Neural Network Dynamics for Barium Titanate (BaTiO3) Moire Structures

Experiments involving stacking freestanding perovskite layers of BaTiO3 (BTO) of thickness of about 10nm to 15nm at different twist angles, 3°, 6°, 10.4°, and 50° show the emergence of polarization vortices and antivortices. The couplings across the interface between the twisted layers drive large strain gradients in the ferroelectric layers, resulting in vortex-like modulations of the homogeneous polarization state due to the flexoelectric effect. It is noted that the periodicity of the 2D vortex pattern can be tuned mainly by controlling the twisting angle. Current empirical forcefields are unable to model the interaction at such interfaces, making it difficult to investigate this phenomenon using molecular dynamics simulations. In this project, we have developed a large training set of DFT trajectories for various twist angles that are used to train Legato-Nequip model to run Neural Network Molecular Dynamics simulations to investigate structure correlations in BTO moire structures.