[ase-users] problem removing rotation and translation in NEB calculations

[Ask Hjorth Larsen]

Hi,

Am Mi., 28. Aug. 2019 um 15:49 Uhr schrieb Peterson, Andrew via
ase-users <ase-users at listserv.fysik.dtu.dk>:
>
> I've encountered the same thing (but with NWChem). There is no problem in how the algorithm runs, it's just that the final energy loses its information on energy and forces, which makes our standard tools for looking at NEB results break.
>
> The problem is: at each step along the NEB calculation, the `minimize_rotation_and_translation` function is called on each image except the 0th (initial state). That is, image 1's rotations and translations are minimized with respect to image 0, then image 2 with respect to image 1, etc., until the final image's rotations and translations are minimized with respect to its previous image. From ASE's perspective, that final image's atomic positions have changed, which automatically deletes the energy and force information. (Even though we "know" that the energy should stay the same and the forces should stay the same, within some rotation/translation.) By default ASE's NEB algorithm never re-calculates the final image, so in the resulting trajectory that final image never has saved energy/force results (except possibly at step 0).
>
> It's not obvious to me what the solution is. We certainly don't want to re-calculate that final image at every NEB step just to be able to make plots. Artificially adding back the energy and forces to that image also seems problematic. Perhaps we could make the plotting tools more tolerant to missing images?

I have a suggestion for a general solution: Have an object which does
the job, and only that job :)

The NEB run would return some new object called NEBPath which is the
preferred representation of the output (or an intermediate state) of
an NEB calculation.

The NEBPath contains the images and makes sure to know everything it
needs to know, and also to provide common functionality as methods: It
can plot() and read/write as JSON, and can return basic things like
the energy barrier.

That means we avoid all the circumstantial questions like whether or
not the initial and final images happen to be in the trajectory,
whether those have energy/forces.

Best regards
Ask




>
> On Tue, Aug 20, 2019 at 3:45 PM Charles-Alexandre Mattelaer via ase-users <ase-users at listserv.fysik.dtu.dk> wrote:
>>
>> Dear ASE community
>>
>> I've stumbled upon a peculiar behaviour of the NEB module when removing rotation and translation. When enabling the removal, I don't get an energy for the final image which results in no plots (when e.g. subtracting this energy from the rest to obtain relative energies) and also errors when trying to apply the NEB analysis.
>>
>> I've added example input that generate said problem (trajectory files seemed to big, even when extracting the last NEB path). No problem analyzing the trajectory where rotation and translation was allowed. I do believe that this problem occurs with other (external?) calculators as well. (Haven't tried it for this particular case.)
>>
>> It seems that rotation and translation becomes a problem during this calculation ( i see drifting of the molecule which leads to convergence issues) so it would be handy if I could properly enable this option.
>>
>> Any help would be appreciated!
>>
>> Kind regards
>>
>> Charles-Alexandre Mattelaer
>> _______________________________________________
>> ase-users mailing list
>> ase-users at listserv.fysik.dtu.dk
>> https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users
>
>
>
> --
> Andrew Peterson
> Associate Professor
> Brown University School of Engineering
> Barus & Holley 247
> 184 Hope Street
> Providence, RI 02912
> (401) 863-2153
> http://brown.edu/go/catalyst
> _______________________________________________
> ase-users mailing list
> ase-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users