Hi,
I'm trying for the first time an stapt transition state search on the excited state landscape (jobex -ex -trans) and I'm puzzled by the results and could use some guidance. This is what I did:
1) I prepared an initial guess for the TS by interpolating the two minima on either side of the TS.
2) I performed a frequency calculation on the TS using Numforce -ex 1 -c. This frequency calculation yields one imaginary frequency and the associated eigenvector makes sense chemically.
3) I performed a set of single point dscf/egrad calculations on geometries created using screwer by distorting the structure along the eigenvector of the imaginary frequency in both positive and negative directions (by entering positive and negative temperatures). These calculations show that in one direction the energy increases (uphill, positive distortion) while in the other direction the energy decreases. Moreover, the same calculations showed that the overall gradient (as obtained through grep cycle gradient) decreases in the uphill directory and increases in the downhill direction.
4) I then attempted a transition state search using jobex -ex -trans (with itvec set to 1 in the input file and a variety of maximum step-sizes) and in all cases in the first step the energy goes down and not up as I would have expected from step (3). Having said that the gradient does go down.
Now I'm aware that in a transition state search the energy does not necessarily goes down (although I'm sort of expecting it do in the case there's only imaginary frequency and you're following it) but based on (3) I would at least have expected the first step of the transition search (especially with very small maximum step sizes of e.g. 1e-3) to be uphill.
I would be very happy if someone could please point out to me what I'm missing here.
Thanks in advance,
Martijn