when will the NEB calculation coverge?

Message

whzhang · #1 Post by **whzhang** » Thu Jul 21, 2005 2:16 pm

If I set EDIFFG = -0.02 eV/A,when will the NEB calculation converge? Is all the component of CHAIN + TOTAL (eV/Angst) smaller than 0.02eV/Angst ?

#2 Post by **admin** » Thu Jul 21, 2005 3:29 pm

EDIFFG <0 sets a force convergence criterium, it is fulfilled if the LARGEST
remaining force is less that the specified value.

aarondesk · #3 Post by **aarondesk** » Mon Jul 25, 2005 6:57 pm

One thing you might try is using the climbing image neb method. See http://theory.cm.utexas.edu/vtsttools/ .
I tried for a very long to get convergence with NEB, but finally switched to the CI-NEB method. Convergence still takes a long time (~200 steps), but I did get a transition state.

whzhang · #4 Post by **whzhang** » Tue Aug 30, 2005 3:30 am

I use four images to perform the calculation, from the OUTCAR I find the FORCE on atoms decreasing at the initial steps but it begins to increase after about 200 steps, I don't know how to do now.

Below is my input file
SYSTEM =defts

LREAL = .TRUE.
PREC = Medium
#Electronic Relaxation 1

ENCUT = 450 ev
NELM = 200
NELMIN = 4
NELMDL= -8
EDIFF = 1E-05
ISPIN = 2
INIMIX = 0
AMIX = 0.05
BMIX = 0.0001 ! almost zero, but 0 will crash some versions
AMIX_MAG = 0.2
BMIX_MAG = 0.0001 ! almost zero, but 0 will crash some versions

VOSKOWN = 1

IMAGES = 4
SPRING = -5
LORBIT = 11

#Inoic relaxation
IBRION = 1
ICHAIN = 0
#LCLIM = .TRUE.

NSW = 500

ISIF = 2
#IWAVPR = 11

#DOS related values
ISMEAR = 0
SIGMA = 0.01

#Electronic relaxation 2
IALGO = 38
LWAVE = .FALSE.
LCHARG = .FALSE.

#5 Post by **admin** » Tue Aug 30, 2005 9:43 am

it may be helpful to increase the convergence to
1) start with low dimensionality (decrease NFREE ; NFREE = 2)
for the first few steps
2) reduce the step width (POTIM) from the default value.
3) increase PREC to high to assure that the forces are accurate
enough

superyoyo · #6 Post by **superyoyo** » Thu Sep 15, 2005 4:01 am

I have the same question
now I 'm calculating a barrier use NEB. I use INCAR like this
ISMEAR = 0
SIGMA = 0.05
POTIM = 0.05
NSW=40

NELMIN = 4
EDIFF = 1E-4
EDIFFG = -0.1

IMAGE =4
SPRING= -5

I try to calculate over 200 step (copy CONTCAR to POSCAR when 40 step finished)
Total energy is become positive (+0.5...) and then negative(-510.2...) and now I have no idea that will it converge? and when?

my second question is
Can VASP4.4 calculate the climbing NEB?
thanks

mhyman · #7 Post by **mhyman** » Thu Sep 15, 2005 3:05 pm

POTIM = 0.05 is a really small time step. I would try POTIM = 0.2 for at least 100 cycles. Then look at the energies of each step. First make certain that you are starting with end points that are minima. I've wasted much time not checking this.

Then, check to see how the energies are changing. If they are steadily changing, just keep running more cycles. If they are oscillating over a range of say 0.2 eV, you need a smaller time step.

superyoyo · #8 Post by **superyoyo** » Fri Sep 16, 2005 4:04 am

[quote author= 0.05 is a really small time step. I would try POTIM = 0.2 for at least 100 cycles. Then look at the energies of each step. First make certain that you are starting with end points that are minima. I've wasted much time not checking this.

Then, check to see how the energies are changing. If they are steadily changing, just keep running more cycles. If they are oscillating over a range of say 0.2 eV, you need a smaller time step. [/quote]
Thanks for your suggestion. I 'm a student in Thailand and start to learn NEB for 1 month. I like this board and this forum

#9 Post by **admin** » Fri Sep 16, 2005 11:30 am

the climbing and chain options were not developed and are not
supported by us (CMS, Institut fuer Materialphysik d. Univ. Wien),
therefore please ask those colleagues who
developed it