Abstract
The local density approximation and a range of nonhybrid gradient corrected density functionals (PW91, BLYP, PBE, revPBE, RPBE) have been assessed with respect to the prediction of geometries and spin‐state energy preferences for a range of homoleptic Fe(II)L~6~ and Fe(III)L~6~ complexes, where L = Cl^−^, CN^−^, NH~3~, pyridine, imidazole, H~2~O, OCH~2~ and tetrahydrofuran. While the qualitative spin‐state energies from in vacuo structure optimizations are reasonable the geometries are relatively poorly treated, especially for [FeCl~6~]^3−/4−^. Structural results for all the complexes are significantly improved by including environmental effects. The best compromise between structural and spin‐state predictive accuracy was obtained for the RPBE functional in combination with the COSMO solvation approach. This approach systematically overestimates the energetic preference for a low spin state, which is partly due to the well‐known effect of the lack of exact exchange in nonhybrid functionals and partly due to the larger solvation stabilization of low‐spin complexes that have shorter bond lengths and thus smaller molecular volumes than their high‐spin partners. Calculations on low spin [Fe(bipy)~3~]^2+^ and [Fe(phen)~3~]^2+^ and their ortho methyl substituted analogs, which are high spin at room temperature but cross over to low spin at low temperature, suggest the RPBE/COSMO combination generates low spin states which are too stable by approximately 13 kcal mol^−1^. © 2004 Wiley Periodicals, Inc. J Comput Chem 25: 1840–1848, 2004