Structure-Driven Prediction of Magnetic Order in Uranium Compounds

TL;DR

This paper introduces a fast machine learning method using structural data to predict magnetic states in uranium compounds, overcoming the limitations of traditional DFT calculations for correlated materials.

Contribution

It presents a novel, computationally inexpensive random forest classifier that accurately predicts magnetic ground states solely from structural features.

Findings

Achieved 60.2% accuracy in predicting magnetic states.

Outperformed random chance significantly.

Provides a faster alternative to DFT for material discovery.

Abstract

The advancement of machine learning technologies has revolutionized the search and optimization of material properties. These algorithms often rely on theoretical calculations, such as density functional theory (DFT), for data inputs and validation, which are not always effective for uranium-based materials due to their strong electron correlations. This study presents a computationally inexpensive machine learning approach, specifically a random forest classifier, to predict the magnetic ground states of uranium compounds using only structural inputs. Our model, trained on a curated dataset of experimentally-verified magnetic orders, achieves a mean accuracy of 60.2%, significantly outperforming random chance. By excluding computationally intensive DFT calculations, our method offers a faster and reliable alternative for discovering new materials with desirable magnetic properties, addressing the challenges posed by strong correlations in quantum materials.