I am interested in thermodynamic property of a metal (Fe) complex and using B3LYP* as the functional and def2tzvp as the basis set. Even after using opt=(verytight,calcfc), there is no improvement of the result and some (six) negative frequencies are present. I read in other posts that by displacing the atomic coordinates in Gaussview by manual displacement and saving the structure for further opt can help but I don't have access to gaussview so is there any other way we can do that ? using other softwares such as avogadro or molden ? If anyone knows any other free software for this purpose kindly let me know.
Thanks in advance
I hope you have already read the procedure to overcome this by changing QM level/basis combinations. Additionally, you can extend your trial by using opt(calcall) and quadratically convergent scf procedure using slightly lower level/basis combinations (if your system is not huge to think such an expensive process !). Proceed for some steps and kill the job, then extract the geometry and start again optimizing without calcall by using your desired QM/basis (eg, B3LYP*/def2tzvp). Many times this solves my problems. You can try this.
If you are using Gaussian, you can use its IRC function, then you should just "walk downwards" from your saddle point: https://gaussian.com/irc/
However, when you say you have six negative frequencies: there should be six zero frequencies due to translational and rotational degrees of freedom. If you have six very small (
Jürgen Weippert, thank you for your comment. I have used IRC function and done the calculations by the keyword opt=(verytight,calcfc) but still the negative frequency persists. Now I am running opt=(verytight,calcall) also but it is taking too much of computational time. If there is any other keywords which I should try, kindly let me know. The negative frequencies are quite large starting from -70 to -20 (six negative frequencies).
OK, but do you have the "zero cm-1" modes? Otherwise something about the projection might still be fishy.
So, do you have 3N or 3N-6 modes on the whole? If you have 3N modes and no 0cm-1 modes, then some part about projecting the other degrees of freedom out is truly off. I don't know what it is exactly in Gaussian, since I studied at KIT we were taught most things with Turbomole examples.
OK, so it's projected out and those are real and not artifacts.
If I read your previous posts correctly, you have already imposed stronger criteria for the geometry optimization. Have you considered using a harsher SCF convergence criterium as well.
As I said, I was taught things with Turbomole, but here are the keywords for the SCF:
https://gaussian.com/scf/
Maybe the TightLinEq options might help you?
- Badges
- Science topic
- Similar topics
- Engineering
- Materials Engineering
- Powders