Split-localized orbitals can yield stronger configuration interaction convergence than natural orbitals

Abstract

The convergence of configuration interaction (CI) expansions depends upon the orbitals from which the configurations are formed. Since their introduction half a century ago, natural orbitals have gained an increasing popularity for generating rapidly converging CI expansions and the notion has become widespread that they always yield the fastest CI convergence. It is shown here that, in fact, certain localized orbitals often yield a better CI convergence than natural orbitals, as measured by a wave function criterion as well as by an energy criterion.