Efficient geometry optimization of molecular clusters

Abstract

By combining valence coordinates (stretches, bends, and torsions) to describe intramolecular degrees of freedom with inverse distance coordinates for intermolecular degrees of freedom, we derive an efficient set of coordinates for the geometry optimization of molecular clusters. We illustrate the efficacy of our new coordinates by considering randomly generated clusters of dihydrogen and water molecules. Compared to optimizations in Cartesian coordinates, the number of cycles required for convergence is reduced by up to a factor of 30. In addition, for the dihydrogen clusters, optimizations using our new cluster coordinates consistently converge to lower energy structures than the corresponding Cartesian optimizations. Our method is far more efficient than optimizations using Z-matrix coordinates, and it avoids all problems with near-linear bond angles that are endemic with a Z-matrix description of the cluster geometry. Additionally, by constraining all the intramolecular degrees of freedom in a completely automated manner, we are able to carry out full rigid-body optimizations. © 2000 John Wiley & Sons, Inc. J Comput Chem 21: 69–76, 2000