PREDICTION OF ATOMIC IONIZATION-POTENTIALS I-III USING AN ARTIFICIAL NEURAL-NETWORK

A simple three-layer back-propagation neural network is shown to learn the complex relationship between the electronic structure and the fir st three ionization potentials (log{IP(eV)}) of 222 atoms and ions for which spectroscopic data have been determined. The neural network was trained to a root mean square error of 0.043 log units and was subseq uently used to predict ionization potentials which were not included i n the training set and values which have not been experimentally deter mined. The neural network predictions are in very good agreement with experimental values not included in the training data set. In addition , the neural network predicts IP for elements 73, 104, and 105 which a re also in good agreement with those values predicted by highly sophis ticated multiconfiguration Dirac-Fock (MCDF) calculations.