What one can do to eliminate basis set superposition error (BSSE) if there is no chance to use larger basis sets or counterpoise correction.

  • $\begingroup$ Can you clarify what you mean by no way to use counterpoise correction? In general if the system can be computed at a given complexity, the counterpoise correction is normally cheaper / the same cost as the original calculation. $\endgroup$ Aug 24, 2020 at 15:31
  • 1
    $\begingroup$ +1. Wow! A very interesting question indeed! Can you tell us an example of a molecule you're interested in? The answer will vary extremely depending on whether you're talking about systems like these: mattermodeling.stackexchange.com/a/555/5 or systems like these: mattermodeling.stackexchange.com/a/1079/5 or one of many other types of systems. Are you only trying to reach the CBS limit for energies, or polarizabilities too? There's the F12/R12 methods: mattermodeling.stackexchange.com/q/1134/5, there's transcorrelation methods, there's extrapolations methods, wavelets.. $\endgroup$ Aug 24, 2020 at 15:34
  • 1
    $\begingroup$ @TristanMaxson It's true that the counterpoinse correction wouldn't cost much more than doing the original calculation, but there is a "doubling" of the total cost because you have to do the calculation with ghost atoms (this is the calculation that you say is equal to or less costly than the original calculation, but it still is an extra calculation that normally you wouldn't want to do). F12 methods for example, can help you get to the CBS limit without doing a "second" calculation on ghost atoms, or using larger basis sets. They do have disadvantages which I gave in my last comment. $\endgroup$ Aug 24, 2020 at 15:39
  • $\begingroup$ That is all good to know, I have found in my experience that normally BSSE calculations are fairly inexpensive to run due to the fact normally the relaxation of the molecule is also a large factor in computational time. I could see how when doing high level theory with low level geometry that BSSE could really double the computational time though. $\endgroup$ Aug 24, 2020 at 15:42
  • 1
    $\begingroup$ I think BSSE is usually associated with SCF-level calculations, where it's especially a problem in small basis sets. Note that numerical basis sets (or highly contracted Gaussian basis sets) are much less susceptible to BSSE: since the minimal basis is exact for the non-interacting atom, there's no energy to be gained by using the functions on another atom. I don't think there's an issue for post-HF calculations, since they anyway require much larger basis sets to yield accurate correlation energies. At the minimum you'd use something like QZ, and BSSE shouldn't be an issue anymore. $\endgroup$ Aug 25, 2020 at 10:06