Abnormal NEB energy and Convergence problem

[Yuanxi]

Hi,
I performed a NEB calculation.
But in OUTCAR, I found all images
had unreasonably lower energies (-1E6) comparing with the energies
of endpoints (-1E3). I used the POSCARs in 01 to 04 directory to perform energy
calculation, and obtained normal energy. What's the problem ?

I pasted the INCAR below, can you help me to check.
Besides,this calculation cannot converge, and VASP aborted the iteration after 60 electronic steps.
I also tried some of the solutions for convergence problem you gave on this forum,
such as set IBRION = 3 and POTIM = 0.01, but none of them works.

Another question, you said that we should indicate elements in POSCAR.
Where should we put the symbols?

Thanks!

INCAR

General
SYSTEM = NEB Test
ISTART = 0
ICHARG = 0
ICHIAN = 0
IMAGES = 1
SPRING = -5

Electronic Minimization
ENCUT = 400.0
ISMEAR = 2
SIGMA = 0.1
ISIF = 0
NELMIN = 8
EDIFF = 1E-5
MAXMIX = 80

Ionic Relaxation
EDIFFG = -0.02
NSW = 100
IBRION = 1
NFREE = 10
ALGO = Fast
LREAL = A

[Yuanxi]

The force is also incredibly large (etc -1070 ev/Angst)

[graeme]

Start by checking the initial geometries. You can make a movie of the unrelaxed calculation with nebmovie.pl.

Try a single point energy calculation for one of the images - for example in the 01 directory. Does this calculation make sense?

Why do you have IMAGES=1 when you have directories 01 and 04? Your endpoints should be in (for example) 00 and 05 if you want 4 images (IMAGES=4).

Anyways, it seems like a few things could be messed up in your setup.

[Yuanxi]

sorry, I corrected IMAGE settings on the server.

Thanks, I will try the nebmoive.pl.

[alick1102]

Start by checking the initial geometries. You can make a movie of the unrelaxed calculation with nebmovie.pl.

Try a single point energy calculation for one of the images - for example in the 01 directory. Does this calculation make sense?

Why do you have IMAGES=1 when you have directories 01 and 04? Your endpoints should be in (for example) 00 and 05 if you want 4 images (IMAGES=4).

Anyways, it seems like a few things could be messed up in your setup.

i want to ask graeme that why do you suggest us set IBRION=3 with a small POTIM value, say 0.01! when the convergen problem happens. what's the reason? thank youuuuuuu!

[graeme]

If the initial NEB is far from the minimum energy paths, the forces on the images can be very high. By this, I mean forces greater than 10 eV/Ang. In this case, second-order optimizers like LBFGS and CG (IBION = 2 and 1 respectively) are poorly behaved - they rely on calculations of the second derivatives which can be out of control in high force regions. Using IBRION=3 with a small time step is a safe way to bring the forces down to about 1 eV/Ang. At this point, you can switch to a more efficient optimizer.

A controlled descent from the initial high forces is important. If the images move too much and get scrambled, it can take a lot of unnecessary computational time for it to sort out and find the minimum energy path. This initial phase can be very quick - typically just 10-20 iterations. Note that I am not recommending using IBRION=3 for the entire calculation. If you have our latest code, we are pretty excited about new implementations of LBFGS, CG, and the FIRE optimizers for the NEB. If you are interested in the details, we have a recent publication about the efficiency of methods for finding minimum energy paths: http://theory.cm.utexas.edu/henkelman/p ... 134106.pdf

[alick1102]

thank you !
there is another question:
i have recompiled the VASP code with VTST tools following your advises on the web. and i have browsed carefully the topics related to this in your forum. but i remain really not confident that whether i have successfully recompiled VASP with addtion the vtst codes.

i have noted that you said that " graeme on Mon Mar 26, 2007 11:31 pm

In the latest versions of our code, there will be a print statement in the OUTCAR that looks like

VTST: version x.xx

In the current and earlier versions, you can tell if the magnitude of the forces are printed in the OUTCAR

FORCES: max atom, RMS

Also, if you have access to the code directory, you can see if any of the additional modules are present, including, neb.F, dimer.F, dynmat.F, and in the most recent code, opt.F, cg.F, and lbfgs.F" on the forum .

i had tried the new optimizing method by adding "POTIM=0.0, IBRION=3, IOPT=1", but i could not find the above strings in my OUTCAR file.


can you give me an example about this ? a daily working directory would be more appreciated!
ps: i have sent an email to you, waiting for your kindly reply . and thanks again

[graeme]

If you are not getting the following line in your OUTCAR,

VTST: version x.xx

then you have not linked our code into your vasp binary. There are instructions about how to do this on the page: http://theory.cm.utexas.edu/vtsttools/downloads/

Sending a sample vasp calculation will not help with linking in the vtsttools, but we can give you an example of the NEB working with the new optimizers. I'll ask Dan to post a link to one of the calculations in the paper I mentioned.

[alick1102]

many thanks.
one question maybe, after read your recently paper of JCP.
in your paper, you suggest that GL-BFGS optimizer is very efficient, however, your latest code of VTST seems not yet provide this tag? if yes, when we can use it with a very latestttt version ?

[dano]

The optimizers are incorporated in into our latest VTST package. The flags for using these optimizers can be found at: http://theory.cm.utexas.edu/vtsttools/optimizers/ When using the IOPT flag you must set IBRION=3 and POTIM=0.0 to disable the optimizers intrinsic to VASP.