Tight-Binding Linear Scaling Method Applications to Silicon Surfaces

Abstract

The past years have witnessed impressive advances in electronic structure calculation, especially in the complexity and size of the systems studied, as well as in computation time. Linear scaling methods based on empirical tight-binding Hamiltonians which can describe chemical bonding, and have a computational time proportional to the number of atoms N in the system, are of particular interest for simulations in material science. By contrast conventional diagonalization schemes scale as $N^{3}$. In combination with judiciously fitted parameters and an implementation suited for MD, it is possible to apply such \emph{O(N)} methods to structural, electronic and dynamical properties of large systems which include up to 1000 atoms on a workstation. Following a brief review of a tight-binding based linear-scaling method based on a local orbital formulation and of parametrizations appropriate for covalently bonded systems, we present recent test calculations on \emph{Si(111)-5 $\times$}5 and \emph{Si(001)-c(4$\times$}2) reconstructed surfaces in this framework, and compare our results with previous tight-binding and \emph{ab-initio} calculations.