Abstract

Semiempirical (NDDO-based, MNDO -like) molecular orbital (MO) theory has led a life in the shadows for several decades, probably because “more respectable” techniques, such as density-functional theory, have become applicable to quite large molecules, including transition-metal complexes. However, the great strengths of semiempirical MO theory (speed, scaling, one-electron properties, and excited states) remain.

Two independent parameterizations (PM6 and AM1 *) for the first-row transition metals are now available, so that these elements can be treated successfully. Classical dispersion potentials have been added to standard parameterizations in the same spirit as DFT -D.

We have developed a massively parallel code that brings calculations on 100,000 atoms within reach on 1,000 processors with high efficiency, but also gives super-scalar performance on 8-32 processors, for instance on dual or quad-core nodes. Similarly, geometry optimizations on datasets of 100,000 drug-sized molecules and more are possible in a weekend on an eight-core node. These capabilities open new possibilities and applications of semiempirical MO theory.

Among these is the use of Pulay’s UNO -CAS technique to give reliable band gaps for semiconductors, which, combined with a newly developed direct optimization algorithm for the UHF density matrix, also allows calculations on conductors for the first time.

These developments provide a powerful tool for materials modeling, in particular for spectroscopic and electronic properties, but also for high-quality quantitative structure-property relationships without resorting to an atoms-and-bonds picture of molecules.