Halide perovskites have captivated scientific interest due to their exceptional optoelectronic properties and potential in next-generation energy applications. However, their propensity for instability poses significant challenges for practical use. Defects dynamics play important role in the stability however is very challenging to study due to the complex chemical environment in halide perovskites. Addressing these challenges, Mike and Victor trained force fields by using on-the-fly trained machine-learned (MLFFs) methods. This technique enables iterative refinement of force fields through Molecular Dynamics (MD) simulations, continuously informed by density functional theory (DFT) calculations, thus minimizing the need for human input. The synergy of computational efficiency with near-first-principles accuracy allows us to probe the intricate dynamics of phase transitions and defects in perovskites with unprecedented detail.

A schematic workflow of the generation of MLFF and its application for phase transition of halide perovskites.

The work is published on Chemical Communications and more details is documented in Victor’s master thesis, which was supervised by Mike. Congratulations on this great joint effort to both!

Similar integration of machine learning and quantum mechanical calculations presents immense opportunities to study the intricate interplay of electrons, ions, and lattices in pivotal materials, including halide perovskites for energy conversion and storage. Our group continues on expanding this work to study more complex processes, and other energy materials. Stay tuned for future updates.