NEB still not convergence

[ColinDCG]

Dear Professor Graeme,
I have been finding solutions to make NEB job convergent for a long time, but still find nothing helpful. I using the optimizer both from your group's and VASP's, but still cannot get good result. I also reading some posts on this forum and using QM at first and then turn to LBFGS or QM with smaller step, but it cannot convergence either. When I use the IOPT=3, forces on some images decline rapidly, but the middle image's force keep increasing. at the end, the middle three images' forces become larger and there is no hope to convergence.
I use 5 images, I think it maybe has something to do with the large amount of degrees of freedom, since I am calculating the phase transition barrier of MoS2，every atom is relaxed except one for reference.
The attachment was my NEB job with IOPT=3, the images of which are all below 0.5eV/Ang, Then I always turn the IOPT to 1 or reduce the TIMESTEP, but nothing work. Could you please have a look and help me check out what was wrong with my work.This is my INCAR file:

ISTART=0
ISYM = 0
ISPIN=1
ISIF=2
ENCUT = 450
PREC= Normal
ALGO=Fast
EDIFFG = -0.5
ISMEAR = 0; SIGMA = 0.05
NSW =200
IBRION=3
POTIM=0
IOPT=3
TIMESTEP=0.1
ICHAIN = 0
LCLIMB = .TRUE.
IMAGES = 5
NPAR = 2
LCHARG = .FALSE.
LWAVE = .FALSE.
LVDW = .TRUE. !switches on disp. correction (not necessary if IVDW is set)
IVDW=1 !switches between 0: off, 1: DFT-D3 and 2:TS-VDW (default=1)
VDW_RADIUS=50.0 !cutoff radius for considering neighboring imagecells
VDW_CNRADIUS =20.0 !cutoff radius for calculating the CoordinationNumber

Any suggestion will be appreciated.
Thank You
Colin Dong
East China University of Science and Technology

[graeme]

You should think carefully about the transition that you are trying to calculate. It is certainly possible to find bulk solid-solid phase transitions with the NEB, but you have set up a solid-solid phase transition in a very thin layer of a material. I think that it is very unlikely that there will be a single saddle point associated with a concerted transformation mechanism. Instead, because of the additional freedom provided by the surfaces, I expect that the path will involve many local atom rearrangements which eventually lead to the final structure that you impose.

If you really think this is a sensible calculation, then use plenty of images and do a careful relaxation using IOPT=3 and POTIM of about 0.02. Check to see if and when intermediate minima show up and minimize them so that you can break up the band into different steps. Make sure that symmetry is off - you have a fairly symmetric initial path and I expect the symmetry will break as local events show up and the overall barrier lowers.

But again, I'm not sure this calculation makes sense. You have set up a calculation that implies a simple concerted transformation of the entire 2D material, and I doubt there is a single saddle point associated with the transition.

[ColinDCG]

Thank you Professor.