Artificial neural network applied for predicting rainbow trajectories in atomic and molecular classical collisions

A simple artificial neural network (ANN) is developed and applied to collision processes. A general discussion of how ANNs can be introduced to study general phenomena in scattering problems is presented and neural networks are proposed to predict classical rainbow trajectories in atomic and molecular collisions, As a result of modeling the collision process, based on the neural network approach analytical equations were obtained to calculate classical atomic and molecular rainbow trajectories, However, these analytical results just translate the behavior of the input/output data and do not contain any general physical meaning. Although a fitting procedure could be easily used in the present case, the cost of function approximation using ANNs increases only linearly with the number of input variables. This contrasts with classical polynomial fitting procedures for which the computational cost increases exponentially with the input space dimension. This makes the ANN approach worth considering when modeling scattering processes, as shown throughout this paper. At last, an articial network strategy is pointed out to study inversion problems in collision processes. (C) 1997 American Institute of Physics. [S0021-9606(97)02847-X].