Machine-Learning Enabled Study of Surface Reconstructions in Magnetic Topological Insulator MnBi₂Te₄

Using first-principles calculations together with molecular dynamics simulations accelerated by machine-learning on the fly, we focus on the rather unexplored issue of how surface reconstructions in magnetic topological insulator MnBi₂Te₄ (MBT) thin films, which are likely to occur in experimental realizations, can be studied. We demonstrate that an interstitial-2H and peripheral-2H type atomic reconstructions are thermodynamically stable and are responsible for modifying the exchange gap and surface characteristics of MBT thin films, with important implications for the topological indices and the nature of quasi one-dimensional side-wall edge states dominating quantum transport. Surface reconstruction of peripheral-2H type is proposed to be the origin of additional Rashba surface states seen in Angle-Resolved Photoemission Spectroscopy (ARPES) measurements. Our analysis provides a theoretical framework to elucidate the nature and effect of reconstructions in MBT thin films, with predictions for the experimental realization.