Hello Dr. Henkelman
I am running a climbing NEB calculations on case of DOM over Pt/TiO2. I have tried differernt methods (including iopt = 0 iopt=2 and iopt=3), but I don't see any improvement. Please help me with the convergence
Attached below are the folders. Thanks in advance
CI-NEB convergence
Moderator: moderators
CI-NEB convergence
- Attachments
-
- CI-NEB.tar.gz
- (11.63 MiB) Downloaded 672 times
Re: CI-NEB convergence
This calculation is off the rails. Normally I would just redo a calculation like this and post it, but you have such a large system that I don't want to spend the computer time. So I will simply tell you what I would do and have you run it.
1. Re-relax your endpoint structures with just the bottom two layers of TiO2 frozen. I don't understand why there are random frozen Ti atoms in the topmost layer.
2. Switch to C_s, O_s, and the regular Ti (not Ti_sv) potentials. Use a cutoff of 274 eV. Set EDIFFG=-0.05 instead of -0.5
3. Remake the NEB band with 8 images. For some reason, your intermediate images have frozen atoms in different coordinates than the endpoints. I don't know how this happened, but it should not if you use our nebmake.pl script. Again, the frozen atoms (just the bottom two layers of TiO2) should be frozen in exactly the same position in every image, including the endpoints.
4. Run the CI-NEB as you did, but with 8 images using IBRION=3, POTIM=0, IOPT=3, TIMESTEP=0.1.
If that does not give sensible convergence, please post back with the calculation attached.
1. Re-relax your endpoint structures with just the bottom two layers of TiO2 frozen. I don't understand why there are random frozen Ti atoms in the topmost layer.
2. Switch to C_s, O_s, and the regular Ti (not Ti_sv) potentials. Use a cutoff of 274 eV. Set EDIFFG=-0.05 instead of -0.5
3. Remake the NEB band with 8 images. For some reason, your intermediate images have frozen atoms in different coordinates than the endpoints. I don't know how this happened, but it should not if you use our nebmake.pl script. Again, the frozen atoms (just the bottom two layers of TiO2) should be frozen in exactly the same position in every image, including the endpoints.
4. Run the CI-NEB as you did, but with 8 images using IBRION=3, POTIM=0, IOPT=3, TIMESTEP=0.1.
If that does not give sensible convergence, please post back with the calculation attached.
Re: CI-NEB convergence
Thank you very much. I will try your strategy and post the results.
CI-NEB+Dimer convergence
Hello Dr. Henkelman
I followed your strategy excately. The TS convergence is really speeded up. In order to reduce the total CPU time, I used CI-NEB (EDIFFG= -0.1) combinded with Dimer (EDIFFG= -0.03)method to search TS. CI-NEB is converged, however, Dimer method can not converge with strict value. Please help me with the convergence. Attached below are the folders. Thanks in advance
I followed your strategy excately. The TS convergence is really speeded up. In order to reduce the total CPU time, I used CI-NEB (EDIFFG= -0.1) combinded with Dimer (EDIFFG= -0.03)method to search TS. CI-NEB is converged, however, Dimer method can not converge with strict value. Please help me with the convergence. Attached below are the folders. Thanks in advance
- Attachments
-
- NEB+DIMER.tar.gz
- (2.54 MiB) Downloaded 676 times
Re: CI-NEB convergence
Great to hear that things are working better. A few comments:
If you have a converged NEB, you do not need to do a subsequent dimer calculations. The climbing image will be at the saddle if the NEB converges.
For me to see what is happening between the NEB calculation and the Dimer calculations, it would like to see the converged NEB calculation. Second, and this is a little embarrassing, but I'm not familiar with what the vasp guys coded up as far as the dimer method in IBRION=44. If you have our VTST code linked into vasp, which I think you do, then you can run the neb2dim.pl script in your completed NEB calculation and then it will get up a subsequent dimer calculation with a good estimate of the saddle and reaction coordinate. This will be using ICHAIN=2 rather than IBRION=44, see:
https://theory.cm.utexas.edu/vtsttools/dimer.html#
If you have a converged NEB, you do not need to do a subsequent dimer calculations. The climbing image will be at the saddle if the NEB converges.
For me to see what is happening between the NEB calculation and the Dimer calculations, it would like to see the converged NEB calculation. Second, and this is a little embarrassing, but I'm not familiar with what the vasp guys coded up as far as the dimer method in IBRION=44. If you have our VTST code linked into vasp, which I think you do, then you can run the neb2dim.pl script in your completed NEB calculation and then it will get up a subsequent dimer calculation with a good estimate of the saddle and reaction coordinate. This will be using ICHAIN=2 rather than IBRION=44, see:
https://theory.cm.utexas.edu/vtsttools/dimer.html#