ISYM=0 for endpoint calculations?

[yuka]

Dear Prof Graeme,

I have read that you recommend ISYM=0 for NEB calculations. I have recently started doing this. However, I would like to know if the endpoints need to have been converged with ISYM=0 as well?

I believe it should not affect the convergence of the band, however, my primary concern is whether this would affect the dimer calculations since the starting MODECAR (from neb2dim.pl) depends on the results of nebresults.pl, which uses also the energies of the start and end points of the NEB calculation. My understanding is that if the images have different energies compared to the endpoints due to the different settings, this might cause neb2dim.pl to choose a wrong start point for the dimer and MODECAR, is this right?

Thanks!

[yuka]

In addition, could I also quickly ask if it is certain that images with angles close to 90 are not going to converge, and so once I see that extra images should be added immediately? How about images with angles less than 90 e.g. ~50?

[graeme]

I would say that ISYM should only be used to speed up the convergence of known symmetric structures. In any case where the structure would relax and break the symmetry, ISYM should be turned off. It is just like using ISPIN=1 to restrict the spin; in any case where spin polarization is required, the imposed symmetry should be turned off. So yes, for the minima as well.

Turning symmetry off for the dimer and NEB calculations is particularly important because there can be structures very close to symmetric, which will not be allowed to break the symmetry if ISYM is on.

In regards to the NE question, I think it depends upon the situation. One reason for small angles along the path is the presence of intermediate minima. For that case, minimizing to find the minimum and doing separate band for the segments along the path is most efficient. Second, check to see if the band is converging, or if the high curvature along the path (small angles) is due to poor convergence. If so, a more conservative optimizer is recommended. Finally, and only if the band looks reasonable apart from a poor resolution of the path, would I add more images.

[yuka]

Thank you for your reply! I think I will put some time to recalculate my initial states and final states with ISYM=0.

The reason I ask the second question is because I have been trying to get the TS for this reaction whereby the band is converged to about 0.1eV/A but then does not go any lower. I then noticed that the angle for the image near the saddle point
(image 2) is a very worrying 90.x degrees.

By looking at the movie, the path seems to be reasonable. There also does not seem to be intermediate minima from mep.eps. The 90 degree angle at image 2 I would suppose is due to the abrupt shift in the position of the hydrogen atom from image 2 to image 3. It thus seems to be that I would either need to add more images, or change the position of my hydrogen in the initial state. Would this be a problem if I continued to further optimize the band?

If you had some time, could you take a look and see if what I am saying is reasonable? Thanks!

[graeme]

The path does look reasonable, and long enough that it could use a few more images.

One issue that I see is that your frozen atoms should be the same in your initial and final configurations.

[yuka]

Thanks! I will try out either more images or switch to the dimer method to see how it goes.