improved dimer method paramters tuning

[sxd375]

I have been running the TS state using improved dimer method since cl-NEB failed to converge to the Force criterion -0.05 ev on just one or two atoms, though they came out close to -0.06 ev on these but did not go down further. Hence I did vibration calculation on highest image before using the improved dimer method. However, after just a few runs the forces went even higher to -0.1 ev on these atoms(the curvature was negative though)
The INCAR parameters for DIMER are the default that I am using


# DIMER PARAMETERS
FINDIFF = 2
DIMER_DIST=0.01
MINROT=0.01
STEP_SIZE=0.01
STEP_MAX=0.1

since Forces to go down are quite close to the given criteria, I guess I need to know what parameters in the above can be tuned to very low values so that the structure remained close to the one I gave in the start and it only adjusts the atoms on which the forces are not within the criteria. Please let me know.
Thanks

[graeme]

If you can post both the neb and dimer calculation, I can take a look.

[sxd375]

Please find here attached the files for Dimer run and NEB calculation with IBRION 1( which I am running with POTIM 0.5, though I tried with very low POTIM like 0.1, 0.2 or other optimiser also). The Dimer calculation has been done on the Highest image 03 of the NEB which will be close in structure to the dimer POSCAR here since I could not find the NEB_run from where I took the dimer starting geometry.

Please let me know.
Thanks

[sxd375]

I tried attaching this file though it didnot show up. so here is the file for NEB.

[graeme]

I ran the two attached calculations. In neb.tar.gz, the NEB calculation has converged. The saddle is also quite a ways from the calculation that you attached, so I believe that yours was not sufficiently converged. The dim.tar.gz is a dimer calculation started from your NEB saddle. Again, the calculation has converged, but note that the saddle does not appear to be the same as in my NEB calculation, again, probably because it was started from a significantly different geometry.

Hopefully, with these calculations as examples, you will be able to calculate the transition states for your system. Note that I reduced some of the settings to improve convergence speed. I recommend this -- once you have a converged pathway or saddle, it is trivial to reconverge with more accurate settings. On the other hand, it can be painful, as well as unnecessary, to search for saddles with your most accurate settings.

[sxd375]

Thanks a lot for your effort.
I was stuck on this for a month.
The lowering of KPOINTS really helps in making the iterations faster.