Performance assessment of semiempirical molecular orbital methods in the structural prediction of Sb(III) and Bi(III) complexes.

In this paper we carried out a systematic study in order to assess the quality of some semiempirical methods (AM1, PM3 and PM6), comparing predicted structural properties of many Sb(III) and Bi(III) complexes with the corresponding experimental data, indicating which one is more appropriate to describe the structure of such compounds. Root-mean squared deviation (RMSD) and unsigned mean error (UME) were used to evaluate the accuracy of the semiempirical methods to predict the ground state geometries of complexes with many ligand types. Our results have shown that, in general, PM3 predicts more accurately the geometry of Sb(III) complexes, being considered by us as the method of choice to study Sb(III) complexes with a great variety of ligands. PM6 is indicated as the method of choice to study Bi(III) complexes with many types of ligands and also to study Sb(III) thiocompounds, even though PM6 showed an inability to reproduce Sb-N bonds for complexes with flexible ligands, presenting an average deviation of 71.5 % compared the X-ray data.