Domain Reconstruction of Twisted Bilayer and Heterobilayer transition metal (di-)chalcogenides via large-scale DFT and Machine Learned Interatomic Potentials

For moiré 2D materials such as twisted bilayers and heterostructures, it is vital to fully characterise the relaxations and corrugations that occur due to interlayer interactions. One can then predict how electronic and optical properties depend on twist angle and the large-scale Moiré pattern [1]. As the relative twist angle between the layers approaches 0º (parallel stacking), or 60º, (antiparallel (AP) stacking), reconstructions occur to maximise the area of low-energy stacking domains, with a lattice of solitons of high-energy stacking connected by domain walls (DWs). We show that Machine Learned Interatomic Potentials (MLIPs) utilising higher-order equivariant message passing, as implemented in MACE [2], can provide the highly-accurate energy dependence on stacking, strain, shear and interlayer distance required to model this behaviour in a way that exactly reproduces vdW-corrected DFT, at much larger scales than ab initio methods can handle. We quantify the domain reconstruction patterns for transition metal (di-)chalcogenides MoS₂, MoSe₂, WS₂, WSe₂ and InSe down to twist angles approaching 1º. We demonstrate effects including DW-bending in AP systems, and the “twirling” that occurs at the solitons in heterobilayers. We also explore the properties of 1T'-WTe₂, explaining the formation of a moiré-induced striped electrostatic potential landscape in the twisted bilayer.

[1] S. Carr et al, Phys. Rev. B 95, 075420 (2017). [2] I. Batatia et al, Adv Neural Inf Process Syst (2022).