My Community

Dear all,
When using the built-in NEB method in VASP, I noticed that increasing the number of intermediate images does not necessarily result in a finer resolution of the transition path.

In a typical example, I performed NEB calculations on the same initial and final states with two different settings: IMAGES=7 (Figure a) and IMAGES=18 (Figure b). The energy profiles are plotted below:

(Public link if figure not shown)https://imgur.com/a/4tihxkF

In both Figures a and b, each point corresponds to one image. The horizontal spacing between neighboring points represents the root sum squared distance between configuration files, calculated using the VTST script nebresults.pl.

Surprisingly, the result from IMAGES=18 does not appear more refined than that from IMAGES=7. This is because many of the added images do not reflect any significant structural transition. In fact, images 0–7 and 12–19 (Figure c) correspond mostly to trivial translations. The actual transformation occurs only within a narrow range (images 7–12), which is not much improved compared to the IMAGES=7 case.

This issue resembles a well-known problem in traditional elastic band algorithms, where intermediate images tend to slide toward the endpoints. In my case, this still happens despite the increased number of images.

My question is: Is there a way to get a more uniform distribution of the intermediate images along the actual transformation path and see more on the transition path? Otherwise, simply increasing the number of images becomes ineffective.

Thanks!

Dear kai huang,

If I understand correctly you are trying to get a denser grid between the distance of 0.5 and 1.5 in your plot. In principle you could just repeat your calculation and taking the structure at 0.5 and the structure at 1.5 as start and end point NEB calculation. Then repeat the calculation by setting IMAGE=18 in this region. Like this you should get more data points in the region of interest. Afterwards you can combine these results with the previously obtained results to get the whole structure range.

All the best Jonathan

Dear Jonathan,

Thank you for your response. I followed your suggestion, but as shown in the new results, the issue persists.

Specifically, I used the structures from image 7 (at 0.5) and image 12 (at 1.5) from the previous calculation as the new endpoints (images 0 and 17). However, after running the NEB calculation, I again observed trivial translations near the endpoints, with only 5 images (image 6~10) in the middle showing the actual transition.

This appears to be the typical issue with the standard elastic band method, where the images slide toward the endpoints. Replacing the start and end structures does not seem to resolve the problem.

Do you think increasing the SPRING constant (e.g., from the recommended -5 to -10) might help? I’ve seen a paper where the authors used a larger spring constant in the middle images and smaller ones near the ends to achieve a denser distribution in the central region of the transition path. Is it possible to implement a similar approach in VASP?

Thanks,
Kai