An Atomic Cluster Expansion Potential for Twisted Multilayer Graphene

Abstract: Twisted multilayer graphene, characterized by its moire patterns arising from inter-layer rotational misalignment, serves as a rich platform for exploring quantum phenomena. While first-principles calculations are computationally prohibitive and empirical interatomic potentials often lack accuracy, machine-learning interatomic potentials (MLIPs) present a promising alternative, offering (near-)DFT accuracy at a significantly reduced computational cost. Despite their success in two-dimensional monolayer materials, MLIPs remain under-explored in twisted multilayer graphene systems. In this work, we develop an Atomic Cluster Expansion (ACE) potential for simulating twisted multilayer graphene and test it on a range of simulation tasks. We propose an approach to construct training and test datasets that incorporate all possible twist angles and local stacking, including incommensurate ones. To achieve this, we generate configurations with periodic boundary conditions suitable for DFT calculations, and then introduce an internal twist and shift within those supercell structures. We further refine the dataset through active learning filtering, guided by Bayesian uncertainty quantification. Our model is validated for accuracy and robustness through a wide range of numerical tests.