Dear all,
I want to perform a neb calculation,and firstly I set the spring constant=0, after calculation I think all the images locate the MEP, then
I use these images as the initial images to set the spring constant=-5.0 to search the saddle point.
Because my calculation is so slow, I think this way could accelerate the calculation. It is right?
Thanks a lot.
first set spring=0, then set spring=-5.0
Moderator: moderators
I don't think that it makes sense to break up the optimization into two steps. These kind of approaches usually introduce inefficiencies.
One way to speed up the convergence of the NEB is to use our new force based optimizers. In particular, the FIRE optimizer is similar to quick-min, but tends to be a little faster, and the conjugate gradient optimizer is usually significantly faster than both of these methods.
Another strategy is to use the NEB to find the pathway, but then converge upon the saddle using a method that requires fewer images. If the band is fairly well converged, you can run a single image NEB (using the climbing image) with endpoints chosen to be those neighboring the saddle. A safer approach, particularly if you might not be close to the saddle, is to start a min-mode following (dimer) calculation from the approximate saddle. We have a script called neb2dim.pl to set up this kind of run for you.
One way to speed up the convergence of the NEB is to use our new force based optimizers. In particular, the FIRE optimizer is similar to quick-min, but tends to be a little faster, and the conjugate gradient optimizer is usually significantly faster than both of these methods.
Another strategy is to use the NEB to find the pathway, but then converge upon the saddle using a method that requires fewer images. If the band is fairly well converged, you can run a single image NEB (using the climbing image) with endpoints chosen to be those neighboring the saddle. A safer approach, particularly if you might not be close to the saddle, is to start a min-mode following (dimer) calculation from the approximate saddle. We have a script called neb2dim.pl to set up this kind of run for you.