I am performing a CBS-QB3 calculation with Gaussian 09 to find the bond dissociation energy of a radical reaction. The radical species is designed to distribute the radical character across the entire molecule. When I perform the calculation, the job immediately fails possibly due to severe spin contamination with the following error. The input geometry has been pre-optimized with a smaller basis set.
Annihilation of the first spin contaminant:
S**2 before annihilation 1.0390, after 0.8293
Convergence failure -- run terminated.
However, the job runs perfectly fine if I use Gaussian's ROCBS-QB3 method as expected. This makes sense since the restricted open-shell method will fix the spins. Is this an appropriate solution? I know that it is often better to use an unrestricted method for open-shell radicals so I would like a second opinion. I do not know why the spin contamination is so bad for this particular species. Thank you!!
There is no direct relationship between spin contamination and SCF unconvergence problem. There are many useful keywords, which could facilitate SCF convergence in the unrestricted CBS-QB3 calculation.
If ROCBS-QB3 can get converged, then using ROCBS-QB3 instead is also a viable choice. In the original paper of this method (JCP, 125, 094106 (2006)), it is shown that ROCBS-QB3 in many cases works marginally better than unrestricted CBS-CB3, however for some properties the ROCBS-QB3 is slight worse than unrestricted CBS-QB3.
There is no direct relationship between spin contamination and SCF unconvergence problem. There are many useful keywords, which could facilitate SCF convergence in the unrestricted CBS-QB3 calculation.
If ROCBS-QB3 can get converged, then using ROCBS-QB3 instead is also a viable choice. In the original paper of this method (JCP, 125, 094106 (2006)), it is shown that ROCBS-QB3 in many cases works marginally better than unrestricted CBS-CB3, however for some properties the ROCBS-QB3 is slight worse than unrestricted CBS-QB3.
Bojidarka Ivanova,
The job still ended with the same error. I updated the attached "output.txt" with the new log file. My route now looks like this:
# cbs-qb3 geom=connectivity iop(5/13=1)
If I am not mistaken, iop(5/13=1) will allow the job to continue without terminating even though it hasn't converged. Is there a serious risk of producing inaccurate results with this or is the effect on accuracy negligible? Thank you!
- Badges
- Science topic