"New Theoretical Insights into Moiré Solids and 2D Magnets from Machine Learning Assisted First-
Principles Calculations"
This talk will show our recent theoretical and computational investigations into moiré superlattices and 2D magnets. We start by demonstrating that a deep neural network guided by first-principles data can be used to examine moiré structural reconstruction in various homobilayers and heterobilayers of transition metal dichalcogenides. Going beyond the capacity of direct DFT calculations, our machine-learning enabled workflow discovers salient structural features and key topological characters controlled by twist angles, layer composition, and other tuning knobs. This knowledge can be
used to inform accurate continuum model, and to predict new forms of moiré potential and moiré topology. In the second part, we show that the magnetism in 2D magnets like CrSBr is highly tunable by strain, pressure, chemical doping, and interfacing with other vdW materials like graphene. This allows for multi-dimensional control of magnetic orders and noncollinear magnetism. We demonstrate the strong light-matter interactions and unique excitonic structures in CrSBr provide possibility to control material transparency by ultrafast means. Finally, we connect our theoretical discoveries to
experimental results and explore potential applications.
