[CP2K-user] Wrong Endpoint Energy in NEB Calculation

[Fabian Ducry]

Dear Martin,

The culprits are 
                ALIGN_FRAMES .TRUE.
                ROTATE_FRAMES .TRUE.
If you change those to
                ALIGN_FRAMES .FALSE.
                ROTATE_FRAMES .FALSE.
both endpoint energies agree:
-934.347897297       -934.316490032       -934.289348456       
-934.274499318       -934.304688658       -934.347897299

As far as I know, these two keywords are only to be used with molecules 
without periodic boundaries.

Best,
Fabian

On Saturday, 14 March 2020 14:02:47 UTC+1, Martin Brehm wrote:
>
> Dear Community,
>
> we performed a CP2K NEB calculation of a solid state system. The starting 
> and ending configuration of the band are equivalent after wrapping 
> coordinates into the primary image of the cell (just two atoms have changed 
> place, one of them moving out of the primary image). This can easily be 
> confirmed by running single point calculations of the starting and ending 
> calculations - they yield identical energies in CP2k.
>
> When we now perform the NEB calculation, the ending point unexpectedly 
> yields a different total energy (too high by 13.1 kJ/mol). The starting and 
> ending configuration cannot move during the NEB optimization, as we 
> specified "OPTIMIZE_END_POINTS .FALSE.". And the energy of the starting and 
> ending configuration remains constant during the NEB optimization - the 
> strange higher energy appears from the beginning on. So this problem is 
> probably not related to the NEB calculation itself.
>
> Attached ("image_energy.png") you can find a plot of the total energy of 
> the NEB images at some point during the NEB optimization, where the first 
> and the last value should be identical.
>
> We already checked the following:
>
> (*) We tried to simply swap the starting and ending geometry. Then, again, 
> the energy of the ending geometry is higher by the same amount (13.1 
> kJ/mol), and the energy of the starting geometry is correct this time. This 
> confirms that both configurations are identical. Probably there is a bug 
> which affects the energy calculation of the last image in a NEB calculation.
>
> (*) We were able to reproduce this error on different computer systems, 
> using different CP2k versions, from 2016 to recent versions.
>
> (*) Even the contribution of the dispersion correction to the last NEB 
> image is different than to the first one. This indicates that the issue is 
> not related to the SCF calculation, but that somehow the geometry of the 
> last image is not correct in the internal data structures.
>
> The following files are attached to this post:
>
> image_energy.png - Plot of the total energy of the NEB images at some 
> point during the NEB optimization.
> band.inp - Input file for the NEB run.
> start.xyz - Starting configuration (first NEB image).
> end.xyz - Ending configuration (last NEB image).
> coord.xyz - Again the coordinates for the "&COORD" section.
> li12si7-BAND01.out - Beginning of the Log file for the first NEB image.
> li12si7-BAND16.out - Beginning of the Log file for the last NEB image.
>
> If you want to try to reproduce this strange behavior, you do not need to 
> wait until the NEB calculation is converged. Already one or two geometry 
> steps will be sufficient to check if the first and last (fixed) images have 
> identical energies (as they should) or not.
>
> Any recommendations are welcome.
>
> Best regards,
> Martin
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200314/fa0b07d0/attachment.htm>