Systematics of Masses and Half-Lives of Nuclei with Neural Nets

Abstract

Ih the last few years a phenomenological approach to nuclear systematics based on multilayer feedforward neural networks has been under development[1,2,3]. Using suitable training sets,back- propagation and other related algorithms[4,5] are applied to teach such networks a given nuclear property. The networks are then asked to predict the property using test nuclei absent from the training set. Training and test sets are provided by the Brookhaven Nuclear Data facility. With proper architecture, coding schemes for input and output data,activation and error functions as well as pruning techniques,a number of networks can be produced that demonstrate high quality of performance in learning. Their predictive power can be competitive with that of traditional theoretical approaches. In this work,we study the nuclear masses and the half-lives of unstable β-decaying nuclear ground states using the following number of nuclides: 1882 and 1260 as learning sets and 627 and 423 as test sets respectively. Concerning masses, our work is a continuation of the work reported in refs [1,2]. It uses an enriched data basis and tries to achieve the highest possible performance with the smallest number of parameters. So far, our results for the root mean square error are comparable to those of ref. [2] and those derived by the mass fits of Mässon- Jänecket6! and Möller et al[7]. However, the number of the parameters used in the later fit is significantly smaller. Concerning half-lives,up to now there is no global model based on conventional nuclear theory. There are some models mainly for beta-decay. At present, most of our computer experiments have focused on the study of this decay channel.Our initial learning and test sets include 575 and 191 nuclei respectively. The performance of the models developed in learning is comparable to that of Klapdor et al[8]. The next step is to improve the predictive performance and to study and include the nuclides with other decay channels. The aim of this work is the production of global models of masses and half-lives of very good quality