NEURAL MODELING OF TORSIONAL POTENTIAL HYPERSURFACES IN NONRIGID MOLECULES

This paper presents a study on the ability of neural networks to model torsional potential hypersurfaces in non-rigid molecules. Using the p otential function for the methyl torsion of acetaldehyde, we find that standard models do not represent accurately periodic function in its full range of definition. However, the functions are correctly describ ed in any non-periodic zone. This behaviour arises from the periodic n ature of the actual function and from the local character of the train ing methods. A new periodic activation function is defined, which enha nces greatly the results for the periodic and non-periodic cases. The new activation function permits description of any periodic function o f any number of arguments using a neural network with only one hidden layer. (C) 1998 Elsevier Science Ltd. All rights reserved.