wrong transition state with vasp

[fme]

Thanks for the forum.
I have a question regarding a recent problem I am having with neb calculations. I am trying to calculate CO dissociation barrier on small Pd cluster in gas phase (1 kpoint). The barrier I am getting is unexpectedly quite small. More detailed analysis indicates that transition state C-O distance is much larger than what has been reported earlier. That means vasp is completely missing the transition state. I have tried to add more images around that point and I do capture more or less right TS (distance wise) but when I run NEB to convergence same thing happens and C-O distance at converged position is again much larger and have a wrong barrier. There is a big jump from image TS to TS-1 (~ 1A). Again adding more images in between does not help and it converges to same final TS (almost).
I have tried to run first few iterations without climbing image and get good initial path. For e.g. after first 10 iteration the CO distance at TS is 1.86 A and barrier is about 1.6 eV (which is very close to what is expected from literature). Then I turn on climbing image and after couple of hundred iteration that barrier become very small like 0.4 eV and TS CO distance become like 2.7 A.
We also did some calculations with Gaussian and found barrier ~ 1.3 eV and C-O distance at TS 1.8 A on exact same system.
Do anyone knows why this could be happening and any general suggestion to how I can fix this. I have played around with number of input parameters in INCAR and did quite few tests but still no use.
I will greatly appreciate any help or suggestions. Let me know if you need any more information from me.

[andri]

I'm not really sure what the situation is. But I recall some years ago I was trying to minimize the structure of CO on a Pt surface. If I started with the CO bond length larger than the experimental value, VASP wouldn't not converge and eventually the molecule would disassociate. But if I started with the bond length too small it converged without any problems. Perhaps a similar strategy will work here as well? Can you take the run the NEB for 10-20 steps and start a dimer from that point? Then reduce the bond length to something less than 1.8 A and converge.

What minimizer are you using now with the NEB?

[fme]

Thanks for your suggestions. I am using quasi-Newton method for neb calculations. I will try dimer method to see if it improve TS.

Thanks again,

Faisal

[andri]

Is that with IBRION = 1? Is so, you should really look into using the optimizers developed by the Henkelman group instead. See http://theory.cm.utexas.edu/vtsttools/optimizers/ and JCP , 128(13), 134106, 2008