Machine learning reveals strain-rate-dependent predictability of discrete dislocation plasticity

Predicting the behaviour of complex systems is one of the main goals of science. An important example is plastic deformation of micron-scale crystals, a process mediated by collective dynamics of dislocations, manifested as broadly distributed strain bursts and significant sample-to-sample variations in the response to applied loading. Here, by combining large-scale discrete dislocation dynamics simulations and machine learning, we study the problem of predicting the fluctuating stress-strain curves of individual small single crystals subject to strain-controlled loading using features of the initial dislocation configurations as input. Our results reveal an intriguing rate dependence of deformation predictability: For small strains predictability improves with increasing strain rate, while for larger strains the predictability vs strain rate relation becomes non-monotonic. We show that for small strains the rate-dependence of deformation predictability can be captured by considering the fraction of dislocations moving against the direction imposed by the external stress, serving as a measure of strain-rate-dependent complexity of the dislocation dynamics. The non-monotonic predictability vs strain rate relation for large strains is argued to be related to a transition from fluctuating to smooth plastic flow when strain rate is increased