NEB calculation, Drift importance

[alexandrosvasi]

Good day,
I would like to ask you a question regarding the importance of drift in NEB results. I am considering a hop from the minimum configuration of my system to a local minimum nearby 88meV higher in energy (indicated in the attached figure) (system TiO2, hopping ion Na)
Process: I relaxed to find energy minimum configuration with ENCUT=520 (1.3*ENMAX), ISIF=3, ISYM=0, IBRION=2, NSW=1000. Once I did, I created and relaxed start and end image with the following INCAR
ENCUT=520
ISTART=0
PREC=ACCURATE
NSW=1000
IBRION=2
LREAL=A
ISMEAR=-5
NPAR=4
ICHARG=2
ISYM=0
ISIF=2 (so that the xyz lattice parameters are locked for the minimum energy configuration coming from ISIF=3 previously)

then I created 4 images (visible in the attached figure) and tried to find NEB with the following INCAR file
-------------------------------------------------------------
ISTART=0
ICHARG=2
ISPIN=0

ENCUT=520
EDIFF=1E-5
EDIFFG=-1E-2
ISIF=0
ISMEAR=1
SIGMA=0.2
LREAL=A
LWAVE=F
LCHARG=F
NSW=5000
IBRION=1
POTIM=0.5
IALGO=48

PREC=ACCURATE
ICHAIN=0
IMAGES=4
SPRING=-5
LCLIMB=.TRUE.
LTANGENTOLD=.FALSE.
LDNEB=.FALSE.
--------------------------------------------------------------
Results: neb.dat (barriers in the third column in eV)
0 0.0000 0.0000
1 0.0000 0.1783
2 0.0000 0.3784
3 0.0000 0.7282
4 0.0000 0.5703
5 0.0000 0.0889
nebef.pl
0 0.024
1 0.023
2 0.028
3 0.034
4 0.022
5 0.028 (all below 0.05)

From OSZICAR (energy convergence in all images dE < 0.2E-05)

TOTAL DRIFT
image 1 0.09 0.015 0.52
image 2 0.07 0.067 0.59
image 3 -0.125 0.1642 -0.37
image 4 0.031 0.1498 0.045

1) I would like a comment on the actions I performed, are they reasonable?
2) How important is drift? what can I do to reduce it?
3) Can I trust this result?

Thank you very much for your time
Alexandros

[graeme]

I have a few comments. First, the fact that I don't see distances in your neb.dat makes me worried that you don't have the vtst code linked into your vasp executable. To check, see if there is a line containing VTST in your OUTCAR file. Without the vtstcode you will not be running the climbing image NEB.

Second, if you do have a low force at a saddle point, and you care primarily about the barrier, then you do not have to worry about drift. The problems that large drifts are related to long paths with poor resolution particularly around the saddle. It will also mess up our automatically generated spline that you will see in the mep.eps file. I don't see any evidence of theses problems in your calculation.

[alexandrosvasi]

Thank you very much for your quick reply,

I checked that the vtst is installed
(in OUTCAR "VTST: version 3.0c)

for some reason the last column of the neb.file did not get copied, the full neb.file looks like this.

0 0.0000 0.0000 0.0000
1 0.0000 0.1783 -0.083557
2 0.0000 0.3784 -0.103077
3 0.0000 0.7282 -0.034344
4 0.0000 0.5703 0.012718
5 0.0000 0.0889 0.000

I hope that the result is still reliable
Thank you again
Alexandros

[graeme]

Ok, that makes me feel better. I am still concerned about the column of zeros. I think that is the distance along the path, which for some reason is not being calculated. My first guess is that you may not have all of our vtstscripts in the same directory so that when the nebresults.pl script calls another script to calculate the distance between images, it is not getting anything back. Apart from that, it looks fine to me.

Oh, one comment other that I try to make whenever I can: When exploring the potential energy surface, including calculating pathways, I suggest using modest simulation parameters. I have, for example, never seen a barrier energy change significantly between Prec=Normal vs Accurate. I encourage people to use soft potentials, cutoffs around 300 eV and normal precision when calculating pathways. Remember that only energy differences matter, so you should not look for convergence in absolute energies -- this is overkill for what you care about. Modest settings allow you to find pathways much faster. When you know which barriers are important for your science, it is trivial to reconverge the saddles at a higher precision. I feel that this is important not only because it reduces the total computer time but also because I think that the community focusses a little too much on precision at the expense of not being able to afford to explore the potential energy surface around our (most often) assumed reaction mechanisms.

[alexandrosvasi]

I will put all vtst tools in one folder, I will also follow your advice regarding precision in the future
Thank you once again for your valuable help
Best Regards Alexandros

[alexandrosvasi]

Dear admin
You were correct. By adding dist.pl to the path the distances between the images are calculated normally.

I would like to ask you one final thing

Recently I tried more calculations for the same jump described in my previous posts.
Based on the vtst site I tried a Solid State Nudged Elastic Band by setting the parameters IBRION=3 and IOPT=3 along with ISIF=3. I was expecting the results to vary but what I got was extremely different.

same as original but with
ISIF=3
IOPT=3
IBRION=3
LREAL=F

results neb.dat
0 0.000000 0.000000 0.000000 0
1 0.338126 0.084463 -0.090409 1
2 0.683640 0.123269 -0.115008 2
3 1.028764 0.155797 -0.058613 3
4 1.373238 0.163789 0.007587 4
5 1.722057 0.088956 0.000000 5

nebef.pl
0 0.024075 -851.540200 0.000000
1 0.002203 -851.455800 0.084400
2 0.004797 -851.417000 0.123200
3 0.009339 -851.384400 0.155800
4 0.013188 -851.376500 0.163700
5 0.028736 -851.451300 0.088900


You can notice that not only the value of the barrier varies greatly but also appears in a different distance along the path compared to the ISIF=2 calculation. I checked the volume changes for each image and they are small. Order of magnitude --> from 1265.97 (A^3) to 1261.84 etc (starting and final are with ISIF=2 so same start and end volume)
(Drifts, Energies and Forces look good)

same as original but now start and end image are also relaxed with ISIF=3
ISIF=3
IOPT=3
IBRION=3
LREAL=F

neb.dat
0 0.000000 0.000000 0.000000 0
1 0.339882 0.098189 -0.094216 1
2 0.679564 0.136714 -0.116569 2
3 1.019120 0.169155 -0.062888 3
4 1.358417 0.177948 0.007037 4
5 1.699036 0.090211 0.000000 5
Pretty similar results with before

same as before but LREAL=A
ISIF=3
IOPT=3
IBRION=3
LREAL=Auto

0 0.000000 0.000000 0.000000 0
1 0.340500 0.033083 -0.092626 1
2 0.680627 0.071254 -0.114671 2
3 1.020399 0.103299 -0.061123 3
4 1.359914 0.111693 0.005614 4
5 1.703562 0.090211 0.000000 5

Have you ever experienced such large variations? basically with ISIF=2 nothing moves (barrier=0.72eV) but with ISIF=3 the jump is possible (barrier=0.16eV). The latter values with ISIF=3 make more sense as at the dilute limit the space offered is enough for an easy jump, also this jump is essential for a phase transition well observed experimentally.
Thank you very much
Alexandros

[graeme]

You can ignore the SSNEB calculation where the endpoints were not relaxed with ISIF=3; there is it looking for, but not finding energies for those endpoints.

For the solid state NEB, using the nebefs.pl and check the nebefs.dat file after running nebresults.pl. I can't remember exactly the form of the output but there could be an issue related to printing the barriers as eV/atom for the solid-state calculations. Anyway, see if these scripts help make sense of the output. If the volume did not change significantly, the barrier should also not change significantly.