Success in Automatic Anomaly Detection of Fermi Surfaces via Explainable AI: Automatic Detection of Spin Polarization and Nodal Lines in Heusler Alloys
A research group including Tokyo University of Science developed a new Explainable AI (XAI) method to automatically analyze the Fermi surface of Heusler alloys. By combining PCA and outlier detection, they successfully identified spin polarization peaks and the appearance of nodal lines.
📋 Article Processing Timeline
- 📰 Published: April 28, 2026 at 19:00
- 🔍 Collected: April 28, 2026 at 10:31
- 🤖 AI Analyzed: April 28, 2026 at 15:01 (4h 29m after Collected)
【Research Abstract and Points】
- Developed an analysis method for the Fermi surface of Heusler alloys using an Explainable AI approach.
- Focused on 'jumps' in Principal Component Analysis (PCA), revealing they correspond to extreme values and inflection points of the spin polarization rate.
- Successfully detected the appearance positions of nodal lines by reconstructing outlier data.
- Robust against low-quality data, expected to expand into various material systems as an experimental Fermi surface topology analysis method.
【Research Overview】
A joint research group consisting of Daichi Ishikawa (2nd-year Master's student), Dr. Kentaro Fuku (then Postdoctoral Fellow), and Professor Masato Kotsugi of the Department of Materials Science and Technology, Faculty of Advanced Engineering, Tokyo University of Science; Professor Yoshio Miura of Kyoto Institute of Technology; Associate Professor Yasuhiko Igarashi of the System Information Department, University of Tsukuba; and the National Institute for Materials Science (NIMS), has successfully established a machine learning method for automatically analyzing the Fermi surface of the Heusler alloy Co2MnGaxGe1-x (Cobalt-Manganese-Gallium-Germanium).
The Fermi surface plays a vital role in understanding the electrical, magnetic, and topological properties of materials. Because the shape of the Fermi surface changes complexly depending on these functions, analyzing fine shape changes is difficult and manual analysis involves significant labor. In this research, using an Explainable AI approach, the team established a method to automatically detect shape changes in the Fermi surface by combining Principal Component Analysis (PCA) and distance learning-based outlier detection. This method enabled the automatic visualization of the extreme values of spin polarization rates and the appearance of nodal lines. Furthermore, they verified robustness against noise and blurring, building a foundation for experimental AI analysis methods. This method contributes to the realization of AI4Science as a basis for intelligent analysis of functional materials.
This result was published online in the international academic journal 'Scientific Reports' on April 27, 2026.
【Research Background】
The Fermi surface is a crucial source of information that determines the physical properties of materials. Its shape changes complexly based on crystal structure, elemental composition, and band dispersion, giving rise to various material functions such as carrier density, magnetic behavior, and spin polarization.
Angle-resolved photoemission spectroscopy (ARPES) has been widely used for experimental Fermi surface analysis. Recent technological advances have significantly improved angular and energy resolution. Additionally, with the advent of next-generation radiation sources, high-throughput measurement equipment is being developed.
However, analyzing the Fermi surface remains a cumbersome process requiring high-level expertise. Quantifying shapes often depends on researchers' subjective judgments, and the arbitrariness of analysis has been a challenge. Moreover, while various machine learning methods have been proposed recently, many remain black-box approaches or are limited to simple physical property predictions. Therefore, 'Explainable AI' that analyzes mechanisms with high interpretability is required.
In this study, the team applied PCA and distance-based outlier detection to the spintronic material Co2MnGaxGe1-x. They developed a method to quantify shape changes in the Fermi surface and visualize the changes in function and their origins.
【Detailed Research Results】
Fermi surface data for Co2MnGaxGe1-x were created via first-principles calculations, preparing a dataset that well reproduces prior ARPES research. Next, to quantitatively analyze the complex shape changes, PCA analysis and distance-based anomaly detection were performed. As a result, systematic changes in the Fermi surface depending on composition could be visualized, and characteristic jumps from I to VI shown in Fig. 1 were confirmed. These jumps matched the extreme values or inflection points of the spin polarization curve, showing that distance in the PCA space is an effective index for describing changes in spin polarization.
Significant outliers were observed between Ga compositions x = 0.94 and 0.95 (Jump VII). Reconstructing the data for these outliers revealed they correspond to the appearance positions of nodal lines. In Co2MnGa, it is known that nodal lines exist at energies slightly higher than the Fermi energy, meaning the team successfully detected information that is the origin of the anomalous Hall effect and anomalous Nernst effect.
- Developed an analysis method for the Fermi surface of Heusler alloys using an Explainable AI approach.
- Focused on 'jumps' in Principal Component Analysis (PCA), revealing they correspond to extreme values and inflection points of the spin polarization rate.
- Successfully detected the appearance positions of nodal lines by reconstructing outlier data.
- Robust against low-quality data, expected to expand into various material systems as an experimental Fermi surface topology analysis method.
【Research Overview】
A joint research group consisting of Daichi Ishikawa (2nd-year Master's student), Dr. Kentaro Fuku (then Postdoctoral Fellow), and Professor Masato Kotsugi of the Department of Materials Science and Technology, Faculty of Advanced Engineering, Tokyo University of Science; Professor Yoshio Miura of Kyoto Institute of Technology; Associate Professor Yasuhiko Igarashi of the System Information Department, University of Tsukuba; and the National Institute for Materials Science (NIMS), has successfully established a machine learning method for automatically analyzing the Fermi surface of the Heusler alloy Co2MnGaxGe1-x (Cobalt-Manganese-Gallium-Germanium).
The Fermi surface plays a vital role in understanding the electrical, magnetic, and topological properties of materials. Because the shape of the Fermi surface changes complexly depending on these functions, analyzing fine shape changes is difficult and manual analysis involves significant labor. In this research, using an Explainable AI approach, the team established a method to automatically detect shape changes in the Fermi surface by combining Principal Component Analysis (PCA) and distance learning-based outlier detection. This method enabled the automatic visualization of the extreme values of spin polarization rates and the appearance of nodal lines. Furthermore, they verified robustness against noise and blurring, building a foundation for experimental AI analysis methods. This method contributes to the realization of AI4Science as a basis for intelligent analysis of functional materials.
This result was published online in the international academic journal 'Scientific Reports' on April 27, 2026.
【Research Background】
The Fermi surface is a crucial source of information that determines the physical properties of materials. Its shape changes complexly based on crystal structure, elemental composition, and band dispersion, giving rise to various material functions such as carrier density, magnetic behavior, and spin polarization.
Angle-resolved photoemission spectroscopy (ARPES) has been widely used for experimental Fermi surface analysis. Recent technological advances have significantly improved angular and energy resolution. Additionally, with the advent of next-generation radiation sources, high-throughput measurement equipment is being developed.
However, analyzing the Fermi surface remains a cumbersome process requiring high-level expertise. Quantifying shapes often depends on researchers' subjective judgments, and the arbitrariness of analysis has been a challenge. Moreover, while various machine learning methods have been proposed recently, many remain black-box approaches or are limited to simple physical property predictions. Therefore, 'Explainable AI' that analyzes mechanisms with high interpretability is required.
In this study, the team applied PCA and distance-based outlier detection to the spintronic material Co2MnGaxGe1-x. They developed a method to quantify shape changes in the Fermi surface and visualize the changes in function and their origins.
【Detailed Research Results】
Fermi surface data for Co2MnGaxGe1-x were created via first-principles calculations, preparing a dataset that well reproduces prior ARPES research. Next, to quantitatively analyze the complex shape changes, PCA analysis and distance-based anomaly detection were performed. As a result, systematic changes in the Fermi surface depending on composition could be visualized, and characteristic jumps from I to VI shown in Fig. 1 were confirmed. These jumps matched the extreme values or inflection points of the spin polarization curve, showing that distance in the PCA space is an effective index for describing changes in spin polarization.
Significant outliers were observed between Ga compositions x = 0.94 and 0.95 (Jump VII). Reconstructing the data for these outliers revealed they correspond to the appearance positions of nodal lines. In Co2MnGa, it is known that nodal lines exist at energies slightly higher than the Fermi energy, meaning the team successfully detected information that is the origin of the anomalous Hall effect and anomalous Nernst effect.