NEB distances in the MEP

[var]

Hello,
in a NEB calculation the distances in the MEP between images must be the same (within 2 decimals ?). The distance in the MEP between any of the two end points and the nearest image should be the same as between images after optimization ? Initially, all the distances between nearest neighbors (image-image, image-end point) are the same number, after linear interpolation.
Thanks a lot !.
Best regards,

[graeme]

Yes, that is all correct. The precision to which the equal-distance restraint is achieved is determined by the spring constant between images and the force criterion which defines convergence.

If you use the climbing image, the equal-distance criterion is maintain on either side of the climbing image, but not across the entire band.

[var]

Many thanks, graeme ! ... I am calculating a diffusion with 3 images and SPRING = -5. After optimization the last of the 4 distances is shorter than the other (equal) 3. Should I change the SPRING ? to -4 or -6 ? I am trying both but I am not sure how to think 'weaker' or 'stronger' springs ?

Any help would be appreciated,
var

[graeme]

You should not have to change the spring constant (ever).

Are you using the climbing image? If so, this behavior is expected.

[var]

Hello graeme (and forum), thank you for your answers. I am not doing cNEB. Please, give a look at the INCAR and at the "grep prev, 01/OUTCAR" and at the "grep prev, 03/OUTCAR" and tell me what is wrong in the INCAR or in my interpretation or something else. It is the diffusion of a H2O molecule on a metal surface.

System = water Rh111 diff top-fcc
NWRITE = 2
PREC = Accurate
LREAL = auto
ROPT = 2e-4 2e-4 2e-4
ISTART = 0
ALGO = Fast
# IALGO = 48 ! 8: CG, 48: DIIS algorithm for electrons
SPRING = -5
IMAGES = 3

Electronic relaxation
NELM = 100
NELMDL = -7
EDIFF = 1e-5
ISMEAR = 1 ! 0: Gaussian, electron smearing
SIGMA = 0.200
ENCUT = 400
%NBANDS = 60

Ionic relaxation
EDIFFG = -0.05 ! - means force is less than
NSW = 200 ! # of steps in optimization (default 0!)
IBRION = 2 ! 1: quasi-NR, 2:CG algorithm for ions
#NFREE = 10 ! number of DIIS vectors to save
POTIM = 0.25 ! reduce trial step in optimization
ISIF = 2 ! 0: relax ions, 1,2:relax ions,calc stresses, 3:relax ion+cell
ISYM = 0

#LWAVE = .FALSE. ! write WAVECAR?
#LCHARG = .FALSE. ! write CHGCAR?
#LVTOT = .FALSE. ! write LOCPOT?

Parallel
LPLANE = .TRUE.
NPAR = 2
NSIM = 2


grep prev, 01/OUTCAR

NEB: distance to prev, next image, angle between 0.736455 0.736455 179.999999
NEB: distance to prev, next image, angle between 0.740643 0.739667 176.180498
...
NEB: distance to prev, next image, angle between 0.922476 0.928102 128.978894
NEB: distance to prev, next image, angle between 0.923495 0.928974 129.011998

looks fine to me, both distances (00-01 and 01-02) in the MEP are the same number, but

grep prev, 03/OUTCAR

NEB: distance to prev, next image, angle between 0.736455 0.736455 180.000000
NEB: distance to prev, next image, angle between 0.734956 0.740863 175.884477
...
NEB: distance to prev, next image, angle between 0.931261 0.425674 171.843088
NEB: distance to prev, next image, angle between 0.932211 0.423212 171.953843

the last distance (03-04) is much smaller than the other three. Of course, you noticed that initially, the four distances are equal to 0.736455

Many thanks, whether you can help me or not ...
Regards.

[graeme]

This does look fishy, but I can't tell what's going on without seeing the band.
If you tar.gz that calculation, and make it available, I'll take a look.

[xph]

The default is LCLIMB = True
You did not set LCLIMB = False, so these results are expected.