A comparison of geometry optimization with internal, cartesian, and mixed coordinates

Improvements in MO programs and computer speeds have dramatically increased the size of molecules that can be optimized by MO methods. This has put a considerable strain on existing geometry optimization techniques. For a given molecular system, the performance of a geometry optimization method can vary significantly depending on the coordinates used. Nonredundant internal coordinates (e.g., Z-matrix coordinates) are easily constructed and behave well for acyclic molecules but can be quite troublesome for cyclic systems. Cartesian coordinates behave well for rigid, cyclic molecules but can be quite poor for flexible, acyclic systems. Mixed cartesian and internal coordinates have been suggested as a means of combining the best of both approaches and are particularly convenient for cyclic systems with flexible substituents. The efficiency of these different approaches is compared for a number of molecules.