Application of Deep Neural Network in Finding of Repulsive Part of Molecular Potential Based on Dispersed Emission Spectra

A new method of obtaining parameters of the repulsive part of the analytical potential energy curve of a diatomic molecule, based on the deep neural network approach, is presented. The method was successfully tested on artificially created spectra as well as an experimental bound -> free spectrum of the G0(u)(+) (6(1) P-1)(nu' = 39) -> X0(g)(+) transition in Hg-2. The method was tested for a Morse and four-parameter expanded Morse oscillator potentials but can be easily adaptable to other forms of analytical potential energy curves.