説明可能AIのアプローチでフェルミ面の異常検知に成功 ～ホイスラー合金のスピン偏極とノーダルラインを自動検出～

2026-04-28 東京理科大学,京都工芸繊維大学,筑波大学,科学技術振興機構

図1 PCAによるフェルミ面画像の次元削減結果。I～VIIのジャンプはフェルミ面における非系統的変化を示しており、スピン偏極率の極値と変曲点に対応していた。最も大きなジャンプ（VII）データを再構成した結果、ノーダルラインの出現位置を可視化できた。

＜関連情報＞

• https://www.tus.ac.jp/today/archive/20260417_0478.html
• https://www.nature.com/articles/s41598-026-39115-0

解釈可能な機械学習によるCo2MnGaxGe1-xにおけるフェルミ面形態の異常検出 Anomaly detection of fermi surface morphology in Co2MnGaxGe1-x via interpretable machine learning

Daichi Ishikawa,Kentaro Fuku,Yoshio Miura,Yasuhiko Igarashi,Yuma Iwasaki,Yuya Sakuraba,Koichiro Yaji,Alexandre Lira Foggiatto,Takahiro Yamazaki,Naoka Nagamura & Masato Kotsugi
Scientific Reports  Published:27 April 2026
DOI:https://doi.org/10.1038/s41598-026-39115-0

Abstract

The Fermi surface provides indispensable insights into the electronic structure of materials. Here, we present a robust analysis framework employing an interpretable machine learning approach to investigate Fermi surfaces. Complex Fermi surface images of the Heusler alloy Co₂MnGa_xGe_1−x, along with corresponding spin polarization, were analyzed using simple principal component analysis (PCA). Our results reveal that pronounced “jumps” in the PCA space correlate strongly with extrema and inflection points in the spin polarization. Notably, compositions near Ga = 0.94–0.95 exhibit significant changes attributable to the emergence of nodal lines. And the position of nodal lines in momentum space were automatically detected by differential analysis of outlier. Robustness evaluations demonstrate that our method remains effective even under conditions of increased image broadening and noise, mimicking ARPES experimental data. This method can contribute to the analysis of large-scale datasets by detecting non-systematic outliers.