CINEB-problem-deviated from the expected path

[itachww]

Dear Professor:
I am using vtst-CINEB to calculate the dissociation reaction path of a chlorophenol on graphene oxide. Since this is a long path, I set up 8 images between the initial state and the final state. After the initial convergence of eddifg=0.3, it further converges to eddifg=0.05. However, judging from the results obtained, the reaction path seems to have deviated from the expected assumption. I have the following questions:
1. From the energy point of view, 05image should be a structure close to the saddle point, but the frequency calculation found that the vibration direction of the imaginary frequency of the structure is not the expected OH vibration, but the vibration of the chlorine atom. Does this mean that the calculation is not positioned to the expected path? So if Dimer calculation is carried out from this structure, is it meaningless?
2. I used the idpp script to generate the reaction image points. Is my initial guess of the reaction path problematic, but besides, is there any more reasonable way to generate the initial guess of the path?
3. From the principle of vtst-CINEB, the tangent force of the highest energy point should be zero. I have a related script that can display the tangent force of each image point at each step. Then the tangent force of the 14th step of 05image is 0.00001. From this perspective, the structure should be close enough to the saddle point. So the closer the tangent force of the highest energy point is to 0, the closer the point is to the transition state. Do you think this is reasonable?
4. Will CINEB calculations locate to other reaction paths? If so, how to solve this problem?
I have attached the relevant result file, the result of rough convergence is in the folder 0.3. Thank you very much!

[graeme]

I'm guessing that you ran other calculations before the one that you posted. I'm basing this upon the initial path, which shows almost molecule dissociation and re-adsorption to the surface. I tried making a new path (attached) and it is a much smoother and shorter path.

I am concerned about your choice of IOPT=1 for a system that has many soft degrees of freedom. My suggestion is to start from a new initial path and see what happens if you use IOPT=3 , TIMESTEP=0.1. Just see if it moves towards a more reasonable path.

If nodes free up on one of our clusters, I will also try this.

[itachww]

Thank you very much for your suggestion! I did a preliminary convergence before that, and the result is in the folder named 0.3. I used IOPT=7 for pre-convergence.The initial path inside may be more reasonable. After using vfin.pl, I performed subsequent calculations.But how did you generate the new path, did you just insert fewer points and adjust it manually? Also, could I take a look at your new path for reference. I will try to use IOPT=3 to calculate!

[itachww]

I used 5 images, IOPT=3, and performed a calculation of eddfig=-0.3, but it seems that the path obtained is not much different from the path obtained by IOPT=7. I think it will not change much if I continue to converge to higher accuracy.

[graeme]

Here is the start of my calculation. I just did linear interpolation, but idpp looked very similar. This is not converged, but convergence is looking very smooth to me. There appears to be a barrier getting lower than 1.2 eV for the dissociation mechanism, and it is close to your initial path.

With very soft systems like this, with free molecules and no constraints, I find that a gentle convergence is best. In the run you posted using IOPT=1, atoms were moving too aggressively and forces were fluctuation which led to elongation in the path.

I will let you continue this calculation. If you do see trouble as the forces get lower, please let me know.

[graeme]

As I suspected, the band did converge to the specified accuracy. It remains a short path with a 1.1 eV barrier.

[itachww]

Yes professor, thanks a lot for your help! I also did the same calculation and got the same result. But I still have some doubts. For image5, I did a frequency calculation, but the imaginary frequency obtained does not involve -OH, only the slight rotation of the benzene ring, then can I consider this structure to be the saddle point of this reaction? It is disturbing that I cannot see the cleavage process of the hydroxyl group. Does it make sense to use it as a first guess for Dimer calculations? I attached the result of the frequency calculation.

[graeme]

Yes, I know what you mean - it looks the bond breaking step occurs between images. It is possible that there is insufficient resolution in the band to resolve it. If you look at the image below, you can see that an interpolation between the images (using force information) is consistent with missing the saddle.

If you do want to start a dimer calculation, I recommend using the neb2dim.pl script which will use this interpolation to estimate the geometry of the saddle and the negative mode.

Other options are to either (i) increase the resolution of the band with more images; (ii) increase the resolution around the saddle with a band from (for example) image 02 to 05.

Good luck!

mep.png (18.35 KiB) Viewed 14019 times

[itachww]

thanks for your time on this. I will try your suggestion!