High‐throughput screening of spin states for transition metal complexes with spin‐polarized extended tight‐binding methods

Abstract The semiempirical GFNn‐xTB () tight‐binding methods are extended with a spin‐dependent energy term (spin‐polarization), enabling the fast and efficient screening of different spin states for transition metal complexes. While GFNn‐xTB methods inherently can not differentiate properly between high‐spin (HS) and low‐spin (LS) states, this shortcoming is corrected with the presented methods termed spGFNn‐xTB. The performance of spGFNn‐xTB methods for spin state energy splittings is evaluated on a newly compiled benchmark set of 90 complexes (27 HS and 63 LS complexes) containing 3d, 4d, and 5d transition metals (termed TM90S) employing DFT references at the TPSSh‐D4/def2‐QZVPP level of theory. The challenging TM90S set contains complexes with charges between 4 and +3, spin multiplicities between 1 and 6, and spin‐splitting energies that range from 47.8 to 146.6 kcal/mol with a mean average of 32.2 kcal/mol. On this set the (sp)GFNn‐xTB methods, the PM6‐D3H4 method, and the PM7 method are evaluated with spGFN1‐xTB yielding the lowest MAD of 19.6 kcal/mol followed by spGFN2‐xTB with 24.8 kcal/mol. While for the 4d and 5d subsets small or no improvements are observed with spin‐polarization, large improvements are obtained for the 3d subset with spGFN1‐xTB yielding the smallest MAD of 14.2 kcal/mol followed by spGFN2‐xTB with 17.9 kcal/mol and PM6‐D3H4 with 28.4 kcal/mol. The correct sign of the spin state splittings is obtained with spGFN2‐xTB in 89% of all cases closely followed by spGFN1‐xTB with 88%. On the full set, a pure semiempirical vertical spGFN2‐xTB//GFN2‐xTB‐based workflow for screening purposes yields a slightly better MAD of 22.2 kcal/mol due to error compensation, while being qualitative correct for one additional case. In combination with their low computational cost (scanning spin states in seconds), the spGFNn‐xTB methods represent robust tools for pre‐screening steps of spin state calculations and high‐throughput workflows.