How Does VTST Define Saddle Point When There Are Multiple Saddle Points

[JiQi535]

Dear developer,

I would like to ask how the VTST would try to get to convergence when there are multiple saddle points in a NEB path. For example, a path can have a "M" shape, which has two maximum energy points and one local minimum energy point. Learning from the website documentations, I know that the CI-NEB is trying to find the point with maximum energy. Does it try to find maximum energy for all images or just one? I believe it is just one. Then how would the CI-NEB pick the one point to climb if there are multiple local peaks to climb?

Meanwhile, I see that a CI-NEB calculation converges when the maximum atomic force on the saddle point image is smaller than convergence criteria we set, while the maximum atomic force on the other images doesn't need to be converged. Is that correct? In this case, the energy for the local minimum might not be that accurate, as its force might not be well converged. Am I correct?

Thanks very much for your help, and please direct me to existing posts if any.

[graeme]

Yes, I think that all of your statements are correct.

For a path with intermediate minimum, it is best to minimize a nearby image to get an accurate energy of the intermediate minimum. You can also then run an NEB between the intermediate minimum and a final state to find the energy of a lower energy saddle.

[JiQi535]

Hi Prof. Henkelman,

If I understand correctly, when there are multiple local peaks in the path, CI-NEB will only identify one peak with the highest initial energy and climb that image. I think this is reasonable, as image with highest initial energy can be assumed to be the most close to the actual saddle point.

Thanks very much for your confirmation of my understanding and suggestion about how to find accurate energy of intermediate minimum.