A PARALLEL IMPLEMENTATION OF LIGHT-BINDING MOLECULAR-DYNAMICS

We introduce an efficient and scalable parallel implementation of tigh t-binding molecular dynamics which employs reordering of the atoms. Th is implementation also exploits the sparse character of the Hamiltonia n matrix. Reordering allows our algorithm to scale well on parallel ma chines since it assures data locality. A sparse storage format and a s tabilized parallel Lanczos eigensolver allow consideration of large pr oblem sizes relevant to materials science. We present a showcase appli cation to a 1000-atom silicon sample.