Fig. 1: The DDD+ML approach.
From: Learning dislocation dynamics mobility laws from large-scale MD simulations
a A schematic showing two consecutive dislocation networks \({{\mathcal{G}}}^{t}\) and \({{\mathcal{G}}}^{t+\Delta t}\) extracted using DXA from two MD trajectory snapshots at times t (solid lines) and t + Δt (dashed lines). As returned by DXA, two network configurations are defined by nodal positions r = {ri} and Burgers vectors of their dislocation segments b = {bij}, but are generally non-isomorphic and contain no information about mechanical forces driving dislocation motion. Not having the same nodes present in both networks leaves it uncertain how to define nodal velocities. For instance, three arrows of different colors pointing out of node 2 show three possible ways to define its velocity V2. b A schematic of inference and training loops of our proposed workflow for developing a GNN dislocation mobility function. c An illustration of the Nye’s tensor field-matching approach circumventing the ill-defined problem of matching nodal velocities. As exemplified here, our field-matching procedure is agnostic to details of line discretization and network topology.
