Global optimization of additive potential energy functions: predicting binary Lennard-Jones clusters.
ABSTRACT We present a method for minimizing additive potential-energy functions. Our hidden-force algorithm can be described as an intricate multiplayer tug-of-war game in which teams try to break an impasse by randomly assigning some players to drop their ropes while the others are still tugging until a partial impasse is reached, then, instructing the dropouts to resume tugging, for all teams to come to a new overall impasse. Utilizing our algorithm in a non-Markovian parallel Monte Carlo search, we found 17 new putative global minima for binary Lennard-Jones clusters in the size range of 90-100 particles. The method is efficient enough that an unbiased search was possible; no potential-energy surface symmetries were exploited. All new minima are comprised of three nested polyicosahedral or polytetrahedral shells when viewed as a nested set of Connolly surfaces (though the shell structure has previously gone unscrutinized, known minima are often qualitatively similar). Unlike known minima, in which the outer and inner shells are comprised of the larger and smaller atoms, respectively, in 13 of the new minima, the atoms are not as clearly separated by size. Furthermore, while some known minima have inner shells stabilized by larger atoms, four of the new minima have outer shells stabilized by smaller atoms.