A general rule for predicting the magnetic moment of Cobalt-based Heusler compounds using compressed sensing and density functional theory

TL;DR

This paper introduces a new descriptor derived from compressed sensing and density functional theory to predict magnetic moments of Co-based Heusler alloys, applicable to both half-metallic and non-half-metallic compounds, improving over the Slater-Pauling rule.

Contribution

A novel machine-learning descriptor for predicting magnetic moments of Co$_2$YZ Heusler alloys, applicable regardless of their half-metallicity, validated by DFT calculations.

Findings

The new descriptor outperforms the Slater-Pauling rule in generality.

It accurately predicts magnetic moments of known and hypothetical Heusler compounds.

DFT confirms the stability of the predicted compounds.

Abstract

We propose a general rule for estimating the magnetic moments of Co$ 2$(cobalt)-based Heusler alloys, especially when doped with late transition metals. We come up with a descriptor that can characterise both pure Co$_2$YZ compounds and the doped ones with the chemical formula Co$_2$Y$_{1-x}$M$_x$Z (M is the dopant) using online data for magnetic moments of Heusler alloys with Co$_2$YZ structure and compressive sensing approach. The newly proposed descriptor not only depends on the number of valence electrons of the compound also it depends on the number of unoccupied d-electrons in the doping site. A comparison of the performance of the proposed descriptor and the Slater-Pauling rule is made. Unlike the Slater-Pauling rule, which is only effective for half-metallic Heusler compounds, our machine-learning approach is more generic since it applies to any Co$_2$YZ Heusler compounds, regardless of whether they are half-metals or not. We use this new rule to estimate the magnetic moments of a few yet-to-be-discovered Heusler compounds and compare the results to density functional theory (DFT) based calculations. Finally, we use DFT and machine learning investigations to prove their stability.