Identifying\nRare Events in Quantum Molecular Dynamics\nof Nanomaterials with Outlier Detection Indices

Nanoscale and condensed matter systems evolve on multiple\nlength-\nand time-scales, and rare events such as local phase transformation,\nion segregation, defect migration, interface reconstruction, and grain\nboundary sliding can have a profound influence on material properties.\nWe demonstrate how outlier detection indices can be used to identify\nrare events in machine-learning based, high-dimensional molecular\ndynamics (MD) simulations. Designed to order data-points from typical\nto untypical, the indices enable one to capture atomic events that\nare hard to detect otherwise. We demonstrate the approach with a nanosecond\nMD simulation of a grain boundary in a metal halide perovskite that\nis extensively studied for solar energy and optoelectronic applications.\nThe method captures the initial grain boundary sliding and a spontaneous\nfluctuation half a nanosecond later, both events giving rise to persistent\ndeep electronic trap states that impact charge carrier lifetime and\ntransport and material performance. The approach offers a generalizable\nand simple method for identifying outlier events in complex condensed\nmatter, molecular, and nanoscale systems.