On the size consistency of multireference CEPA methods

The size consistency of various multireference coupled electron pair approaches (MRCEPA) is analysed. These methods include the MRCEPA0, MR averaged coupled pair functional (MR-ACPF), MR averaged quadratic coupled cluster (MR-AQCC), MR (singles and doubles) CEPA (MRDCEPA) and the recently proposed MR averaged CEPA (MR-ACEPA) methods. Test calculations on the O2 dimer show that a correct treatment of the variationally included (VI) terms such as in the MRDCEPA and MR-ACEPA methods is necessary for obtaining nearly size-consistent results. On the other hand, the results strongly suggest that it is impossible to obtain exact size consistency with any method employing the diagonal shift formalism. The analysis also shows that the size consistency error is not significantly affected by the exclusion principle violating (EPV) terms.