4
$\begingroup$

I'm planning to analyse an extremely complex molecule on NWChem(it's practically the only free computational package available package here, and the molecule is so large that running a seemingly simple HF/STO-3G calculation the Gaussian module on e.g. the WebMO Demo server would certainly go way past the 30s limit on the Demo), via DFT, using the Karlsruhe basis sets and subsequent complete basis set(CBS) extrapolation.

After not being able to find any CBS methods (aside from the one using MP2/aug-cc-CBS, which is much too expensive for a gigantic molecule) implemented in the NWChem user manual, I decided to look for a way to extrapolate to the CBS limit by hand.

A simple Google search led me to a page talking about how cc-pVNZ SCF (resp. corr) energies converge at a speed of O(exp(-a√n))(resp. O(z^(-n))), a(resp. z) being fixed constants, but I could not find any such equation, to solve by hand, for the def2 basis sets.

The only extrapolation I could do by hand at the moment is the one that assumes an O(exp(-b*n)) convergence with both the b and the constant implicit in the big-O expression being unknown. This requires a three-point extrapolation(I know what I'm talking about, I'm literally majoring in mathematics); since the def2 basis sets are (def2-)SVP, TZVPP, and QZVPP, this requires including a part of the def2-SVP energies in the extrapolation, a thing that is not favored in QC(N.B. the automatic CBS extrapolation that ORCA says it supports uses a two-point extrapolation, which would mean that the constant b, or its equivalent thereof should the equation be slightly different from what is written above, is explicitly known).

So I'm stuck at a dead end, unless I get to know how to extrapolate NWChem results to the CBS limit(even if I end up having to export the NWChem results and put them in a seperate package to do it- one already needs to do it that way to compute NBO's via NWChem IIRC), or (more simply) find an explicit systematic convergence rate for the def2 basis sets. What should I do?

$\endgroup$
3
  • $\begingroup$ What is your need? Evaluate the geometry, energy or other properties? $\endgroup$
    – NickZ
    Jul 29, 2022 at 7:21
  • $\begingroup$ I have done quite a few extrapolations to the CBS limit, and I would say a good approach is to literally fit all of the functional forms people have suggested and just report an uncertainty. All extrapolations to the CBS limit are empirical techniques. If you can afford to do the quadruple-zeta calculation, then all the extrapolations should be very similar. Also, why not use the Dunning correlation-consistent basis sets, since they're made for this? $\endgroup$
    – jheindel
    Jul 31, 2022 at 6:40
  • $\begingroup$ There is a reason why I said "via DFT". $\endgroup$ Sep 19, 2022 at 1:58

0