Modeling dynamics and phase transitions in perovskites via machine learned potentials
Erik Fransson, Chalmers University, Sweden
Perovskites constitute a prominent class of materials with a wide range of properties and applications. Their performance is often connected to particular soft phonon modes that are most apparent in driving a series of phase transitions. These pheonmena are challenging to investigate using first-principles methods, as this requires large systems and long time scales. Here, to overcome this we employ large-scale molecular dynamics simulations using machine-learned potentials. We explore differences and similarities between oxide, halide, organic-inorganic, and two-dimensional (2D) perovskites, with a focus on predicting and interpreting experimental results.
First, in inorganic halide perovskites we find that the soft phonon modes driving the phase transitions exhibit strong over-damped behaviour. This overdamping is driven by strong anharmonicity, and shows up as large excess intensity in inelastic neutron scattering spectra.
Next, we explore 2D hybrid halide perovskites which are composed of a small number of perovskite layers stacked on top of each other and separated by organic cations that act as spacers. We demonstrate that the thickness of perovskite layers and the choice of organic linker molecule strongly impact on the dynamics and phase transitions in these materials. We show that this is fundamentally linked to the surface layers exhibiting separate transitions compared to the interior bulk layers.
Lastly, we analyze the oxide perovskite barium zirconate, one of the few perovskites reported to retain a cubic structure down to zero Kelvin. The local and global structure have however been heavily debated in the literature. Here, we predict the phase diagram, Raman spectra and diffuse scattering spectra in order to understand experimental observations and provide a coherent picture for both the local and global structure in barium zirconate.