A dynamical simulated annealing approach to the electronic structure of liquid metals

Abstract

The authors present a dynamical simulated annealing approach to the self-consistent calculation of the electronic structure of liquid metals. Models for the atomic structure are generated using a classical microcanonical molecular dynamics simulation based on interatomic potentials derived from pseudopotential perturbation theory. The electronic structure is calculated by numerically integrating the dynamical simulated annealing equations of motion for the electron states at fixed atomic coordinates, using the same pseudopotential. Detailed results are presented for liquid Al, Si, Ge, As and Te. As far as experimental information is available, the calculated electronic density of states is in good agreement with the photoemission spectra. The dynamical simulated annealing calculations are compared with electronic structure calculations based on minimal basis sets such as the linear-muffin-tin-orbital method. The authors find that a comparable accuracy, the dynamical simulated annealing approach reduces the computational effort for 64-atom models by about a factor of ten. Compared to a full density-functional molecular-dynamics approach the present method achieves self-consistency between the atomic and the electronic structure only at the level of a linear-response approach. For good liquid metals such as Si the result of the present approach based on a combination of perturbation and ab initio methods leads to results equivalent to those based on full density-functional molecular-dynamics calculations, but it requires only about 1% of the computational effort. For molten materials close to a semiconductor/semimetal transition (liquid As and Te) covalent bonding effects however are expected to have a nonnegligible influence on the electronic density of states at the Fermi level. Here the combination of the present approach with full density-functional MD calculations should help to get accurate results at a much lower computational effort.