5
$\begingroup$

I am having trouble with a difficult convergence in Gaussian16. I am running two separate PES scans for hindered rotors in my molecule that will eventually be used in subsequent rate calculations. I decided to back down the optimization convergence criteria for one of the scans from verytight to tight, as looking at the criteria in the failed runs shows that it is getting very close but not close enough to finally converge, at which point the displacement and RMS displacement begin to increase.

I've also thought about changing the SCF algorithm or increasing my integration grid, however, the grid is already at ultrafine and I am trying to minimize the use of SUs.

The issue is that I already have the other PES scan that has been running for a while and is using the opt=verytight criteria and I'm worried that using both PES scans together for the rate calculation might not be valid. However, since they are both using the same model chemistry, (i.e. functional, basis set, integration grid, etc) just with different convergence thresholds, it may not be the worst thing in the world (Albeit, not ideal).

My Question

  • Is it valid or invalid to calculate the PES of two hindered rotors on the same molecule with different convergence criteria?

My Rationale

  • If someone can confirm/reject the following, that would be great also:

    I'm currently under the impression that since this is not a direct comparison between
    two energies, rather, the optimization is being done to find the PES for a rotor, its not that big of a deal.

$\endgroup$
3
  • 3
    $\begingroup$ Ideally all the calculations should be on the same level of theory. In particular when you are trying to find the reaction rate, the overall curvature of the PES would matter and combining energy values with different convergence criteria might introduce more errors and spurious features on the PES. Maybe you can be post the title card for your calculations and atomtypes in your molecule (for ex:- C,H,O,..), so that we might to help with reducing the computational cost of the calculations? $\endgroup$
    – mykd
    Jul 30, 2023 at 4:08
  • $\begingroup$ @mykd unfortunately I cant change the functional or basis set for a fews reasons. Besides that, its just the ultrafine grid (necessary for DFT functionals) and the verytight convergence criteria. I ended up having to re-run both PES energy scans because the other one wouldn't converge either. I even tried calculating the force constants. The system only has C,H,O in it. (18 atoms total). I am definitely open to any other advice for reducing computational cost or dealing with tough convergence. $\endgroup$
    – user1441
    Jul 30, 2023 at 20:11
  • $\begingroup$ If you are having convergence issues, maybe you can post the title card here, so we can suggest further optimization if possible. Also if your system is a hydrocarbon and you are working at ground state , have tried the same scan with PM6 or DFT-B? It might give you a decent idea about the PES without much computational load $\endgroup$
    – mykd
    Jul 31, 2023 at 2:06

1 Answer 1

4
$\begingroup$

That’s a bit up to interpretation… I guess it depends on what you want to get out of the scans exactly (i.e., are the geometries important, or just the energies on those geometries?). If you want the scan points themselves (the geometries), I would say that the convergence criteria is probably quite important. If you’re more interested in the potential energy curve itself, the geometry just has to be ‘good enough’ for the energy to be ‘accurate’. That being said… you can still say that all your scans were converged to tight criteria. So, in that sense, it actually depends on whether you think it’s important that they’re optimised to vtight or not. And that depends on the exact nature of the calculation.

All this being said, an alternative solution would be to calculate the PES geometries using one method and then calculate energies on the resulting geometries at a different (higher) level of theory. You get the best of both worlds, that way.

$\endgroup$
5
  • $\begingroup$ Makes sense and I agree. I think for this specific calculation, the deviation in the RRKM rates wont be significant enough for me to justify tight vs vtight criteria (unless there is a large difference obtained from the difference convergence criteria that I am not ancipating). If i was looking to calculate frequencies for future spectroscopic investigations, then using vtight is more important but for rough RRKM calculations, I think tight would be okay. I ended up restarting the second calculation and backing it down from vtight to tight anyways so comparing its not a problem anymore. $\endgroup$
    – user1441
    Jul 30, 2023 at 20:15
  • $\begingroup$ @isolated_matrix thank you for the input $\endgroup$
    – user1441
    Jul 30, 2023 at 20:16
  • $\begingroup$ @Lazarus98 I don’t think you needed to stop the vtight calculation; after all, the vtight calc will still be converged to the tight criteria. But, whatever works! $\endgroup$ Aug 1, 2023 at 3:57
  • $\begingroup$ @isolated_matrix I should've specified that unfortunately I had to due to the same reasons for the other calculation (i.e. coming close but not converging). thank you for your advice. $\endgroup$
    – user1441
    Aug 1, 2023 at 14:16
  • $\begingroup$ @Lazarus98 ah, fair enough. Hopefully they work! If you’re using DFT and still having trouble, you can also try integral=superfine, which might help $\endgroup$ Aug 2, 2023 at 3:08

You must log in to answer this question.