solid state CI-NEB, not converge

[rlzhou]

I'm running a solid state CI-NEB. But after more than two hundred steps, it does not converge. The energies and forces of some images jump dramatically. What's the problem? My INCAR is:
NWRITE = 2
ISTART = 0 job : 0-new 1-cont from WAVECAR 2-samecut
ICHARG = 2 0-from ini-wf 1-read CHGCAR 2-superposition of ACD
ISPIN = 1
ISYM = 0 0-nonsym 1-usesym
SYMPREC = 1e-6
PREC = ACCU
LCHARG = F
LWAVE = F

Electronic Relaxation
ENCUT = 520
EDIFF = 1E-05 stopping-criterion for ELM
NELM = 500
NELMIN = 4

Ionic Relaxation
ISIF = 3
IOPT = 3
MAXMOVE = 0.2
TIMESTEP = 0.1
POTIM = 0.0
FMAGFLAG = .TRUE.
EFIRST = -76.090398
ELAST = -76.438477
NSW = 500
IBRION = 3 -1-fixed 0-MD 1-quasi-New 2-CG 3-DMD
EDIFFG = -0.05 stopping-criterion for IOM
PSTRESS = 300
ISMEAR = 1
SIGMA = 0.05

Algorithm:
IALGO = 48


ICHAIN = 0
IMAGES = 16
SPRING = -5
LCLIMB = .TRUE.
LNEBCELL = .TRUE.


The output files of nebconverge.pl are attached..

[xph]

I have some suggestions:
1. Use fewer images, usually 5 images between initial and final minimum are enough. More images make it more difficult to converge.
2. Make sure that ENCUT is more than 1.3*ENMAX. ENMAX is the largest ENMAX in POTCAR.
3. Turn off climbing image in the beginning and turn it on when the path almost converge.
4. If the above three doesn't help, try to set a small NSW and resubmit the job when it finishes. In this way, the basis sets can be reset during cell changing and the forces and stress are more accurate.

[rlzhou]

Thank you for your suggestion. I read some posters in the forum. They said when the neb can not converge the image should be increased. The ENCUT I set is larger than the 1.3*ENMAX. I will try to turn off the CI-NEB according to your suggestion.

[xph]

Could you try the calculation without any pressure? The SSNEB in VTST has not been tested under pressure. The TSASE version does.

Still, if there are intermedia minimums between the initial and final states, it is more difficult to converge. Rather than increase the number of images, it is more efficient to chop the whole band at the intermedia minimums into several independent bands. 16 images are way too many for SSNEB.

[rlzhou]

Do you mean that I first do calculations with fewer image (e.g. 4), and when it converges, then insert more images between adjacent converged images?

I have read some papers where the authors predicted the phase transformation path under various pressure using solid state CI-NEB method combined with VASP (PRL, 106, 075501 (2011)). So, there should be no problem with the pressure.

[xph]

There is no need to insert more images if the resolution is satisfactory. 5 images are enough for most cases.
Just checked that PSTRESS was good in the latest VTSTCODE (updated a month ago).

About that paper, they were not using our SSNEB method. To repeat their results, I think you can just set like this:
ISIF = 3
IBRION = 3
POTIM = 0.05
ICHAIN = 0
LNEBCELL = False (Turn off SSNEB, which is the default)
IMAGES = 5

This means apply NEB to atom freedoms and optimize the cell all along the way. Here IOPT must not be set.
The above is a kind of drag method. For concerted mechanism, the results could be the same as our SSNEB.
The comparison of SSNEB and drag methods can be found here:
http://theory.cm.utexas.edu/henkelman/p ... 074103.pdf

[rlzhou]

Thank you very much. I will try it.

[rlzhou]

The VTSTCODE on the download website is updated in August,2011. Do you have a new version? Where can I download it?

[graeme]

The code on our webpage has been update: http://theory.cm.utexas.edu/vtsttools/downloads/

[rlzhou]

The new version failed to compile with error. The subroutine of chain_force dosn't contain 'stress' varable. Please update a correct one.

mpif90 -FR -lowercase -assume byterecl -heap-arrays -FR -O0 -c main.f90
main.f90(2555): error #6784: The number of actual arguments cannot be greater than the number of dummy arguments. [CHAIN_FORCE]
CALL CHAIN_FORCE(T_INFO%NIONS,DYN%POSION,TOTEN,TIFOR, &
-----------^
main.f90(2556): error #6633: The type of the actual argument differs from the type of the dummy argument. [B]
TSIF,LATT_CUR%A,LATT_CUR%B,IO%IU6)
------------------------------------^
main.f90(2556): error #6634: The shape matching rules of actual arguments and dummy arguments have been violated. [B]
TSIF,LATT_CUR%A,LATT_CUR%B,IO%IU6)
------------------------------------^
compilation aborted for main.f90 (code 1)
make: *** [main.o] Error 1

[graeme]

Look at the change that is required to main.F for the ssneb:
http://theory.cm.utexas.edu/vtsttools/downloads/

[rlzhou]

yes, I have changed main.F according the instruction. In chain.F file, the subroutine chain_force is : SUBROUTINE chain_force(nions,posion,toten,force,a,b,iu6). It should contain a 'stress' variable.
The sentence in main.F is changed into
CALL CHAIN_FORCE(T_INFO%NIONS,DYN%POSION,TOTEN,TIFOR, &
TSIF,LATT_CUR%A,LATT_CUR%B,IO%IU6)
TSIF should be corresponding to 'stress'

For the later version, I also encountered this problem. It was corrected after I sent a message to you. Please see the poster:"Problem of Compilation of version 2.04" . So, I think you upload a wrong version again.

[rlzhou]

I compared every file in vtstcode-2.04a downloaded before and vtstcode-2.04b. I found version of dimer.F is v1.36 in vtstcode-2.04a while it is v1.33 in vtstcode-2.04b. So, I think you just upload an older version of vtstcode. In this package, it does not contain ssNEB.

[graeme]

Ok, round two: things should be updated now.

[rlzhou]

When I use 5 images, the jobs run well. After the calculation finish, I use the nebresult.pl script to get the barrier. I opened the CI-NEB by setting the 'LCLIMB = .TURE ." However, I found that no image sit on the saddle point. Is that because I used small number of images or any other reasons?
The mepss.eps is attached.