UT theoretical chemistry code forum

Hi, I met with a weird problem when I used NEB to calculate the barrier of a surface reaction. I follow the procedures of NEB exactly. 1. use nebmake.pl to generate some imgaes. 2. use nebmovie.pl to see whether these images are reasonable. 3. run vasp. 4. check the images again using nebmovie.pl.

The initial movie.xyz was ok, all the images looked reasonable. But after tens of steps, the images excluding the two endpoints became totally distorted, which led to the slowdown of convergence. Why this happened? I met with this problem before, but found no solution.

This is my INCAR:
SYSTEM = Silicon (100)-(2x1) and (111)-(2x1).(2x2) surface
Startparameter for this Run:
NWRITE = 2 verbosity write-flag
LPETIM= F write-flag & timer #
ISTART = 0 0-new 1-cont 2-samecut job
ISPIN = 1 1-no, 2-yes
INIWAV = 1 0-jellium WF; 1-random numbers (recomm.)

Electronic Relaxation
! ENCUT = 400.00 #Ecut (eV)
PREC = Medium * #precission: #Low/Medium/High
NELM = 150 max. # of electronic SC steps
NELMDL = 5 # of non-consistent steps at the beggining:>0 or <0
EDIFF = 1.0e-4 stopping-criterion for electronic upd.
VOSKOWN = 1
ISMEAR = 0
SIGMA = 0.1
AMIX = 0.2
BMIX = 0.0001

Ionic relaxation
EDIFFG = 1.0e-3 stopping-conditions for ionic relax.loop (EDIFF*10)
NSW = 250 * # of steps for ionic update (def:0)
IBRION = 2 * ionic relax: -1:no move; 0-MD;1-Newton;2-CG;3-damped;4-?
ISIF = 2 * calc.stress:0-no;1-tr;2-7 yes; 3-change vol.4-change shape
! ISYM = 2 symmetry:0-OFF, 1-ON,2-ON+memory conserving for CD
NBLOCK = 1
POTIM = 0.3 if IBRION=1,2,3:scaling constant forces;IBRION=0,time st.


! LPLANE = .TRUE.
! NPAR = 8
IALGO = 48 * algorithm: use only 8 (CG) or 48 (RMM-DIIS)
LDIAG = .TRUE. sub-space diagonalisation
GGA = 91 * xc-type: PB, PW (86), LM, 91 (Perdew-Wang 91)

LWAVE = .T. * write WAVECAR
LCHARG = .T. * write CHGCAR and CHG
LVTOT = .T. * write the local potential LOCPOT
LELF = .F. create ELFCAR file
LORBIT = .F. create PROOUT
LREAL = Auto

Partial charge density
!LPARD =.TRUE.
!IBAND = 4
!NBMOD = 1
!KPUSE = 1
!LSEPB=.TRUE.
!LSEPK=.TRUE.

! NEB

ICHAIN = 0 !Indicates which method to run. NEB (ICHAIN=0) is the default
IMAGES = 4 ! Number of images in the band, excluding endpoints
! SPRING = -5.0 eV/A2 ! Spring force between images; negative value turns on nudging
! LCLIMB = .FALSE. !Flag to turn on the climbing image algorithm
! LTANGENTOLD = .FALSE. ! Flag to turn on the old central difference tangent
! LDNEB = .FALSE. !Flag to turn on modified doubble nudging

! Dimer

! ICHAIN = 2 ! Use the dimer method (required for the latest code)
! IBRION = 3 !Specify that VASP do MD with a zero time step
! POTIM = 0 !Zero time step so that VASP does not move the ions

I mean the 4 images between the reactant and product became distorted after about 80 SCF steps. I don't know whether it is the problem of the program or there are some useful tips to avoid it?

Hi, I am not sure if my answer will help you (since I am using VTST for some weeks only). Obviously you are using the VASP internal CQ algorithm to optimize the ionic structure, which is energy based (IBRION = 2, POTIM = 0.3). Since the NEB method needs a force based optimizer you shouldn't use it. However some force based ompimizers are included in the VTST code, which you can use by defining
IBRION = 3
POTIM = 0
(MD with zero time step)
IOPT = 0-7 (use 2 for force based CQ)
in the INCAR file.
More details are given in
http://theory.cm.utexas.edu/vtsttools/optimizers/

If this doesn't work you might have to inacurate forces. You could try
EDIFF=1e-7
or
ADDGRID = .TRUE.

I hope this will help you.

It would be good to check the magnitude of the forces in your initial band. If they are >10 eV/Ang, and aggressive optimizer can move the atoms into unphysical regions of configuration space.

Also, the conjugate gradient optimizer that you choose (IBRION=2) that is built into vasp can not be used with the NEB. This could be part of your problem.

I suggest doing 10 or 20 iterations using a conservative optimizer (IBRION=3, POTIM=0.1 or POTIM=0.01). Make sure that the images relax to a sensible band and that the forces drop below 1 eV/Ang. Then you can switch to a more aggressive optimizer to get efficient convergence of the band.