We propose an algorithm that is a combination of systematic variation of the torsions and Monte Carlo (or stochastic) search. It starts with a trial geometry in internal coordinates and with a set of preconditioned torsional angles, i.e., torsional angles at which minima are expected according to the chemical knowledge. Firstly, the optimization of those preconditioned geometries is carried out at a low electronic structure level, generating an initial set of conformers. Secondly, random points in the torsional space are generated outside the "area of influence" of the previously optimized minima (i.e., outside a hypercube about each minima). These random points are used to build the trial structure, which is optimized by an electronic structure software. The optimized structure may correspond to a new conformer (which would be stored) or to an already existing one. Initial torsional angles (and also final ones if a new conformer is found) are stored to prevent visiting the same region of the torsional space twice. The stochastic search can be repeated as many times as desired. Finally, the low-level geometries are recovered and used as the starting point for the high-level optimizations. The algorithm has been employed in the calculation of multi-structural quasi harmonic and multi-structural torsional anharmonic partition functions for a series of alcohols ranging from n-propanol to n-heptanol. It was also tested for the amino acid L-serine.
Keywords: conformational search; geometrical optimization; hindered rotors; stochastic methods; torsional anharmonicity.
Copyright © 2020 Ferro-Costas and Fernández-Ramos.