Capturing 2D van der Waals magnets with high probability for experimental demonstration from materials science literature

Abstract

Abstract 2D van der Waals (vdW) magnets have opened intriguing prospects for next‐generation spintronic nanodevices. Machine learning techniques and density functional theory calculations enable the discovery of 2D vdW magnets to be accelerated; however, current computational frameworks based on these state‐of‐the‐art approaches cannot offer probability analysis on whether a 2D vdW magnet can be experimentally demonstrated. Herein, a new framework can be established to overcome this challenge. Via the framework, 2D vdW magnets with high probability for experimental demonstration are captured from materials science literature. The key to the successful establishment is the introduction of the theory of mutual information. Historical validation of predictions substantiates the high reliability of the framework. For example, half of the 30 2D vdW magnets discovered in the literature published prior to 2017 have been experimentally demonstrated in the subsequent years. This framework has the potential to become a revolutionary force for progressing experimental discovery of 2D vdW magnets.

Keywords

• 2D vdW magnets
• mutual information
• neural networks