Descriptor for positron binding to atoms

Abstract

Systematic theoretical data of positron binding to atoms [C. Harabati, V. A. Dzuba, and V. V. Flambaum, Phys. Rev. A 89, 022517 (2014)] are handled by machine learning techniques. Despite the existence of unsystematic errors in the calculations, it becomes possible to identify, through bound-unbound classification, a descriptor as a function of just the polarizability and ionization potential of the isolated atoms. Details of the bonding mechanism appear to be less relevant and the binding seems to be described by simply these two atomic properties, with large predominance of the polarizability. The descriptor is shown to be useful in the discrimination of binding energies of atoms with approximately equal polarizabilities as well as to correct some possibly wrong energy data. Identified arguably wrong cases are withdrawn from the data set so that an adjusted simple relation for the binding energy as a function of the descriptor is obtained, allowing corrections and predictions of binding energies extended to atoms up to Rf (104), except lanthanides and actinides.