Predicting magnetic edge behaviour in graphene using neural networks

TL;DR

This paper introduces a machine learning method that predicts magnetic edge behavior in graphene, enabling rapid simulation of large, disordered systems with high accuracy, thus advancing spintronic research.

Contribution

The study demonstrates that geometric-input neural networks can accurately estimate magnetic profiles in large graphene structures, surpassing previous computational limitations.

Findings

High agreement with mean-field Hubbard calculations

Accurate prediction of electronic, magnetic, and transport properties

Enables rapid simulation of large, disordered graphene systems

Abstract

Magnetic moments near zigzag edges in graphene allow complex nanostructures with customised spin properties to be realised. However, computational costs restrict theoretical investigations to small or perfectly periodic structures. Here we demonstrate that a machine-learning approach, using only geometric input, can accurately estimate magnetic moment profiles, allowing arbitrarily large and disordered systems to be quickly simulated. Excellent agreement is found with mean-field Hubbard calculations, and important electronic, magnetic and transport properties are reproduced using the estimated profiles. This approach allows the magnetic moments of experimental-scale systems to be quickly and accurately predicted, and will speed-up the identification of promising geometries for spintronic applications. While machine-learning approaches to many-body interactions have largely been limited to exact solutions of complex models at very small scales, this work establishes that they can be successfully applied at very large scales at mean-field levels of accuracy.