Machine learning moiré quantum flatland

Moiré systems are fertile platforms to explore emergent quantum phenomena. Recent experimental discoveries have suggested crucial roles of structural reconstruction in the understandings of the moiré quantum phenomena. However, current modeling and first-principles frameworks face significant challenges in resolving these structural features at the moiré length scale. For example, density functional theory and many-body perturbation theory calculations, while serving as standard approaches for conventional solids, are too expensive in computational cost for most moiré superlattices. In this talk, we employed a neural network method trained by first-principles results to overcome this difficulty. We studied structural reconstructions and lattice dynamics in a variety of moiré structures. Our calculations discovered salient structural features and electronic properties controlled by twist angles, layer constituents, and other tuning knobs. We then connect and compare our findings with available experimental observations. Our work offers new theoretical and computational strategies to study moiré materials.