NEB Convergence problem

[fmolajaf]

Dear Prof. Henkelman,

Thank you for developing the incredible VTST tools.
I am doing a series of CINEB calculations to study Mo Oxide/H-ZSM5 reduction under CH4, and another calculation for the formation of Mo Dimer upon condensation of two Mo monomers. In all of the NEB calculations (about 10 systems), I was able to converge the forces below 0.1 eV/A (most of the calculations converged to about 0.06 or 0.07 eV/A); however, none of them were able to converge to lower values, eddifg=-0.03 eV/A which is the required accuracy for my calculations. I have started almost all of the calculations (5 intermediates) from IBRION=3, IPOT=3, and POTIM=0, and all converged fairly reasonably to values between 0.06-0.08 eV/A, suddenly when I continue the calculations (just using CONTCAR as POSCAR without any changes in the INCAR) for example from 0.06 to 0.05 eV/A, forces go to high values like 0.3 eV/A. This is the case in almost all of my calculations, and I was trying to change settings according to your advice in the forum, like going to IOPT=1 with INVCURV = 0.001, but it made forces even worse. I would appreciate it if you please help me understand how I converge to lower values.

Thank you in advance.

#INCAR (Converged calculations to 0.065 eV/A)

SYSTEM = HZSM5
PREC = Accurate
ENCUT = 500
ALGO = Fast
ISMEAR = 0
SIGMA = 0.05
# parallelization
LPLANE = .TRUE. # distribute real-space mesh
#NPAR = 8 # number of cores per computing node
LORBIT = 11
LREAL=.FALSE.
#NELECT = 50 #7*nions+1
ISPIN = 2
MAGMOM = 298*0
# relax
ISIF = 2 # 2=atoms, 3=2+lattice, 4=shape+ions
IBRION = 3
NFREE = 2
NELMIN = 4 # minimum number of electronic steps
NELM = 150
EDIFF = 1E-6 # SCF energy convergence 1E-7
EDIFFG = -0.065 # max force for ionic steps -0.01
NSW = 800 # max ionic steps
NUPDOWN = 0
IVDW= 11
# outputs
NWRITE = 1 # {0|1|2} default:1
LCHARG = .FALSE.
LWAVE = .FALSE.
LAECHG = .FALSE.
IMAGES = 5 # Number of NEB images between the fixed endpoints
SPRING = -5.0 # The spring constant, in eV/Ang^2 between the images; negative value turns on nudging
ICHAIN = 0 # Indicates which method to run. NEB (ICHAIN=0) is the default
LCLIMB = .TRUE. # Flag to turn on the climbing image algorithm
#LNEBCELL = .TRUE. # Flag to turn on SS-NEB. Used with ISIF=3 and IOPT=3
# Must set IOPT = 3 or 7 when using LNEBCELL=.TRUE.
IOPT = 3 # QM (Quick-Min) force based optimizers is used and good for high force far from minmum
# O(default vasp),1(LBFGS),2(CG),4(SD),7(FIRE)
#IBRION = 3 # Disbale VASP default optimizers which are basesd on Energy
POTIM = 0 # Disbale VASP default optimizers which are basesd on Energy
MAXMOVE = 0.2 # Maximum allowed step size for translation(Default comes with IOPT=3)
TIMESTEP = 0.01 # Dynamical time step (Default comes with IOPT=3)
LSCALAPACK = .FALSE.
NCORE=8
NSIM=8

[graeme]

It does not sound right that the forces should increase between runs. The only reason I could think of when this may happen is if the cell is able to relax, but I see that you have ISIF=2, so that is not an issue. If you can post an example of both the original calculation and the restarted one, I can take a look. I can probably also suggest changes to improve convergence.