Machine Learning Study of the Magnetic Ordering in 2D Materials.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Magnetic materials are crucial for technologies like data storage and quantum devices.
• Two-dimensional (2D) materials offer novel platforms for magnetism, despite theoretical challenges.
• Predicting magnetic ordering in these materials is essential for their technological application.

Purpose of the Study:

• To develop a machine-learning strategy for predicting and understanding magnetic ordering in 2D materials.
• To identify the key atomic and structural features governing magnetism in 2D compounds.
• To discover new 2D magnetic materials for future applications.

Main Methods:

• A machine-learning approach combining random forest for magnetism prediction and Shapley additive explanations (SHAP) for feature importance.
• Development of material maps using the sure independence screening and sparsifying method (SIS) to predict magnetic ordering (ferromagnetic/antiferromagnetic).
• Analysis of atomic features, including 3d transition metals, halides, and spin-orbit coupling (SOC).

Main Results:

• The random forest model achieved 86% accuracy in predicting the existence of magnetism in 2D materials.
• Material maps accurately predicted magnetic ordering with approximately 90% accuracy.
• Identified 3d transition metals, halides, and specific structural arrangements as key contributors to magnetism.
• Spin-orbit coupling (SOC) was found to be a critical feature distinguishing ferromagnetic from antiferromagnetic order.

Conclusions:

• The proposed machine-learning strategy effectively predicts and explains magnetism in 2D materials.
• The study reveals fundamental trends in the chemical and structural space of 2D magnetic compounds.
• This work paves the way for the experimental exploration and design of novel 2D magnetic materials.