[SY-C7] Symposium C-7

[SY-C7] Predictive simulations of crystal plasticity: multiscale or cross-scale?

Invited

^○Vasily V Bulatov¹, Alexander Stukowski², Luis A Zepeda-Ruiz¹, Tomas Oppelstrup¹ (1.Lawrence Livermore National Laboratory, United States of America, 2.Darmstadt University, Germany)

Prediction of crystal plasticity from atomic motion has been regarded over the years as a poster child context for multiscale modeling, given that the relevant scales were inaccessible to direct MD simulations and, at the same time, dislocation lines offer a natural, concise and accurate way to coarse-grain crystal microstructure. The method of Discrete Dislocation Dynamics (DDD) is regarded as a key and most challenging element in the multiscale modeling hierarchy developed to enable such predictions. We will discuss results of a recently completed direct Molecular Dynamic (MD) simulation of dislocation dynamics in which a single crystal of tantalum was compressed at rate 10⁵/s. Termed Livermore BigBig (LBB) simulation, LBB is by far the largest MD simulation ever performed. LBB generated a fully dynamic trajectory of over 2.1 billion atoms simulated over 5 microseconds of compressive straining. The simulation generated nearly 80 exabytes of recordable trajectory data a fraction of which was saved on disk in a highly compressed/post-processed form available for further analysis. As opposed to multiscale, LBB simulation can be regarded cross-scale being sufficiently large to be statistically representative of collective action of dislocations resulting in the macroscopic flow stress and yet fully resolved to every atomic “jiggle and wiggle”. Importantly, LBB is sufficiently large to be used as a critical test of the DDD-based multi-scale approach. To enable an “apples to apples” comparison, we carefully calibrated our DDD model (in ParaDiS) to match our LBB model at the level of single dislocation mobilities, shape and dimensions of the simulation volume, straining rate, temperature, straining axis, positioning and shape of initial dislocation sources, etc. Our first comparisons of the so-calibrated DDD simulations with LBB - which in this case serves as an exact benchmark for comparison - are far from satisfactory suggesting, in particular, that in our DDD model the dislocations do not know how to multiply. The exact origin of the uncovered discrepancies still unclear, we intend to make available to the community all the relevant atomistic simulation data (including the LBB results) thus challenging the DDD practitioners to match our LBB predictions.