Microscopy and Microanalysis | 2019
FerroNet: Machine Learning Flow for Analysis of Ferroelectric and Ferroelastic Materials
Abstract
Ferroelectric and multiferroic materials are among the most fascinating materials classes, due to both broad spectrum of applications and dazzling array of physical phenomena they exhibit. For over 70 years, ferroelectric structure and the nature of the order parameter were explored using the macroscopic scattering methods, providing the information on the average structure and symmetries, as well as correlate disorder. In the last decade, multiple studies of polarization behavior in ferroelectrics were reported using the direct imaging of atomic coordinates via the (Scanning) Transmission Electron Microscopy ((S)TEM), where atomic coordinates were used to map polarization field. However, implied in the analyses to date were the existence of macroscopically-defined polarization field as a (single) order parameter in the system and the relationship between the mesoscopic polarization and local atomic coordinates was postulated based on macroscopic models. In more complex analyses, the bulk form of the Ginzburg Landau free energy was additionally adopted as determined from macroscopic thermodynamic and scattering studies. However, in many materials systems such as morphotropic and relaxor ferroelectrics, the nature of the order parameter itself and hence corresponding free energy expansions are actively debated. Correspondingly, of interest is the question whether these descriptors can be obtained from the experimental data, as opposed to being postulated.