Transition state calculation stop with high force using DIMER methods and IOPT=1

[madmax]

Dear all,
I am running the DIMER method to get the transition state. The ClNEB ahead converged well. But the DIMER method did not converge. The system I work with is the atoms migrate across the crystals. When I set IOPT=2, the DIMCAR was like this:
Step Force Torque Energy Curvature Angle
1 0.05495 4.30704 -242.71845 -0.05476 14.45845
1 0.05495 5.10473 -242.71845 0.47393 9.37700
1 0.05495 2.84423 -242.71845 -0.38317 8.57867
1 0.05495 3.92778 -242.71845 -0.23363 14.50821
1 0.05495 4.18419 -242.71845 -0.32989 14.50717
1 0.05495 3.99605 -242.71845 -0.51007 15.75826
2 0.05134 5.44238 -242.71854 0.07208 22.43279
2 0.05134 5.53726 -242.71854 -0.42588 21.33825
2 0.05134 5.83788 -242.71854 -0.47845 11.72439
2 0.05134 3.43825 -242.71854 -1.57824 3.29695
2 0.05134 3.81248 -242.71854 -1.54923 9.01638
2 0.05134 4.30620 -242.71854 -1.59550 6.27537
3 0.05033 6.04403 -242.71861 0.30567 7.69302
3 0.05033 5.50500 -242.71861 0.08735 4.86787
3 0.05033 5.68554 -242.71861 0.07846 3.91474
3 0.05033 6.29968 -242.71861 -0.13832 0.13438
3 0.05033 6.30260 -242.71861 -0.13641 2.15095
3 0.05033 6.26531 -242.71861 -0.12253 -0.52129
4 0.03156 6.82927 -242.71866 0.29851 -3.80973
4 0.03156 6.79572 -242.71866 0.50704 4.14649
4 0.03156 6.78895 -242.71866 0.30017 -3.88581
4 0.03156 6.79108 -242.71866 0.50769 4.19458
4 0.03156 6.79553 -242.71866 0.30762 -3.97047
4 0.03156 6.81119 -242.71866 0.33508 3.28322
5 1.14673 6.95035 -242.67833 1.49701 -4.12984
5 1.14673 6.84982 -242.67833 1.37533 -0.20904
5 1.14673 6.84428 -242.67833 1.37361 -0.10472
5 1.14673 6.84146 -242.67833 1.37319 -0.05162
5 1.14673 6.84010 -242.67833 1.37297 -0.02550
5 1.14673 6.83954 -242.67833 1.37260 -0.01372
6 2.17290 5.33810 -242.56653 1.36932 -1.31258
6 2.17290 5.34222 -242.56653 1.36167 -0.09788
It is not a good sign. And I found that the crystal near the atom changed instead of the atom move. So I decided to set the IOPT to 1, and it turned out the calculation stoped with high force. This is the DIMCAR:
Step Force Torque Energy Curvature Angle
1 0.05496 4.29851 -242.71845 -0.06107 14.48730
1 0.05496 5.08602 -242.71845 0.52032 9.36333
1 0.05496 2.85052 -242.71845 -0.36601 8.30124
1 0.05496 3.57027 -242.71845 -0.27680 12.64568
1 0.05496 4.41338 -242.71845 -0.37171 16.42338
1 0.05496 4.92712 -242.71845 -0.20891 23.79475
2 0.05496 --- -242.71845 --- ---
The INCARs are attached. What should I do, Thanks a lot.

[graeme]

It looks like the IOPT=1 calculation converged; I don't see any evidence of high forces.

Note that just because the IOPT=1 calculation converged after 1 step, this is really just because the force is sufficiently low from the CI-NEB calculation. The geometry did not change in this calculation - it just found a negative mode and quit. If you want to be more careful, you can lower EDIFFG to =0.01 and increase DRotMax to 10. If you can converge to a lower force tolerance, you can be confident that you found a saddle. If the dimer again wanders away to find a different saddle, I would be concerned that your initial state is not near a saddle.

[madmax]

Dear Professor Graeme,
Thanks a lot for your reply. In the 'IOPT=1' dimer calculation, I set EDIFF = 1E-7 and EDIFFG = -0.01. For dimer parameters, I set ICHAIN = 2,DdR = 0.005,DRotMax = 6,DFNMin = 0.01,DFNMax = 1.0,IOPT = 1. We can see in DIMCAR file that calculation stoped with FORCE larger than 0.01. I am doing the next dimer calculation with parameters "EDIFFG = -0.01, DRotMax = 0.01", and the result will be attached later. Thanks again for your reply.

[madmax]

This is the DIMCAR with EDIFFG = -0.01 and DRotMax = 10:
Step Force Torque Energy Curvature Angle
1 0.05496 4.29851 -242.71845 -0.06107 14.48730
1 0.05496 5.08602 -242.71845 0.52032 9.36333
1 0.05496 2.85052 -242.71845 -0.36601 8.30124
1 0.05496 3.57028 -242.71845 -0.27680 12.64568
1 0.05496 4.41338 -242.71845 -0.37171 16.42339
1 0.05496 4.92712 -242.71845 -0.20891 23.79476
1 0.05496 4.55659 -242.71845 -0.71167 16.50104
1 0.05496 3.25936 -242.71845 -1.17952 13.18656
1 0.05496 3.33705 -242.71845 -1.49650 9.55827
1 0.05496 3.16841 -242.71845 -1.73449 9.55862
2 0.05496 --- -242.71845 --- ---

[graeme]

Can you send me a tar file of the two calculations? I really don't understand why the IOPT=1 is terminating when the IOPT=2 is not, and why there is termination when the force is larger than EDIFFG.

[madmax]

Dear Professor Graeme,
Thanks a lot for your reply. The files about these two calculations are attached. And there are other attached files about processed POSCAR in which we constant some atoms coordinate for your analysis. Because of the unpublished work, the password will be sent to your mail, please forgive me for the inconvenience.

[graeme]

Ok, I am much happier after seeing your calculation files.

First, both IOPT=1 and IOPT=2 converge immediately after 1 iteration. The difference between the force in the DIMCAR and the OUTCAR is due to the fact that the DIMCAR includes a force projection. In both cases the true forces are the same:
FORCES: max atom, RMS 0.009719 0.002054
and below the EDIFFG=-0.01 setting. I was concerned that there was some difference between IOPT 1 and 2 in terms of convergence, but that is not actually the case. I think this is also consistent with your continuing the dimer calculations after CI-NEB calculations, which presumably also converged to the same force criterion.

The other issue raised here, about poor dimer convergence, appears entirely related to removing all frozen atom constraints. So while it was asked in terms of IOPT=1 vs 2, the calculations files show me that it is really about trying to reconverge a saddle after removing the atomic constraints. For those, the optimizer does have an effect. Your IOPT=1 calculation converges to a force of 0.008676 (again, from the OUTCAR, not the DIMCAR). The IOPT=2 calculation has not converged and the geometry may be wandering in search of a different saddle. If you really want to get to the bottom of this, even through you have converged fixed and free saddles with IOPT=1, I would again suggest allowing more rotations for the IOPT=2 setting.

[madmax]

Dear Professor Graeme,
Thanks a lot for your analysis and solution. I now totally agree with your views that it's reasonable for the process and results with IOPT=1. Thanks again for your explications. The calculating files will be reuploaded with no password after finishing the subject.

[madmax]

Dear Professor Graeme, there is still a problem puzzling me. Is this reasonable that "Torque" doesn't stop until reaching max rotating steps, in other words, doesn't converge for IOPT=1 dimer calculations? Looking forward to your reply.

[graeme]

Ah, I see. Yes, this is typical and expected behavior. It is just not worth fully converging the dimer rotation at each translation step, only to then translate the dimer and have to rotate it again. Instead, it is better to simultaneously be converging the rotation and translation, especially if you are far from a saddle. If you are close to a saddle, as you are since you are continuing from a ci-neb calculation, then it is better to spend more time with rotation so that you are certain to find the negative mode and follow that to the saddle. That is why I suggested increase the max number of rotation steps in your calculation.

[madmax]

Dear Professor Graeme,
Oh, I got your words, thanks a lot for your reply :)

[madmax]

Dear Professor Graeme,
Thanks a lot for your suggestions, I tested the parameter “DRotMax” on the dimer calculations and the calculation files have been attached. In the TS1 file, when I set DROTMAX=10, the structure reached the required convergence criteria on step2. When I set DROTMAX=30, the structure reached accuracy with just the rotation process. These two calculations are acceptable because the forces of structures both reached the required accuracy. But for calculations in the file TS3, they could not stop because the forces in the OUTCAR files are larger than 0.01. And there is a strange thing, for IOPT=2, the calculation kept running and the content in the DIMCAR file kept updating, for IOPT=1, the content in DIMCAR kept constant after reaching the rotation steps but calculations were still going on. The calculation files have been attached and the password is sent via e-mail. Looking forward to your reply.

[graeme]

Can you try uploading that file again? It doesn't look like a valid tar file to me.

[madmax]

Dear Professor Graeme,
Maybe you should use the passwords to decompress the file. The password of this file has been sent to your email. Thank you a lot for your reply.

[graeme]

Ah, yes, I did that before and forgot - thanks!