Predicting moiré physics from local proximity effects in van der Waals heterostructures

Abstract While proximity-induced magnetism in van der Waals (vdW) heterostructures is often treated as uniform or analytically predictable, we reveal significant spatial fluctuations that challenge conventional modeling. Through a case study of graphene on Cr 2 Ge 2 Te 6 (CGT), we reveal a profound decoupling between pseudospin-governed polarization and the highly localized nature of induced atomic moments. We resolve this dependency using a machine learning framework trained on density functional theory (DFT) data. By employing atomic environment descriptors, we demonstrate that proximity effects are dictated by the local stacking configuration within ~2 nm 2 , enabling the discovery of complex moiré patterns and dodecagonal states—features that remain computationally prohibitive for standard DFT. This study provides a scalable method for the precise engineering of proximity-driven phenomena in next-generation quantum materials.