Ordinarily, the strength and plasticity properties of a metal are defined by dislocations—line defects in the crystal lattice whose motion results in material slippage along lattice planes1. Dislocation dynamics models are usually used as mesoscale proxies for true atomistic dynamics, which are computationally expensive to perform routinely2. However, atomistic simulations accurately capture every possible mechanism of material response, resolving every “jiggle and wiggle”3 of atomic motion, whereas dislocation dynamics models do not. Here we present fully dynamic atomistic simulations of bulk single-crystal plasticity in the body-centred-cubic metal tantalum. Our goal is to quantify the conditions under which the limits of dislocation-mediated plasticity are reached and to understand what happens to the metal beyond any such limit. In our simulations, the metal is compressed at ultrahigh strain rates along its  crystal axis under conditions of constant pressure, temperature and strain rate. To address the complexity of crystal plasticity processes on the length scales (85–340 nm) and timescales (1 ns–1μs) that we examine, we use recently developed methods of in situ computational microscopy4,5 to recast the enormous amount of transient trajectory data generated in our simulations into a form that can be analysed by a human. Our simulations predict that, on reaching certain limiting conditions of strain, dislocations alone can no longer relieve mechanical loads; instead, another mechanism, known as deformation twinning (the sudden re-orientation of the crystal lattice6), takes over as the dominant mode of dynamic response. Below this limit, the metal assumes a strain-path-independent steady state of plastic flow in which the flow stress and the dislocation density remain constant as long as the conditions of straining thereafter remain unchanged. In this distinct state, tantalum flows like a viscous fluid while retaining its crystal lattice and remaining a strong and stiff metal.
This work was performed under the auspices of the US Department of Energy by Lawrence Livermore National Laboratory under contract W-7405-Eng-48. This work was supported by the NNSA ASC programme. Computing support for this work came from the Lawrence Livermore National Laboratory (LLNL) Institutional Computing Grand Challenge programme and Jülich Supercomputing Centre at Forschungszentrum Jülich, Germany.
Extended data figures
Dislocation multiplication from the initial sources results in the development of a dense dislocation network.
Crystal containing dislocations sources (loops) is subjected to uniaxial compression along the  axis at a constant true straining rate of 2.78.108 1/s (x25). The simulation volume contains about 268 million atoms of tantalum. The video sequence progresses through extension of the initial hexagon-shaped loops, to dislocation collisions resulting in the formation of dislocation junctions, to an increasingly dense dislocation network. Dislocation positions, shapes and Burgers vectors were extracted using the DXA algorithm5. All atoms and defects other than dislocations, such as vacancies, interstitials and clusters, are removed for clarity. The green lines represent ½<111> dislocations and the pink lines depict <100> junction dislocations.
In this MD simulation a crystal containing dislocations sources (loops) was subjected to uniaxial compression along the  axis at a constant true straining rate of 5.56.108 1/s (x50). The simulation volume contains about 33 million atoms of tantalum. This video sequence progresses through extension of the initial loops, to nucleation of embryonic twins on screw dislocations, to rapid propagation and growth of twinning particles. The meaning of colours is as defined in the caption to Fig. 1: the outer surfaces bounding the twins are coloured light grey whereas the insides of twin particles are coloured red, yellow, magenta or cyan depending to each twin's rotational variant.
This MD simulation was performed on a brick-shaped tantalum crystal with the ratio of initial box dimensions 1:2:4. After full compression along Z axis to ¼ of its initial dimension the brick’s shape becomes 2:4:1 (due to Poisson’s expansion in two lateral dimensions, the brick’s volume remains very nearly constant under compression) another MD simulation starts in which the brick is compressed along its now longest Y-axis. After the second compression cycle is completed, the brick is compressed along its now longest X-axis. After three compression cycles the brick recovers its initial shape 1:2:4 and one more Z-axis compression cycle is performed (see related stress-strain plots in Extended Data Figure 5).
This simulation was performed at rate 1.11.107 1/s (x1) from a configuration attained past yield under pre-straining at rate 5.55.107 1/s (x5). Reduction in dislocation density can be observed over the first few frames immediately following the sudden drop in the straining rate from x5 to x1 at time t=0. Subsequently the network attains a dynamic steady state in which dislocation multiplication is balanced by dislocation annihilation. Taken at more frequent time intervals, this sequence reveals various events in the life of dislocations in greater detail than in the other videos. One can observe that dislocation motion is not steady but proceeds in a stop-and-go manner which is also revealed in Extended Data Fig. 9a.