10
$\begingroup$

I am mainly searching for functionals to calculate spectra of pesticides. I found the benchmark test from Goerigk and Grimme$^1$, but don't understand which of the functionals suits my purpose the best. They published different functionals, which were the most accurate and robust in their study.

My question is, can I use these functionals for my calculations of pesticides?

  1. L. Goerigk and S. Grimme J. Chem. Theory Comput. 2011, 7, 2, 291-309 https://doi.org/10.1021/ct100466k
$\endgroup$
3
  • 1
    $\begingroup$ +1 Could you add some more information about what kind of pesticides you are considering, because that might affect the choice? $\endgroup$ – S R Maiti May 6 at 16:28
  • 1
    $\begingroup$ +1. @ShoubhikRMaiti when you posted that comment, I'd already prepared my fairly long answer which is meant to cover the original question which is about pesticides in general. If the user has a specific list of pesticide molecules they'd like to study, I think a separate question on that could be addressed separately. $\endgroup$ – Nike Dattani May 6 at 16:32
  • 1
    $\begingroup$ "Spectra" - UV/vis? IR? Raman? What spectra are you talking about? $\endgroup$ – TAR86 May 6 at 18:55
9
$\begingroup$

B3LYP

My suggestion for you in this case is to simply use the most common funcitonal for basic properties of molecules: B3LYP. I wrote in my answer here that B3LYP is still a decent functional overall, especially considering its relatively low cost for non-periodic systems (e.g. a pesticide molecule).

The first figure in that answer is directly from the paper that you said you're asking about, and it shows that overall, B3LYP is on average:

  • more accurate than all LDA and GGA functionals studied,
  • at least as accurate as all meta-GGAs studied,
  • is in the "middle of the pack" for single-hybrid funcitonals (but is far more popular and universally implemented (it will be available no matter what software you use!)
  • not as accurate as the double hybrids, but far less computationally costly!

Pesticides are extremely diverse!

There is such a vast range, for example pesticides for rodents (rodenticides) such as the organic molecule warfarin, may be very different from pesticides for algae (algicides) such as the transition-metal-containing copper (II) sulphate. We do not know a universal functional that will work well for all pesticides.

You are therefore best off in my opinion, to "keep it simple" and just use B3LYP, which is:

  • not too computationally expensive,
  • is more likely to be implemented in whatever software you're using, than any other functional that has the same balance of accuracy, computational cost, and popularity,
  • is likely the most popular functional used for individual molecules, so there's a greater likelihood for there to be literature to which you can compare (if necessary), and more people will recognize and understand what it is that you're doing.

If you do choose to use something other than B3LYP, you're more likely to need to provide justification for why you're using it, but if you pick B3LYP, most people will agree that it's a decent "default" choice for studying basic properties of individual molecules. The Goerigk-Grimme paper to which you referred in your question, showed (for example in the second figure in my answer here) that B3LYP is quite decent for everything they tested except for reaction energies, and you are not calculating reaction energies!

$\endgroup$
8
  • 5
    $\begingroup$ Be sure to include a modern dispersion correction. $\endgroup$ – TAR86 May 6 at 18:58
  • 2
    $\begingroup$ @NikeDattani - you want to recommend either D3(BJ) or D4 as a default, or B3LYP-3c if available. The Grimme has a more recent (2020) paper: doi.org/10.1021/acs.jctc.0c00877 $\endgroup$ – Geoff Hutchison May 6 at 20:33
  • 2
    $\begingroup$ @ksousa In Geoff's comment right above Susi's, he gave the "2020 version of the Goerigk-Grimme paper". In Tyberius's answer here: mattermodeling.stackexchange.com/a/4876/5, he suggested a 2017 assessment of 200 functionals, and this was also mentioned in this answer: mattermodeling.stackexchange.com/a/11/5. I think a question about references for assessing density functionals ought to be asked as a separate question though, since comments are for things that directly address the post on which the comment is written. $\endgroup$ – Nike Dattani May 7 at 19:11
  • 3
    $\begingroup$ @ksousa there's an updated version of that Goerigk paper (by more authors), which was published in 2017: doi.org/10.1039/C7CP04913G $\endgroup$ – Kristof Bal May 7 at 19:31
  • 2
    $\begingroup$ @ksousa I agree with Nike and Kristof $\endgroup$ – Susi Lehtola May 11 at 19:43
8
$\begingroup$

A general rule of thumb: never trust results obtained with only one functional (or basis set). If two different functionals give you a similar result, then you can trust the results. If the predictions disagree by a lot, it might be that DFT does not work for your problem.

$\endgroup$
2
  • 3
    $\begingroup$ The two functionals to compare should be on the same level of theory, e.g. two hybrids or two GGAs, otherwise, one may just illustrate the concept of Jacob's Ladder and be led to wrong conclusions about the system. $\endgroup$ – TAR86 May 7 at 20:03
  • 1
    $\begingroup$ @TAR86 sure, but the comparison of a GGA vs a hybrid GGA may also say something about the probable validity of the results. If a pure and a hybrid functional give you similar results, then it is probably correct $\endgroup$ – Susi Lehtola May 11 at 19:43
7
$\begingroup$

If you are talking about IR or Raman spectra, then B3LYP is essentially the best choice possible. But if you are talking about UV-Vis, and if the excited states of the pesticide contain significant charge transfer character, then you must use a range-separated hybrid functional (like wB97XD) or a double hybrid functional (like B2PLYP or PWPB95). Functionals like B3LYP describe local excitations well but tend to underestimate the energies of charge transfer states (giving rise to the so-called "ghost states"), range-separated hybrids describe charge transfer states well but usually overestimate the energies of local excitations, and double hybrids do both kinds of excitations well yet at a higher cost.

If your molecule is within ca. 50 atoms, and if you are using ORCA, you may also want to try STEOM-DLPNO-CCSD, which is a wavefunction method that is even more reliable than double hybrids (yet of course with an even higher cost).

Finally please note that, regardless of what functional you use in the UV-Vis calculation, the geometry optimization of your molecules can always be done with B3LYP (and I especially recommend B3LYP for this purpose).

$\endgroup$
2
  • 1
    $\begingroup$ Welcome to the site, nice first answer! Hopefully we will see you around. $\endgroup$ – Tyberius May 7 at 20:23
  • $\begingroup$ Welcome o our community and thank you for your contributions! We hope to see much more of you in the future!!! $\endgroup$ – Nike Dattani May 8 at 17:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.