I'm wondering if this idea is correct...
My idea is: if I do a climbing-image-NEB run with only 1 image.... then I would get an activation energy that is a lower bound to the activation energy I would get from a climbing-image-NEB run with many images on that same system.
In other words, the 1 image [climbing image run] would climb to the nearest barrier/activation energy/saddle point and give the energy of that. But if, for example, before "reaching the final state", there were more barriers to come later, making it a multi-step reaction.... well, then, the single climbing image might have happened to climb to a barrier that is not the highest barrier. In this case using more images would make it more likely that the image that gets assigned to climb will get pushed to what happens to be the highest barrier. (My understanding is that the climbing image NEB algorithm takes the highest energy image after the first electronic relaxation and assigns that to be the image that climbs).
To say this whole idea another way, for climbing image NEB: the less images you use.... the more you introduce the possibility that the climbing image will give you a barrier that is a lower bound barrier but not the highest barrier (i.e. a barrier that couldn't be higher than the highest barrier, obviously, but might in fact be lower). And, by using more images, you increase the odds that the image that gets assigned to climb happens to ascend to the max height barrier (the rate limiting step barrier). Of course, if the reaction is a single step reaction (only one barrier) then 1 climbing image would give you the activation energy and no further images would be needed. (But I better be sure there is only 1 barrier along the way because there is always a gamble I'm missing a higher barrier -if it is a multi-step mechanism- by using only 1 image on the climbing image routine).
Is this all true?
Thanks!!!
climbing image NEB
Moderator: moderators
Re: climbing image NEB
There could be something in this, but there is an important issue to understand.
The major concern about using a single image is that the low resolution of the path may not give an acceptable tangent at the (or a) saddle along the path. In this case, the climbing image many not converge. If it does converge, however, I agree that it is a lower bound to the highest saddle along the (or a) path between reactants and products. Perhaps not the activation energy since there may be intermediate states which are more stable than the reactants into which the system may get trapped.
Second, for pathways with multiple barriers, it is typically more efficient to break the path up into calculations between the intermediate minima. Using many images for such paths is a little unwieldy. For paths with multiple steps, you also have to consider the relevant kinetics -- do you know that the system really takes all these steps in sequence, or might it go somewhere else, or to the specified products in a different way. Using a band with many images may not be the best way to get at the kinetics.
The major concern about using a single image is that the low resolution of the path may not give an acceptable tangent at the (or a) saddle along the path. In this case, the climbing image many not converge. If it does converge, however, I agree that it is a lower bound to the highest saddle along the (or a) path between reactants and products. Perhaps not the activation energy since there may be intermediate states which are more stable than the reactants into which the system may get trapped.
Second, for pathways with multiple barriers, it is typically more efficient to break the path up into calculations between the intermediate minima. Using many images for such paths is a little unwieldy. For paths with multiple steps, you also have to consider the relevant kinetics -- do you know that the system really takes all these steps in sequence, or might it go somewhere else, or to the specified products in a different way. Using a band with many images may not be the best way to get at the kinetics.
Re: climbing image NEB
I have a question about this part of the response:
"The major concern about using a single image is that the low resolution of the path may not give an acceptable tangent at the (or a) saddle along the path. In this case, the climbing image many not converge."
In this particular context when you say "converge" do you define converge to mean: when the VASP run succeeds in coming within the EDIFFG that was set and the run finishes successfully with a complete OUTCAR?
If that is what you mean, then I believe that: as long as the run finishes successfully to within the set EDIFFG (without error, without hitting walltime etc.), it follows that -at least for that particular run- the 1 image low resolution issue did not exist/did not cause a problem... and a true saddle point was found, even though there was only one image.
Is that correct? Or, if that is not correct, then perhaps: OUTCAR can be complete, VASP can come within the EDIFFG tolerance and even still a one image climbing image run can produce an end result where that one image is not at a saddle (due to the low resolution issue).
Thanks in advance for letting me know which is the right way to understand this!!!
"The major concern about using a single image is that the low resolution of the path may not give an acceptable tangent at the (or a) saddle along the path. In this case, the climbing image many not converge."
In this particular context when you say "converge" do you define converge to mean: when the VASP run succeeds in coming within the EDIFFG that was set and the run finishes successfully with a complete OUTCAR?
If that is what you mean, then I believe that: as long as the run finishes successfully to within the set EDIFFG (without error, without hitting walltime etc.), it follows that -at least for that particular run- the 1 image low resolution issue did not exist/did not cause a problem... and a true saddle point was found, even though there was only one image.
Is that correct? Or, if that is not correct, then perhaps: OUTCAR can be complete, VASP can come within the EDIFFG tolerance and even still a one image climbing image run can produce an end result where that one image is not at a saddle (due to the low resolution issue).
Thanks in advance for letting me know which is the right way to understand this!!!
Re: climbing image NEB
You are right; If the calculation finishes in the sense of the climbing image reaching the required ediffg, it has found a saddle to within the convergence criterion. If the tangent is too poorly defined, the force on the climbing image may never drop to the convergence criterion.