NEB optimization fluctuating

[jasius]

After I setup NEB calcs and they reached the maximum energy (in all images) they started oscillating in energy and didn't really converge, although were close. What might be an issue here?

I attached the typical image close to TS but they all ended up not converging

thanks

Jonas

[graeme]

It would be helpful to see all the images. A .tar.gz file without CHG* WAV* files should not be too large.

My first approach to debugging this would be to tweak the optimizer. There are two approaches. One is to use IOPT=3 or 7 with TIMESTEP=0.01. Check for systematic (and slow) convergence, and no more oscillations. Second, you can reduce the INVCURV parameter for IOPT=1 (LBBFGS). Try a small value such as INVCURV=0.001, and again, look for smooth convergence. If you have stiff modes in the system, you may need this more conservative setting.

If there is some other problem, I could probably find it with more data from the calculation.

[jasius]

I will try your suggestion. I also attached the images. Besides the fluctuation, a conceptual question, albeit not very relevant to the software is: how do I create this planar CH3 TS. As it is now, it's simply CH3 shift/transfer, it should be planar CH3 as in SN2 type mechanism

Jonas

[graeme]

I started from a linear band with a conservative optimizer and it looks fine to me (see attached). There is a well-converged saddle in run3/05 with a barrier of 2.59 eV.

You will probably notice that there are shallow minima at the start and end of the path. I recommend minimizing images 02 and 06 to see if there are lower energy initial and/or final states and running a band between those. It will be hard to minimize a band with intermediate minima using only 8 images. It will be better to understand what is going on, and break it up into any separate paths that are relevant.

I'm sure that you can play with the higher order optimizers and get faster convergence.

[jasius]

much appreciated. my attempts with INVCURV=0.001 ran out of cycles