/A. I've tried other algorithms with no effect. If I try Ibrion = 1 and NFREE = 12-26, they system will either oscillate about 0.05 eV/A or diverge. Any attempts with Ibrion = 3 have lead to divergence.

How can I converge my system? These are typically adsorbates on a surface.

Thank you,

Aaron

graeme · #2 Post by **graeme** » Sun Jul 03, 2005 6:14 pm

Try lowering ediff to around 1e-5. This increases the accuracy of the forces, and tends to improve the performance of the optimizers, particularly the conjugate gradient and quasi-Newton algorithms.

#3 Post by **admin** » Mon Jul 04, 2005 2:24 pm

1) it is of utmost importance to ensure that a geometry is electronically converged. (EDIFF = 1e-5 is a good choice for most purposes)
2) it may help to reduce POTIM from the default value for the first ionic steps
3) if very accurate forces are needed,
ADDGRID = .True.
(PREC = High)
can be set in addition.

aarondesk · #4 Post by **aarondesk** » Tue Jul 05, 2005 2:31 pm

Thank you for the ideas. I'll try them out.