GPAW with TSASE

Python framework for working with atomic configurations, potentials, and algorithms.

Moderator: moderators

Post Reply
glenjenness
Posts: 2
Joined: Fri Apr 03, 2015 3:38 pm

GPAW with TSASE

Post by glenjenness » Fri Apr 03, 2015 3:44 pm

I am currently trying to use the dimer routines in TSASE to calculate some reaction barriers with GPAW. However, I run into a small problem wherein I can reach the force convergence, but the curvature is positive. Thus, what I end up with is something that looks like:

4 0.139406 -1.507605 -443.900317 16
alpha=vmag(Fperp)/vmag(F0): 0.946011
curvature: 1.396486
drag up directly
Ftrans 0.110369451764
5 0.046732 1.396486 -443.898168 20
alpha=vmag(Fperp)/vmag(F0): 0.991729
curvature: 0.548309
drag up directly
Ftrans 0.0455703394463
6 0.021699 0.548309 -443.896273 24
alpha=vmag(Fperp)/vmag(F0): 0.988442
curvature: -10.372543
Ftrans 0.353999654585
7 0.154144 -10.372543 -443.901679 28


I am using the internal search function that is included with the ssdimer class.

As anyone else seen a similar behavior?

xph
Posts: 38
Joined: Tue Mar 13, 2012 9:33 pm

Re: GPAW with TSASE

Post by xph » Fri Apr 03, 2015 6:34 pm

The dimer wouldn't stop if the curvature is positive. In your post, the last step (step 7) has a negative curvature. You can set a tighter convergence criteria, for example 0.03, to let the dimer run more steps.
For the first few steps the curvature is positive and the force is small, which indicates the initial guess is very close to a minimum.

glenjenness
Posts: 2
Joined: Fri Apr 03, 2015 3:38 pm

Re: GPAW with TSASE

Post by glenjenness » Sat Apr 04, 2015 2:30 pm

A tighter convergence criteria won't help. It's already at 0.05 eV/Ang, and while it'll dip below 0.05, the curvature won't be negative, and thus it keeps cycling through.

xph
Posts: 38
Joined: Tue Mar 13, 2012 9:33 pm

Re: GPAW with TSASE

Post by xph » Sat Apr 04, 2015 9:08 pm

In that case, you may want to push the initial guess a little further from the reactant minimum. It is better to start outside the concave region to get rid of those concerted mode.

In sum, try a better initial guess. At the same time, set a tighter rotation convergence: rotationMax = 20, phi_tol=3

Post Reply