3
$\begingroup$

I'm investigating reaction mechanism and I want to calculate energies of my reagents, products and TS on a higher level of theory than I use for the geometries and frequencies.

I'm thinking about MP2, but I've read that is has significant basis set dependence and it's advised to investigate convergence to CBS limit. Can I use double-hybrid DFT as a substituent, as far as I understand they aren't so basis set dependent?

Is basis set dependence a thing to consider for regular mechanistic calculations that aren't pretending to be quantitative?

$\endgroup$
10
  • $\begingroup$ I would still check basis set dependence by at least comparing DZ and TZ. Why do you not want to do it? $\endgroup$ – Nike Dattani Jan 19 at 15:21
  • $\begingroup$ @NikeDattani is there a formula to estimate MP2 CBS limit, or it's to just see if the energy changes significantly? I have a pretty large system, and such check is twice as much calculations. Also I'm just curious about proper methodology. $\endgroup$ – roma ichenko Jan 19 at 15:51
  • 1
    $\begingroup$ Yes if you have an energy at 2 different basis set sizes, you can extrapolate to CBS using a few different formulas. I understand that doing DZ and TZ is double the number of calculations but DZ is going to cost maybe 10% of what TZ costs, so really it's not much more. I think you have to do TZ anyway because DZ is usually not enough for anything, even with a double hybrid. $\endgroup$ – Nike Dattani Jan 19 at 15:55
  • 1
    $\begingroup$ For converged results, you would need a QZ basis set even for the double-hybrid. As for extrapolating CBS from (DZ, TZ), at least for non-covalent interactions, some authors warn against it, claiming that TZ by itself is actually more accurate. Overall, using a hybrid functional with a significant amount of Fock exchange (because of the TS calculation) may be the more balanced option. Check e.g. the GMTKN benchmarks for ideas which one to choose. $\endgroup$ – TAR86 Jan 19 at 18:48
  • $\begingroup$ @NikeDattani I've tried DZ and TZ and figured out that MP2/cc-pVTZ calculation doesn't fit my disk quota, so I'll stick to double-hybrid. $\endgroup$ – roma ichenko Jan 20 at 2:22
3
$\begingroup$

Double hybrids contain MP2 so the basis set dependence is similar to MP2; the only thing you gain is the overall weighting factor that is used to damp the MP2 correction.

The reason to use double hybrids is just to get results of better than MP2 quality at a cost similar to MP2. But, in many cases modern range-separated functionals like $\omega$B97M-V out-perform double hybrids at a fraction of their cost...

$\endgroup$
2
  • $\begingroup$ So, ideally, double-hybrids should outperform MP2? $\endgroup$ – roma ichenko Jan 21 at 5:55
  • $\begingroup$ Depends what you mean. Ideally, they should outperform MP2 in accuracy. They may not outperform MP2 in computational cost, since the double hybrid is MP2 + DFT which is more expensive than just MP2. $\endgroup$ – Susi Lehtola Jan 21 at 20:02

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.