Application of neural networks: a molecular geometry optimization study

Abstract

Summary form only given. Optimization algorithms are iterative procedures that evolve from guessed starting points (SP) to the desired global minimum. Their performance can be greatly improved, if a neural network (NN) is created to select suitable SP. In this paper we consider the use of trained NN to select possible ground-state geometries for silicon clusters. A genetic algorithm is initial population energy optimization. For convenience, a cluster's geometry is described as a piling up of plane layers of atoms.