5
$\begingroup$

I have a quite big set of molecules to optimize and study and, after some research in the literature, I selected B3LYP/6-311G* level of theory.

These molecules are quite big and wobbly, so I expected the optimization to be rather difficult. However, for most of them, I achieved the optimization with tight threshold (# opt=verytight b3lyp/6-311g* int=ultrafine). However, for some of them, the calculation does not converge. In particular, the SCFs converge with small number of steps but the optimization seems to never reach a minimum.

For this reason, I tried different things: start from a slightly different configuration to 'escape' from that point in the PES, a slightly larger basis set, to change a little the threshold with IOP keyword, the Opt=CalcFC option. None worked. The only way I could obtain a minimum was to lift the accuracy and use the default threshold for the optimization (# opt b3lyp/6-311g*).

I plan to calculate the UV-Vis of these molecules. My question is: Is there another thing that I could try to optimize these systems and/or the default opt parameters would be just fine in general cases?

Improve this question
$\endgroup$
5
  • 6
    $\begingroup$ What I do is to optimize the system in steps, with different theory/precision. For example, optimize first with semiempiric, then DFT with a lower basis set, lower precision, increase the precision, increase the basis set, etc. $\endgroup$
    – Camps ♦
    Commented Jul 6, 2023 at 11:52
  • 3
    $\begingroup$ I wouldn't expect UV/Vis spectra to depend that strongly on the geometry (to the extent that standard vs very tight convergence will make a difference). The prime example of where you might want really tight convergence is if what you want to compute depends on frequencies and you need to make sure these are properly converged (i.e. no negative frequencies for minima, the appropriate negative frequencies for transition states) $\endgroup$
    – Tyberius ♦
    Commented Jul 6, 2023 at 12:01
  • $\begingroup$ Dear Camps, Thanks for your suggestion! Dear Tyberius, I always check if the vibrations are all positive and if the convergence criteria are satisfied also in the frequency step. If these are satisfied for 'non-tight' parameters, there's no need to go to tighter parameters, right? $\endgroup$
    – Laura
    Commented Jul 6, 2023 at 12:50
  • 1
    $\begingroup$ For computing UV/Vis spectra, the errors from the TD-DFT are likely to larger than those due to geometry differences, so I think standard convergence should be fine. As an aside, while you will receive notifications for all comments on your post, other commenters will only be notified if you tag them like this @Tyberius. Unfortunately you can only tag one user per comment. $\endgroup$
    – Tyberius ♦
    Commented Jul 6, 2023 at 13:10
  • $\begingroup$ Thank you very much for your reply :) If you add a summary of your replies, I can mark is as answered. $\endgroup$
    – Laura
    Commented Jul 6, 2023 at 13:11

1 Answer 1

Reset to default
6
$\begingroup$

As Camps mentions in the comments, a common approach for improving geometry convergence is to start with less accurate method/basis or use lower precision convergence criteria and incrementally build up to your desired level of theory .

However, I wouldn't expect UV/Vis spectra to depend that strongly on the geometry (to the extent that standard vs very tight convergence will make a difference). The prime example of where you might want really tight convergence is if what you want to compute depends on frequencies and you need to make sure these are properly converged (i.e. no negative frequencies for minima, the appropriate number of negative frequencies for transition states).

For computing UV/Vis spectra, the errors from the TD-DFT are likely to larger than those due to geometry differences, so I think standard convergence should be fine.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Re-reading, can you explain what exactly you mean with 'to the extent that standard vs very tight convergence will make a difference'? Moreover, I think fluorescence study in solvent will follow the same comment about the TD-DFT error, right? $\endgroup$
    – Laura
    Commented Jul 6, 2023 at 13:18
  • $\begingroup$ For the convergence tightness, I just mean that while UV/Vis will depend on the geometry somewhat (you can't just throw in any random geometry), the difference between standard and tightly converged geometry should be very small and that small difference shouldn't affect the UV/Vis very much. I think the same argument about TD-DFT error would apply to fluorescence in solvent. $\endgroup$
    – Tyberius ♦
    Commented Jul 6, 2023 at 13:58
  • $\begingroup$ Ok, Thank you very much! $\endgroup$
    – Laura
    Commented Jul 6, 2023 at 15:07

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .