Are frozen-core approximations for the elements of first or second row accurate for potentially extremely short bond distances present in the molecule

Question

Yes, it is about my molecule that you've heard of (but never seen) in my previous questions. The molecule, which contains only C, Al, B, Cl, Mg, N, O, S and Si, is macromolecular with quite crowded monomers- cyanide-like ultrashort bonds, nonbonded distances a mere 0.3Å longer than the sum of covalent radii, bridge-bonded (as in the two borons of diborane) Al-Al distances of less than 2Å and so on.

I'm using a double hybrid so I need to adjust the MP2 part of the total cost- are frozen-core approximations appropriate for bond lengths, vertical ionisation energies/electron affinities, and NBO orbital occupancies/populations of specific orbitals of bonds given as results by NBO, or would a full-electron treatment be more appropriate for the calculation of the aforementioned properties? I'm at the level RI-B2PLYP-D3/def2-TZVPPD/def2-TZVPPD.

Answer 1

Due to the errors associated with your choice of functional (even if it's a double hybrid functional), I think there is not much value in unfreezing the core with regards to getting more accurate energies in the realm of "extremely short bond distances".

The regime of "extremely short bond distances" is one in which no matter what you do, there will likely be a fairly big discrepancy between what you get with two different equally accurate methods, when compared to what you'd find at other less-extreme parts of the potential. DFT (even with double-hybrid functionals) is already so poor at capturing electron correlation, so I wouldn't read very much into its results in the "extremely short bond distance" regime, and I would say the same for the much more accurate coupled cluster methods and even FCI (which is numerically exact for a given basis set and active space).

Therefore I don't see much value in adding so many extra electrons and orbitals to your active space in this regime which is so-much limited by other issues that likely can't be fixed at all.