Studies of the ferroelectric phase transition in the age of artificial intelligence

Artificial intelligence techniques, such as machine learning and deep neural network representations make possible studies of the ferroelectric phase transition, beyond mean field and reduced models, with molecular dynamics (MD) simulations of ab-initio quality. In this lecture, I will illustrate this approach with recent studies of two perovskite materials at ambient pressure: lead titanate (PTO), a prototypical ferroelectric crystal, and lead magnesium niobate (PMN), a chemically disordered perovskite that exhibits glassy behavior, called relaxor, and bears similarities with spin and fragile structural glasses. In both cases, simulations predict thermodynamic and dielectric properties in good agreement with experiment and provide fresh insight into the microscopic mechanisms that are behind the observed changes of the dielectric properties with temperature. In PTO thermal anharmonicity and strain mismatch drive the approach to equilibrium upon crossing the critical temperature, originating a nanoscale transformation that avoids macroscopic nucleation. In PMN, anharmonicity associated with disorder is present at the lowest temperatures and manifests itself in quasi-localized vibrational modes that underlie universal glassy features.