virtual MOs (Doug Smith)

 The ordering and the coefficients of the virtual orbitals will indeed
 change for a sequence of (single determinant) HF jobs with changing
 basis sets. The virtuals resulting from such a calculation will not
 usually be representative of "good" antibonding MO's , since it is the
 occupieds that are being optimized.
 In an MCSCF calculations those orbitals which are placed into the active
 space will resemble actual antibonding MO's more closely, and for a multi-
 reference wavefunction there is no "ordering" of those partially
 orbitals i the active space. For the purpose of interpretation, natural
 orbitals with the appropriate occupation numbers would be more suitable.
 You would expect the HF orbitals to change quite a bit if you use them as
 initial guess for the MCSCF orbitals (within the valence space).
 One needs to be careful in attempting to compare a MR wavefunction to
 a single determinant wavefunction, and density differences would contain
 more information (and meaning) than just the orbitals.