Multimillion atom simulations of dynamics of wing cracks and nanoscale damage in glass, and hypervelocity impact damage in ceramics

We have developed scalable parallel algorithms for first-principles based predictive atomistic simulations of materials. We have achieved parallel efficiency 0.998 for 134 billion-atom molecular dynamics (MD), 1.06 billion-atom reactive force-field MD, and 11.8 million-atom (1.04 trillion electronic degrees-of-freedom) quantum-mechanical MD in the framework of the density functional theory on 131,072 BlueGene/L processors. We have performed up to 540 million-atom MD simulations to study: (1) initiation, growth and healing of wing cracks in confined silica glass; and (2) damage initiation during hypervelocity impact on advanced ceramics. © 2007.