Building ferroelectric from the bottom up: The machine learning analysis of the atomic-scale ferroelectric distortions

Abstract

Recent advances in scanning transmission electron microscopy (STEM) have enabled direct visualization of the atomic structure of ferroic materials, enabling the determination of atomic column positions with approximately picometer precision. This, in turn, enabled direct mapping of ferroelectric and ferroelastic order parameter fields via the top-down approach, where the atomic coordinates are directly mapped on the mesoscopic order parameters. Here, we explore the alternative bottom-up approach, where the atomic coordinates derived from the STEM image are used to explore the extant atomic displacement patterns in the material and build the collection of the building blocks for the distorted lattice. This approach is illustrated for the La-doped BiFeO3 system.