Geometry-Corrected Quadratic Optimization Algorithm for NDDO-Descendant Semiempirical Models

J Chem Theory Comput. 2024 Dec 18. doi: 10.1021/acs.jctc.4c01070. Online ahead of print.

Abstract

The long-held assumption that the optimization of parameters for NDDO-descendant semiempirical methods may be performed without precise geometry optimization is assessed in detail; the relevant equations for the analytical evaluation of the geometry-corrected derivatives of molecular properties that account for changes in the optimum geometry are then presented. The first and second derivatives calculated from our implementation of MNDO are used for a limited reparameterization of 1,113 CHNO molecules taken from the PM7 training set, demonstrating an improvement over the PARAM program used in the optimization of parameters for the PMx methods.