Neural Network Predictions of Atomic Form Factors and Incoherent Scattering Functions

Abstract

In order to predict atomic form factors and incoherent scattering functions which are used to calculate the coherent and incoherent total scattering cross sections, a technique based on artificial neural networks of the multilayer type was implemented. In this context, two neural models have been developed and compared with those in the literature. This study revealed both the accuracy of the results obtained and the effectiveness of the designed model. The mean relative error for the least estimated property does not exceed 16.5 %. The software realized in this way give a prediction of the above parameters for the input variables Z: Atomic number, x: sin(ϑ/2)/λ and E: Photon energy, and it provides users with flexibility for prediction. The advantages of this technique lie in its very fast handling, due to its ease of use, and in the two integrated networks, which it guarantees for a variety of input parameters such as atomic number, photon energy, and momentum transfer variable.