Large scale ab initio calculations of extended defects in materials: screw dislocations in bcc metals

Abstract

Ab initio methods, based on the Density Functional Theory (DFT), have been extensively used to study point defects and defect clusters in materials. Present HPC resources and DFT codes now allow similar investigations to be performed on dislocations. The study of these extended defects requires not only larger simulation cells but also a higher accuracy because the energy differences, which are involved, are rather small, typically 50-to-100 meV for supercells containing 50-to-500 atoms. The topology of the Peierls potential of screw dislocations with 1/2 <111>Burgers vector, i.e. the 2D energy landscape seen by these dislocations, is being completely revisited by DFT calculations. From results obtained in all body-centered cubic (bcc) transition metals, except Cr (V, Nb, Ta, Mo, W and Fe), using the PWSCF code, which is part of the Quantum-Espresso package, we concluded that the 2D Peierls potentials have two common features: the single-hump shape of the barrier between two minima of the potential, and the presence of a maximum – and not a minimum as predicted by most empirical potentials – around the split core. In iron, the topology of the Peierls potential is reversed compared to the classical sinusoidal picture: the location of the saddle point and the maximum are indeed inverted with unexpected flat regions. The first results obtained within the framework of the PRACE project, DIMAIM (DIslocations in Metals using Ab Initio Methods), started at the beginning of 2013, will also be presented. In particular, in order to address the twinning-antitwinning asymmetry often observed in bcc metals and regarded as the major contribution to the breakdown of Schmid's law, we have determined the crystal orientation dependence of the Peierls stress, i.e. the critical stress required for dislocation motion. These computationally most expensive simulations were performed on the PRACE Tier-0 system at Barcelona Supercomputing Center (Marenostrum III). The scalability results using the various parallelization levels of the PWSCF code up to 10 000 cores will be presented.