Low-order moment expansions to tight binding for interatomic potentials: Successes and failures

We discuss the use of moment-based approximations to tight binding. Using a maximum entropy form for the electronic density of states, we show that a general interatomic potential can be defined that is suitable for molecular-dynamics simulations and has several other desirable features. For covalent materials (C and Si), properties where the atoms are in equivalent environments are well converged at low-order moments. For defect environments, which offer a more critical (and relevant) test, the method is found to give less satisfactory results. For example, the vacancy formation energy for Si is too low by ∼2 eV at 10 moments relative to exact tight binding. Attempts to improve the accuracy were unsuccessful, leading to the conclusion that potentials based on this approach are inadequate for covalent materials. We speculate that this may be a deficiency of low-order moment methods in general. For metals, in contrast to the covalent systems, we find that the low-order moment approach is better behaved. This finding is consistent with the success of existing empirical fourth-moment potentials for metals