Global minimization of Lennard-Jones functions on transputer networks

This paper presents a three-phase computational procedure for minimizing molecular energy potential functions of the pure Lennard-Jones type. The first phase consists of a special heuristic for generating an initial atomic configuration. In the second phase a global minimization method is applied to compute a configuration close to the optimal solution. Finally, the third phase uses this configuration as the starting point for a local minimization method. Since the second and third phases are very suitable for parallel implementation, we describe briefly our implementation of the method in a parallel version of C on a transputer network, and we exhibit numerical results for approximate optimization of clusters with up to 20,000 atoms.