UT theoretical chemistry code forum

I obtained the transition state (TS) using the CI-NEB method with 5 intermediate images. The image corresponding to the TS has an imaginary frequency of ~350-400 cm⁻¹ and the highest energy along the pathway. However, when I interpolated more images between the converged 5 images along the reaction path, I noticed that the preceding two images (lower in energy by ~0.04 eV and ~0.2 eV, respectively than the TS) have significantly higher imaginary frequencies (~890 and 1100 cm⁻¹).

When I re-ran CI-NEB with the coordinates of these preceding images (as potential TS candidates), the optimization converged back to the same TS with ~350 cm⁻¹ as the imaginary frequency. I also tried connecting the preceding image and the TS through CI-NEB, but no barrier was observed.

My questions are:

How do I ensure that the TS is correctly identified?

Are there any specific changes needed in my VASP input file to improve convergence or accuracy? Here are my current settings:

LSCALAPACK= .FALSE.
NPAR=1
LREAL = Auto
ISYM= 0
ALGO= FAST
ENMAX=800
EDIFF= 1E-6
EDIFFG= -0.01
POTIM= 0.015
IBRION= 5
NFREE= 2
ISIF= 2
ISPIN= 2
ISMEAR= 2
SIGMA= 0.2
ADDGRID= .TRUE.
LCHARG= .FALSE.
LWAVE= .FALSE.
IVDW= 12


The PES does not seem flat, so I am puzzled about why these preceding images show higher imaginary frequencies. Any guidance or suggestions would be highly appreciated!

I don't see any problem with configurations away from the TS having higher or lower negative curvatures. I can think of situations where either case will occur.

Thank you for your response! Can I confidently report the TS obtained via CI-NEB with an imaginary frequency of ~300 cm⁻¹ as the TS, even though preceding images show higher imaginary frequencies (~800-1100 cm⁻¹)? Would an IRC calculation be necessary to further confirm the TS, or is this setup sufficient to validate my results?

I just don't understand where this concern is coming from. A first order saddle is defined as a point on the potential with zero force and one negative mode. There can be lots of other points on the potential surface with lower negative modes - but who cares?

To your other question: you can calculate the IRC if you want to but it is not 'necessary' to confirm a saddle. Again, a first order saddle is defined as a point on the potential with zero force and one negative mode.

Thank you for the clarification. I now understand that this is not a problem as long as the TS itself satisfies the criteria of a first-order saddle point.