Ground-state of silicon clusters by neural network assisted genetic algorithm

The recently proposed Neural Network Assisted Genetic Algorithm, NAGA, is applied to investigate the ground-state geometries of silicon clusters ranging from 16 to 21 atoms. NAGA combines the ability of artificial neural networks to restrict the search space with the speed of the genetic algorithm. The potential energy surface is calculated according to the tight-binding approximation. Results obtained through NAGA are compared to other candidate ground-state geometries available in the literature. NAGA's performance was also compared to standard genetic algorithm. It proved to be at least six times faster than the genetic algorithm and it did find the desired minimum in every trial for every silicon cluster size tested. (C) 2003 Elsevier B.V. All rights reserved.