Sorry for delayed reply.
Unfortunately, what you propose is impossible both for technical and, well, physical reasons. You cannot automatically transfer the orbitals from the calculation with smaller basis set (even from a single-reference calculation). Some other programs can use the result of the calculation in a smaller basis set as a starting guess for a larger one, but not FireFly. However, it is meaningless to do this in a multireference calculation.
So, you cannot do an extrapolation. If you want to study the convergence of CASSCF results wrt basis set, you'll have to perform calculations with each basis set separately paying special attention to the active space (it should be the same in all calculations). There is no general recommendations about reordering: every time you should manually examine orbitals and reorder them when necessary moving unnecessary (e.g., diffuse) orbitals off the active space.
The more important issue in CASSCF is well balanced active space. The most obvious case is that all orbitals constituting a shell (e.g., two degenerate pi orbitals in benzene or 4f shell in a lanthanide) should enter the active space together. However, there are some less evident cases of poorly balanced active spaces. Sometimes they just result in wrong transition energies, and sometimes, in artificial symmetry breaking. So be careful.
As for the use of smaller basis sets, they are not recommended for high-level quantum chemistry. Larger basis sets (at least, double and triple zeta with polarization and, when necessary, diffuse functions) are more flexible and better reproduce the electron density in the molecules both in single-reference and multireference methods.
For your particular molecule (~20 2nd row atoms + ~30 H atoms), 6-31G(d,p) will be OK. Probably, you may try 6-311G(d,p). From my experience, basis sets such as cc-pVDZ and cc-pVTZ do not provide substantial improvements as compared with 6-31G(d,p) and 6-311G(d,p), respectively, although they contain much more functions.
On Sat Apr 4 '15 9:08pm, Добромир Антонов Калчевски wrote
>Is it possible, in the case of CASSCF, to correctly transit a calculation from a smaller to a larger basis set, and by correctly - I mean as Active Space. Considering I have to reorder certain MOs for the calculations with the smaller basis, will I get the correct MOs from the larger one, without new reordering ? Also, since basis sets with diffusive and polarizing functions tend to substitute MOs with not only more, but qualitatively different MOs, which is more visible in the Vacant part, am I to expect that a sufficient active space can be sustained at a larger basis. What would be the way to do it, in terms of correct calculation steps and keywords. Let's say calculating transitions for excited states will give all sorts of differences in the selection of the active MOs, depending on the basis.
>Can I do an extrapolation such as STO-3G -> 4-31g -> 6-31g -> 6-31g* -> cc-pVTZ without polluting the active space with orbitals, which are not supposed to be a part of it, AND without losing any of the ones which should be a part.
>Also, is there a significant improvement of time requirements if I do this trick or not ? I know a larger basis will converge with less steps. Is there a chance of qualitative difference so big, that my calculation can even fail under a smaller basis set ? Let's say I'm searching for a Conical Intersection of a molecule with close to 20 2nd row atoms + close to 30 hydrogen atoms, and quite conjugated, at least at Ground State.
>Also what would be the output of Firefly 810, in case there is no conical intersection ? Roughly said, as a number of steps or proportion of time (compared to a case with conical intersection) am I for a long wait ?
>My computer is an i7 3770k, 16GB. I have access to another one - x5650, 98GB. Can someone with a similar experience give me some kind of an estimate of the time in front of me, if I've picked let's say 8el 8orb active space, and it happens to be correct.
>Thank you in advance,