initial autoneb method

Here is a description of the principle but I will gladly explain it in more detail later if relevant: So the main idea is to stepwise build up the path while evaluating only a few images at the same time in subsequent neb calculations. After each end iteration a new image is added in the gap that either improves the geometrical or energy resolution (preference around the peak). A new neb calculation is initialized on the new image and the adjacent images. This helps to add user relevant images and not images spread equally over the minimum energy path (MEP). At the same time a better guess for the geometry of the newly created image is possible by having “relaxed” adjacent images already on the MEP. This also gives some extra flexibility if you want to do a parallel neb calculation (at least with the possibilities on our cluster). If you for instance want to have 11 images on your MEP you would normally need to run 9 parallel processes and evaluate all images the same amount. Instead you can run for instance on 3 images at the same time thereby only needing 3 parallel processes. For the optional “last step” that will include the climbing image you could fully converge the forces but for the previous neb calculations while adding images you could have a less strict convergence criteria.

Afterwards it is possible to smoothen the pathway from the beginning up to the climbing image and again from after the climbing image to the end. Currently it is done by making the spring constants follow a two Gaussian distribution with centers around the peak. This keeps a higher image density around the peak and spreads the images out over the MEP path. This is only to “prettify” the path and should have no practical implications.