Titre : Temperature dependence of soft phonon modes from first-principles methods and machine learning
Institut Néel, Salle E424
Résumé : Many functional properties in materials, such as ferroelectrity and ferroelasticity, arise from structural phase transitions, in which the material changes its shape and/or atomic structure at a critical temperature. Phenomenologically, this can be explained by a low frequency, or “soft”, mode in the phonon spectrum, which is entropically stabilized above the critical temperature. Naturally, in the vicinity of the phase transition, the phonon spectrum becomes highly temperature-dependent. Density functional theory, as the main theoretical tool to calculate both atomic structures and phonon frequencies, is limited to calculations at absolute zero. To overcome this limit, I present a method combining density functional theory with machine learning to obtain a thermodynamic sampling of the quantum- and thermal displacements of the atoms [1]. This allows us to capture accurately the soft-mode behavior, as we show on the example of the quantum paraelectric perovskite KTaO3, where not only do we have an excellent agreement with experimental data, but we also show that we reproduce the correct low-temperature flattening-off of the dielectric constant, first predicted by Barrett [2].
