Frozen Core/Frozen Virtual Approximation

Freezing core orbitals in a correlated calculation can save a lot of time. More importantly, for most basis sets (except, for example, Dunning's correlation-consistent core-valence basis sets denoted cc-pCVXZ, etc), it's actually a bad idea to correlate core electrons, since artifacts can arise because there aren't any core-correlating basis functions in the basis set. As a practical matter, most molecular properties (e.g., equilibrium geometries, vibrational frequencies) seem to be very insensitive to freezing or not freezing core, at least for first-row atoms, so don't get too worried if our copy of ACES II does not allow frozen core optimizations. However, it is a good idea to freeze core whenever you can.

One can also freeze virtual orbitals (making them never populated in any determinant). This is really only well justified if the orbitals have very high SCF energies (or very low population numbers, if they are natural orbitals). Usually this only happens in Dunning type DZP or TZ2P, etc., basis sets, where the antisymmetric combination of core orbitals is present in the basis and can certainly be removed by the frozen virtual approximation. Pople basis sets and the new correlation-consistent basis sets (cc-pVXZ) do not have these very high lying orbitals.