Discovering two-dimensional magnetic topological insulators by machine learning

TL;DR

This paper develops a machine learning approach to identify two-dimensional magnetic topological insulators, discovering new classes of materials with potential for experimental realization, by combining neural networks with ab initio calculations.

Contribution

It introduces an interpretable neural network-based chemical rule for diagnosing topological properties, enabling high-throughput discovery of new magnetic topological insulators.

Findings

Discovered 15 new magnetic topological insulators, including 7 with full band gaps.

Developed a high-accuracy, interpretable model using chemical formulas and Hubbard U.

Validated the model's effectiveness across different material regimes.

Abstract

Topological materials with unconventional electronic properties have been investigated intensively for both fundamental and practical interests. Thousands of topological materials have been identified by symmetry-based analysis and ab initio calculations. However, the predicted magnetic topological insulators with genuine full band gaps are rare. Here we employ this database and supervisedly train neural networks to develop a heuristic chemical rule for electronic topology diagnosis. The learned rule is interpretable and diagnoses with a high accuracy whether a material is topological using only its chemical formula and Hubbard $U$ parameter. We next evaluate the model performance in several different regimes of materials. Finally, we integrate machine-learned rule with ab initio calculations to high-throughput screen for magnetic topological insulators in 2D material database. We discover 6 new classes (15 materials) of Chern insulators, among which 4 classes (7 materials) have full band gaps and may motivate for experimental observation. We anticipate the machine-learned rule here can be used as a guiding principle for inverse design and discovery of new topological materials.