General questions regarding CNEB followed by dimer method protocol

[tp.joe]

Hello all,

I am very new to vtst calculation. I am looking if anyone has encountered similar system as I currently do. The system is a reaction on a surface which is fairly large and also have to take into account van der Waals force. Convergence by CNEB using 5 images took a very long time (roughly 4 days) for the same convergence criteria as typical structural relaxation (<0.02).

My question is if this image has to be reoptimized by dimer method anyway, do we really need to converge to such extent? If not then, can CNEB be used as just a rough estimate for 'MODECAR' because, in my case just a couple hour runs, it is already clear which image is the saddle point.

Thank you in advance to all expert VASP predecessors. :)

[graeme]

If you have a converged CI-NEB calculation, the climbing image will be at the saddle -- you do not need to do a dimer calculation.

You can, however, take a roughly converged NEB calculation and (possibly using the neb2dim.pl script) launch a dimer calculation from the best estimate of the saddle and negative mode. This can be faster than an NEB calculation because it only involves one image.

Finally, my most common suggestion is to use modest computational settings when finding saddles. It is trivial to increase ENCUT, KPOINTS, and the system size and reconverged upon a known saddle. On the other hand, it can be expensive if you start with highly accurate settings that you might have used for a single minimization and attempt to do an NEB calculation with them.

[tp.joe]

Thank you Prof. Graeme. Appreciated your quick reply.