Direct simulation and machine learning structure identification unravel soft martensitic transformation and twinning dynamics.

Phase transition between ordered phases has garnered attention from the viewpoint of materials science as well as statistical physics. One interesting example is martensitic transformation and the resulting formation of twin structures, in which atoms or molecules that form one crystalline phase move in a concerted and diffusionless manner toward another crystalline phase. Recently martensitic transformation has been observed experimentally also in various soft materials. However, the complex internal structures involving many molecules have eluded direct investigation of the dynamical processes of martensitic transformation. Here, we carry out a direct simulation of mesoscale structural transition of a liquid crystalline blue phase (BP) of cubic symmetry, known as BP II. The dynamics is simulated by a Langevin-type equation for the orientational order parameter with thermal fluctuations. We demonstrate that machine-learning-aided analysis of local structures successfully unravels the transformation process from a perfect lattice of BP II to a twinned lattice of another BP (BP I). The nucleation of BP I is initiated by the breakup of junctions of line defects (disclinations), followed by the deformation of disclination network. We further show that twinned BP I is reversibly transformed to a perfect lattice of BP II by temperature variation. Order-parameter-based simulations with machine-learning-aided local structure identification provide valuable insights into not only the martensitic transformation of soft materials but also a wider class of complex structural transitions between ordered phases.