Machine learning assisted High-Throughput study of M$_4$X$_3$T$_x$ MXenes

TL;DR

This study uses machine learning and high-throughput DFT calculations to analyze the stability, electronic, and magnetic properties of a large set of MXenes, revealing diverse magnetic behaviors across different compositions.

Contribution

It introduces a machine learning framework that accelerates high-throughput DFT studies of MXenes, enabling systematic classification of their magnetic and electronic properties.

Findings

Ti-, Zr-, Hf-, Nb-, Ta-based MXenes are non-magnetic metals.

Sc- and Y-based MXenes show weak ferromagnetism or semiconducting behavior.

Cr- and Mn-based MXenes are ferromagnetic with high spin polarization.

Abstract

In this work, we employ a machine-learning-assisted high-throughput density functional theory framework to systematically investigate the stability, electronic structure, and magnetic ground states of 234 M$_4$X$_3$T$_x$ MXenes. The machine learning model predicts lattice parameters with up to 94% accuracy using a relatively small training dataset and significantly reduces structural optimization time in high-throughput calculations. Based on total energy and density-of-states analyses, we classify the magnetic nature of MXenes across different transition- metal compositions and surface terminations. Ti-, Zr-, Hf-, Nb-, and Ta-based MXenes are found to be non-magnetic metals for all functional groups considered, while Sc- and Y-based systems exhibit a range of behaviors including weak ferromagnetism and semiconducting character. V- and Fe-based MXenes are identified as antiferromagnetic metals, whereas Cr- and Mn-based MXenes yield 16 ferromagnetic systems with spin polarization ranging from 50% to 100%.