Search

The melting of a solid, like other first-order phase transitions, exhibits an intrinsic time-scale disparity: The time spent by the system in metastable states is orders of magnitude longer than the transition times between the states. Using rare-event sampling techniques, we find that melting of representative solids--here, copper and aluminum--occurs via multiple, competing pathways involving the formation and migration of point defects or dislocations. Each path is characterized by multiple barrier-crossing events arising from multiple metastable states within the solid basin. At temperatures approaching superheating, melting becomes a single barrier-crossing process, and at the limit of superheating, the melting mechanism is driven by a vibrational instability. Our findings reveal the importance of nonlocal behavior, suggesting a revision of the perspective of classical nucleation theory. [ABSTRACT FROM AUTHOR]

A review is presented of the article "Boltzmann generators: Sampling equilibrium states of many-body systems with deep learning" by Frank Noé, Jonas Köhler and others on the use of a machine learning methodology to address a problem with atomistic simulations of complex molecular systems using Boltzmann probability distributions to predict equilibrium properties.