The CI expansion is variational and, if the expansion is complete (Full CI), gives the exact correlation energy (within the basis set approximation). The number of determinants in Full CI grows exponentially with the system size, making the method impractical for all but the smallest systems. For this reason the CI expansion is usually truncated at some order, for example CISD, where only singly and doubly excited determinants are considered. Brillouin's Theorem states that singly excited determinants do not mix with the HF determinant [35]. Therefore CISD is the cheapest worthwhile form of CI, yet this method scales as O(N6) where N is the size of the system.
The other main problem with truncated CI is that it is not size consistent. For CISD, an approximate way to correct for these effects is to introduce the Davidson correction [36]
where c0 is the coefficient of the Hartree-Fock wavefunction in the normalized CISD wavefunction.