Hi all,
Until last night I've only ever had this one before. After two days of calculating opt=(ts,noeigentest) freq=noraman my system converges and then goes on to perform a frequency calculation. I was extremely happy until I noticed that all of the frequencies were positive. In particular, the first set of frequencies reported were:
Frequencies -- 25.0896 35.5939 39.9547
Red. masses -- 6.0310 6.9017 5.5635
Frc consts -- 0.0022 0.0052 0.0052
IR Inten -- 7.2234 18.3572 34.5286
I'm very confused as to why it converged elsewhere and not with an imaginary frequency. Any suggestions would be appreciated.
Many thanks
Anthony
You've provided the NoEigenTest keyword. This prevents the code from only continuing with the optimisation if it has one and only one negative eigenvalue (i.e. a TS).
This means you should occasionally expect the behaviour you've seen. You could try removing NoEigenTest, but it can become difficult to locate TSs in large molecules without it.
Hello,
Sometimes Gaussian just not found TS. You should try change geometry of yours input files, and when that not help try use qst2 or qst3 method.
Thanks Kamil,
I'm running multiple jobs in parallel (a few qst3 and ts jobs) from earlier modredundant scan bond distance calculations. I'm just very surprised that gaussian should ever converge if it can't find a ts.
Thanks again.
Hi,
You can also add to your opt options CalcFc of CalcAll which will increase the chances to obtain the TS. opt=(CalcFC, ts, no eigentest) is less expensive, but if you use opt=(CalcAll, ts, no eigentest) it will perform a frequency calculation in each step and it will try to follow the gradient properly.
I hope it helps
Anthony, sometimes it does happen. Why, I don't know, but I saw this issue repeatedly. If you restart the calculation, it usually works. If it doesn't, just try to get a better guess for TS and then start the calculations. This tactics usually works for me.
You've provided the NoEigenTest keyword. This prevents the code from only continuing with the optimisation if it has one and only one negative eigenvalue (i.e. a TS).
This means you should occasionally expect the behaviour you've seen. You could try removing NoEigenTest, but it can become difficult to locate TSs in large molecules without it.
When I have encountered this problem, it often is resolved by trying multiple starting geometries for the TS. It sounds like you are using the position at the peak or near it from a modredundant scan and have been doing this.
Another approach: From the first calculation using a modredundant scan with a bond increasing incrementally creates an energy curve which rises in energy for each point of the calculation to a peak and then begins to decrease in energy. Perform a second iteration modredundant scan beginning with a geometry that is not at the peak but on the downward part of the curve several points from the peak and this time decrease the bond length rather than increasing as the first calculation (a reverse scan direction). This creates a curve which should also increase in energy to a peak for each point followed by a decrease in energy. Try locating the TS from a geometry near or at the peak as I assume you were doing from the first scan. If this again does not succeed, start a third modredundant scan iteration from a geometry significantly far from the peak with incremental increases in the bond length and try to find a TS from the third scan.
As others have mentioned, calcfc and calcall may help. Calcall can be very long and calcfc might be adequate (I try to avoid calcall if the complex is large). Resource limitations are certainly a factor. I have usually found a good TS geometry from subsequent scan iterations when other methods are not successful.
- Badges
- Science topic