ClNEB, TS becoming the anchor image

[ravi.agarwal]

Hi,
I am carrying out an one intermediate, climbing NEB calculation in Mg. I am trying to figure out the transition state for an solute between two interstitial sites Octahedral and Non basal crowdion site. My initial guess is obtained using nebmake.pl. I am using an force convergence of 0.005eV/A. After the convergence I find that the TS is the anchor image i.e. 02 folder POSCAR which is not an saddle state.
Could you please give me some suggestion to rectify this problem
Thanks
-Ravi

[graeme]

Something sounds fishy. If you post the calculation, we'll take a look.

[ravi.agarwal]

Thank you very much for your reply.
I am attaching the required files to run the single image Cl-NEB. Firstly, I ran the 10 steps with CG on POSCAR obtained by nebmake.pl to get a good estimate of POTIM and positions. So, the CONTCAR obtained using CG is used as POSCAR which is finally optimised using QN (IBRION=1) with the POTIM (average) guess obtained from vasp.out.
My INCAR looks like this for CG 10 steps:
System = hcp Magnesium B

ISTART = 0
ICHARG = 2
PREC = Accurate
EDIFF = 1e-8
ALGO = Normal
ENCUT = 500
LREAL=Auto
ISMEAR = 1
SIGMA = 0.25
NBANDS=192
NGX=96
NGY=96
NGZ=144
LPLANE=TRUE
NPAR=2
ISIF = 2
EMIN = 0
EMAX = 10
!NEDOS = 1001
IBRION = 2 !CG
POTIM = 1.0
NSW = 10
EDIFFG = -5e-3
LWAVE=.FALSE.
#NBE
ICHAIN = 0
IMAGES = 1
SPRING = -5.0
LCLIMB = .TRUE.

After that QN is used as optimiser whose INCAR looks like:
System = hcp Magnesium B

ISTART = 0
ICHARG = 1
PREC = Accurate
EDIFF = 1e-8
ALGO = Normal
ENCUT = 500
LREAL=Auto
ISMEAR = 1
SIGMA = 0.25
NBANDS=192
NGX=96
NGY=96
NGZ=144
LPLANE=TRUE
NPAR=2
ISIF = 2
EMIN = 0
EMAX = 10
!NEDOS = 1001
IBRION = 1 !QN
POTIM = 0.75
NSW = 200
EDIFFG = -5e-3
LWAVE=.FALSE.
#NBE
ICHAIN = 0
IMAGES = 1
SPRING = -5.0
LCLIMB = .TRUE.

I am running this on vasp4.6.36
I have attached an zip folder which has all the files for running initial 10 steps by CG optimiser and then folder containing QN optimiser files(CHG and CHGCAR removed).
Is it happening because my initial and final state are close to each other?
Thanks
-Ravi

[graeme]

I don't understand the problem. I see that the 01 image is converged after your QN optimization.

So what is the problem? How do you know that the 01 image is not a transition state? Why are you saying that the 02 image is? Is this based upon the relative energies of the two images? If so, I can't see the OUTCAR files for 00 and 02 so I can't explain that difference. What I can say is that often people use different settings for the endpoints as compared to the NEB calculation. Differences in settings will give rise to an energy shift between the two.

Check that everything is the same between your endpoint calculations and your band. Also, turn symmetry off for both calculations (ISYM=0). If you still see the problem, let me see the OUTCAR for the endpoints, or give me a better explanation for saying that the 01 image is not a saddle and that 02 is.

[ravi.agarwal]

Thank you for your prompt reply.
I am sorry that I did not make myself clearer.I just realized that my first post is not exactly coherent with the second post. Let me explain you from starting.
->00/ and 02/ folders have anchor image (endpoints) which are metastable (I calculated the eigenvalues of Force constant matrix obtained via restoring force analysis, all three eigenvalues are positive).
->01/ folder is the intermediate image which is being optimised by ClNEB, initial guess obtained using nebmake.pl wrt 00/ and 02/ folder image .
->Intermediate image is optimised in two stage:
-->In first stage, NSW=10 of CG (IBRION=2) is used as optimiser to get a good estimate of POTIM and also to arrive near the saddle point(CONTCAR).
-->Second stage which is QN , starts from where CG run has ended with a new POTIM parameter, new POSCAR ,CHGCAR (ICHARG=1) and QN as optimiser (IBRION=1).This step is ran till the calculation converges.
->After convergence I see that 01/ folder image energy and CONTCAR is same as that of 00/ folder anchor image (sorry it's not 02/ folder, my mistake) POSCAR(approx.) and energy. This is confusing because 00/ folder image is an metastable state not an transition state.
-> I had make sure that Endpoint and ClNEB vasp settings are same and the number of bands are also same.
->I haven't tried the endpoint with ISYM=0. Is it because symmetry might be making the endpoint metastable?

I have one more NEB calculation with different configuration which is also behaving the same way where intermediate image is converging to an anchor image.
Hopefully I have made myself clear now.Please let me know what do you think of this case now?
Thanks
-Ravi

[graeme]

Yes, try turning off symmetry and re-relaxing the endpoints. Then do the 1-image band, again without symmetry.

Also, your k-point sampling and encut (precision) are overkill. I find it is much better to use inexpensive settings for these calculations. You can always reconverge them with accurate settings once you have found the critical points.

[ravi.agarwal]

I will re run them with symmetry turn off.
Just want to clarify 1-image band also means 1 image Cl-NEB right?
I will also make note of inexpensive settings.Thanks for the tip.
-Ravi

[graeme]

Yes, one ci-neb image should be fine for this simple diffusion process.

[ravi.agarwal]

I rerun the anchor images and Cl-NEB with ISYM=0. I followed the same procedure for Cl-NEB as described above i.e. CG and then QN. Even after ISYM=0, I am getting the same result as before.I see that the CONTCARS of anchor images are not much different with symmetry off and on case (max change 10^-5 A). Could you please suggest anything else that can rectify it?
Thanks
-Ravi