Dear all,
<span class='smallblacktext'>[ Edited ]</span>
puzzles
Moderators: Global Moderator, Moderator
-
admin
- Administrator
- Posts: 2922
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
puzzles
1) there must have been some differences in the runs of geometry 00. If the structure really is converged and nothing is changed in the inputs, the total energy of the ionic step which gives CONTCAR as result and the step which uses the same CONTCAR as input have to be the same (and they usually are, even if they are run on different machines).
2) concerning the steps along the barrier: if the IMAGES are converged, their total energies also have to be constant up to the convergence criterium given (i.e. 10.-3 eV in your case, the default value of EDIFFG).
3) of course, if you fully relax any of the images, the system will converge into its adjacent minimum (either initial or final state), but again, the total energy has to be the same as for the fully converged calculation of that state.
Please do the following:
1) use ALGO=Normal for all runs .
2) have a look at the forces, especially on the initial and final state geometries, to check the convergence
3) the transition state has to be interpolated from the NEB curve, it is -most probably- not exactly one of your images.
2) concerning the steps along the barrier: if the IMAGES are converged, their total energies also have to be constant up to the convergence criterium given (i.e. 10.-3 eV in your case, the default value of EDIFFG).
3) of course, if you fully relax any of the images, the system will converge into its adjacent minimum (either initial or final state), but again, the total energy has to be the same as for the fully converged calculation of that state.
Please do the following:
1) use ALGO=Normal for all runs .
2) have a look at the forces, especially on the initial and final state geometries, to check the convergence
3) the transition state has to be interpolated from the NEB curve, it is -most probably- not exactly one of your images.
Last edited by admin on Mon May 05, 2008 12:45 pm, edited 1 time in total.
-
admin
- Administrator
- Posts: 2922
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
puzzles
probably I did not.
let me summarize what you wrote:
1) for the same geometries and the same INCAR, you get total energies differing by about 2eV.
2) if you use different electronic optimization algorithms, some of your images differ in the total energy.
3) if you fully relax one of the intermediate images again, you retutrn to the initial structure, however, with an even different third total energy.
4) for a supposed transition state (if you had a look at the run without using IALGO=48, you immediately see that image2 cannot be the TS) you don't get the expected imaginary frequency in the vibration spectrum.
1) for the same converged geometry and nothing changed in the setup (as done by using the same INCAR1 twice, and hopefully running on the same machines) , the total energies have to be the same.There is no reasonable explanation why the total energy should differ by about 2eV by repeately running the same structure with the same INCAR. VASP results are always reproducible.
2) as already discussed before in the forum, ALGO=V is risky for systems with a dense eigenvalue spectrum. Please check if eigenvalues are missing in the spectrum
3) image 3 (not 2) is a possible transition state, please plot a reaction path diagram, using the results obtained after you had removed IALGO=48 and check whether image03 is the true TS or whether it is just close to the TS.
4) again, if you do a full ionic relaxation (INCAR1) on any of the intermediate images, the system has to converge back to the adjacent minimum. The fact that your total energies differ so strongly for geometries which you say are similar must be due to an error in your calculations. This behaviour has never been reported for vasp before.
it would be helpful if you made your results accessible to have a look.
let me summarize what you wrote:
1) for the same geometries and the same INCAR, you get total energies differing by about 2eV.
2) if you use different electronic optimization algorithms, some of your images differ in the total energy.
3) if you fully relax one of the intermediate images again, you retutrn to the initial structure, however, with an even different third total energy.
4) for a supposed transition state (if you had a look at the run without using IALGO=48, you immediately see that image2 cannot be the TS) you don't get the expected imaginary frequency in the vibration spectrum.
1) for the same converged geometry and nothing changed in the setup (as done by using the same INCAR1 twice, and hopefully running on the same machines) , the total energies have to be the same.There is no reasonable explanation why the total energy should differ by about 2eV by repeately running the same structure with the same INCAR. VASP results are always reproducible.
2) as already discussed before in the forum, ALGO=V is risky for systems with a dense eigenvalue spectrum. Please check if eigenvalues are missing in the spectrum
3) image 3 (not 2) is a possible transition state, please plot a reaction path diagram, using the results obtained after you had removed IALGO=48 and check whether image03 is the true TS or whether it is just close to the TS.
4) again, if you do a full ionic relaxation (INCAR1) on any of the intermediate images, the system has to converge back to the adjacent minimum. The fact that your total energies differ so strongly for geometries which you say are similar must be due to an error in your calculations. This behaviour has never been reported for vasp before.
it would be helpful if you made your results accessible to have a look.
Last edited by admin on Thu May 15, 2008 2:38 pm, edited 1 time in total.