We present a method for total-energy minimizations and molecular-dynamics simulations based either on tight-binding or on Kohn-Sham Hamiltonians. The method leads to an algorithm whose computational cost scales linearly with the system size. The key features of our approach are (i) an orbital formulation with single-particle wave functions constrained to be localized in given regions of space, and (ii) an energy functional that does not require either explicit orthogonalization of the electronic orbitals, or inversion of an overlap matrix. The foundations and accuracy of the approach and the performances of the algorithm are discussed, and illustrated with several numerical examples including Kohn-Sham Hamiltonians. In particular, we present calculations with tight-binding Hamiltonians for diamond, graphite, a carbon linear chain, and liquid carbon at low pressure. Even for a complex case such as liquid carbon—a disordered metallic system with differently coordinated atoms—the agreement between standard diagonalization schemes and our approach is very good. Our results establish the accuracy and reliability of the method for a wide class of systems and show that tight-binding molecular-dynamics simulations with a few thousand atoms are feasible on small workstations.

• Received 13 April 1994

DOI:https://doi.org/10.1103/PhysRevB.50.4316