curvature turns to positive value - dimer

[sxd375]

Hi Dr. Graeme

I have been running dimer calculation from NEB (when the forces dropped to 0.1 ev) with IOPT 3 (deafult values )and 10^-7 EDIFF.
However as found below in the DIMCAR (a part of it), the curvature turns to positive value in the 2nd dimer step and remains positive.

1 3.20328 1.19596 -559.32186 -1.42461 1.62196
1 3.20328 1.56093 -559.32186 -1.51672 1.48015
1 3.20328 1.40739 -559.32186 -1.55160 1.64754
1 3.20328 1.58550 -559.32186 -1.52112 1.56827
1 3.20328 1.44857 -559.32186 -1.57361 1.57066
1 3.20328 1.46491 -559.32186 -1.59732 1.46059
1 3.20328 1.34865 -559.32186 -1.64307 1.44938
1 3.20328 1.33273 -559.32186 -1.66296 1.34360
1 3.20328 1.23333 -559.32186 -1.70246 1.35402
1 3.20328 1.23887 -559.32186 -1.72034 1.23610
1 3.20328 1.11012 -559.32186 -1.77015 1.20633
1 3.20328 1.08400 -559.32186 -1.78433 1.14711
1 3.20328 1.08233 -559.32186 -1.78553 1.21828
1 3.20328 1.11245 -559.32186 -1.78957 1.07243
1 3.20328 0.95861 -559.32186 -1.83143 1.11884
2 1.36951 1.18359 -559.38358 0.12926 1.09578
2 1.36951 0.98116 -559.38358 0.09646 0.87254
3 1.36890 1.12184 -559.38327 0.26832 1.60432
3 1.36890 1.37849 -559.38327 0.22727 1.01618
3 1.36890 0.85105 -559.38327 0.20411 0.84590
4 1.36938 1.08355 -559.38280 0.15492 1.05364
4 1.36938 0.97540 -559.38280 0.14382 0.99097
5 1.36976 1.29505 -559.38215 0.29377 1.23676

please let me know what I need to change here to remain near a saddle (I kept DROTmax to 20 to make sure curvature goes negative in 1st step and torque is low). I have let it run till another 10 or more dimer steps but still positive.
Also is there anyway I can restart the dimer from step 1 where it ended or I need to redo from the beginning.

Thanks a lot
I have been learning a lot from you.

[graeme]

I think you have done everything right. Unfortunately, there are times when negative modes disappear through inflection points. This indicates that your initial point is not in the basin of attraction of a saddle point. You can keep it running and it may recover, but if it wanders far away from your expected saddle, you may need to restart with a better initial guess, or perhaps continue the NEB calculation before switching to the dimer.

[Claunch]

Hello Dr. Graeme,

Are there telltale signs of when it's best to just restart everything with a better initial guess as opposed to just keep it running? I'm just curious.