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

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