Extending the active space in multireference configuration interaction calculations of magnetic coupling constants

In the last two decades, difference dedicated configuration interaction (DDCI) calculations have been outlined as a rather accurate and a successful approach for the evaluation of magnetic coupling constants J. The DDCI space contains all single and double semi-active excitations on the top of the complete active space, which is only composed of the magnetic orbitals. The main drawback of the procedure is its computational cost, in particular for systems involving several metallic centres with S > ½. To overcome this constraint, an alternative strategy has been proposed in the past which consists in restricting the CI expansion to single excitations on the top of an extended active space, involving a few doubly occupied and in some cases virtual orbitals of ligand character. This strategy reduces considerably the size of the CI matrices, and then makes possible to deal with larger and more complex systems. However, the so obtained J values are overestimated compared to the corresponding DDCI and to the experimental values. We analyze here the origin of this behavior in the light of a recent paper where the role of ligand-to-metal charge transfer excitations on the quantitative description of J has been re-examined.