I am trying to speed up the discovery of a transition state. I know what the TS looks like, so I thought I would try and get the system to co-operate more by specifying what bonds should stretch.
I have two identical structures. On the one I perform a bond scan, ten steps, -0.1 angstrom (so they are being brought closer together, say a hydrogen off water to the oxygen on a carbonyl group), and on the other I do exactly the same but I include opt=(calcfc,ts). Yet when I compare how many modredundant steps each has gone through, the non-TS system has stepped through 4 of 11, yet the modredundant+TS is still on the first step.
Is this total counter intuitive? I am thinking that during a regular modredundant bond scan, Gaussian moves to the next step after finding a minimum. I am now wondering whether by specifying opt=ts that a modred bond scan only progresses to the next step when a saddle point is found. If this is the case, this probably will never happen.
Thanks
Thanks for the conformation, Travis. How would you typically use scan mod redundant when looking for a TS which involves, say, two motions i.e., hydrogen of a carbon onto the oxygen of water, which in turn looses its own hydrogen to protonated the oxygen of a carbonyl group (tautomerisation)? Would you typically find the highest point for the first motion, then do the same for the second motion, then use their predicted distances (i.e., distance of the hydrogen between the two atoms which pertains to a high energy state) to generate a new model? Or perhaps find the first highest energy state (for the first hydrogen motion), then use bond coordinate restraints on those whilst performing the second mod red coordinate scan? The highest energy structure at this point would be used for a traditional TS.
I have tried mod red bond scan of two motions at once, but this typically involves 100+ steps which is too time consuming.
Interestingly I am trying both. I've had to CG to an even lower level (3-21+) as my system is pretty big, and I am running multiple starting structures along where I assume the TS coordinates would run along. I do eventually find the very low level TS structure and then I use the vibrational frequency calculation results for the start of a higher level calculation - which really helps, and the TS is rediscovered pretty quickly. The problem is I go through so many individual computational jobs because those that fail will fail with an "Internal coordinate" error, which I assume is a symmetry issue with the Berny algorithm.
Dear Anthony,
If I understood correctly, in the scan optimization you are following a given coordinate, which is frozen while the rest is minimized. This is like optimizing minima using constrains, so a lot of care is advised. However, it may be useful to obtain some TSs, for instance TSs with symmetry. In the normal TS search with opt=(calcfc, ts) the program needs a guess for the Hessian (second derivatives of the energy with respect to the coordinates), which will be performed at the same level as the optimization. The calculation of a Hessian is much more time consuming than a normal step in the optimization, since the latter requieres only gradients (first derivatives). Probably your TS-optimization is still in the Hessian calculation. If you do not want to expend much time in the Hessian generation use opt=(mndofc,ts) or just perform a freq calculation with a semiempirical method and after that read the Hessian from the checkpoint with the RCFC keyword. My advise is to use opt=(calcHFfc, ts) that performs the Hessian at the HF level.
I think, the scan before the TS search will be a more rational choice. You can do the initial scan (as well as the initial TS search) using loose cut-offs for your optimizations. That should speed up the process.
Also, during the initial, loose scan, you may fix certain (not directly connected to your scan coordinate, so that it does not affect your scan greatly) bond angles etc., to reduce 'dancing' and sleep up the scan.
- Badges
- Science topic