ai-identifies-magnetic-structures-driving-high-frequency-energy-loss
AI Identifies Magnetic Structures Driving High-Frequency Energy Loss

AI Identifies Magnetic Structures Driving High-Frequency Energy Loss

Researchers at Tokyo University of Science have developed a physics-interpretable artificial intelligence approach to identify the microscopic magnetic structures responsible for energy loss in soft magnetic materials. The study, published in NPG Asia Materials on November 17, 2026, addresses a critical challenge in modern power electronics, where magnetic components such as transformers and inductors lose energy as heat during operation. As electrical systems in electric vehicles, renewable energy systems, and data centers operate at increasingly high frequencies, understanding the specific origins of these losses is becoming crucial for improving overall energy efficiency and reducing heat generation.

Soft magnetic materials rely on the movement of magnetic domain walls to reverse their magnetization. These domain walls separate regions that are magnetized in different directions. While conventional models can estimate overall magnetic losses, they often fail to identify which specific microscopic structures contribute most to the resistance during this reversal process.

To address these challenges, a research team led by Professor Masato Kotsugi from the Department of Material Science and Technology at Tokyo University of Science developed a novel AI framework. The team’s goal was to create reliable AI methods that explain results through physical principles rather than simply producing predictions. This approach, described as explainable AI, allows researchers to see the invisible by linking complex data to underlying physical mechanisms, providing a clearer understanding of how magnetic materials behave under high-frequency conditions.

The investigation involved large-scale micromagnetic simulations performed using the Fugaku supercomputer. The researchers calculated the behavior of an idealized iron-based nanocrystalline soft magnet at frequencies of 100 kHz, 1 MHz, and 10 MHz. To characterize the evolving magnetic domain-wall networks, they applied persistent homology, a mathematical technique that quantifies shapes and connectivity in images. This method allowed them to capture the topological features of the magnetic structures as they changed during the simulation.

Principal component analysis was then used to compress the resulting image-based descriptors into a two-dimensional representation. The team linked this representation to an image-derived exchange-energy measure and constructed an energy landscape with inverse mapping. This process enabled them to identify the specific structures that contributed most strongly to the energy barrier during magnetization reversal. The integration of supercomputing, topology, magnetic physics, and explainable AI connected complex magnetic structures with their contributions to energy barriers in a way that previous methods could not achieve.

The analysis revealed that magnetic domains became increasingly subdivided as the excitation frequency increased. The researchers identified an energy barrier near the coercive region, where magnetization reverses, that increased with frequency. Most importantly, the inverse mapping showed that high-contribution features were concentrated around propagating reversed-domain tips and selected fine wall segments. These features were not distributed uniformly across the network, suggesting that similar walls can play different roles during the reversal process depending on their specific location and context.

It is important to note that the identified barrier represents an exchange-derived structural energy cost, not a direct measurement of dissipated heat. This distinction is crucial for interpreting the results, as the study focuses on the structural origins of resistance rather than the thermal output directly. By identifying the specific domain-wall segments that contribute most to this energy cost, the study provides a new way to investigate the microscopic structures associated with energy loss in magnetic materials.

The findings could eventually guide the development of low-loss magnetic materials for electric vehicles and power-conversion equipment used in data centers. Understanding the structures linked to greater resistance could help researchers investigate how material composition and processing influence domain-wall behavior. This knowledge is essential for designing materials that can maintain high efficiency even as operating frequencies continue to rise in modern technological applications.

The investigation represents a collaborative effort within the Tokyo University of Science, coauthored by Kosuke Nishioka, who completed a master’s course in the Department of Material Science and Technology in 2025. Dr. Kotsugi, who earned his doctorate from Osaka University in 2001 and joined the Tokyo University of Science faculty in 2015, centers his broader academic program on magnetism, nanostructures, synchrotron radiation, and materials informatics. His ongoing research integrates magnetic physics and explainable artificial intelligence to clarify complex material behaviors, providing systematic pathways to accelerate the discovery and structural design of high-performance functional materials for energy-efficient technologies.

The institutional background supporting this computational breakthrough is rooted in the Tokyo University of Science, Japan’s largest science-specialized private research university. Founded in 1881, the university operates across four campuses located throughout central Tokyo, its surrounding suburbs, and Hokkaido. Operating under an institutional mission dedicated to creating science and technology for the harmonious development of nature, human beings, and society, the university emphasizes merit-based advancement and comprehensive research spanning basic inquiries to applied engineering. Notably, the institution holds the distinction of being the only private university in Japan and the sole private university in Asia to produce Nobel laureates within the natural sciences field.

The research initiative also engaged broader national scientific support across Japan through several targeted funding programs. The project received direct financial backing from the Ministry of Education, Culture, Sports, Science and Technology, known as MEXT, through the Program for Creation of Innovative Core Technology for Power Electronics under grant JPJ009777. Additional financial resources were provided by the Japan Society for the Promotion of Science through KAKENHI grant numbers 21H04656, 22A201, and 22H05107, alongside support from the Japan Science and Technology Agency through its CREST initiative under grant JPMJCR21O1. Furthermore, contributor Chiharu Mitsumata received academic backing from the Tsukuba Research Center for Energy Materials Science at the University of Tsukuba, illustrating the collaborative network supporting the computational investigation.

Formally titled “Physics-interpretable topological machine learning identifies domain-wall contributors to high-frequency loss in a nanocrystalline soft magnet,” the peer-reviewed publication confirms that no financial or nonfinancial conflicts of interest influenced the research or its authors. By demonstrating that specific fine wall segments and propagating tips drive structural energy barriers during magnetization reversal, the study establishes an evidence-based roadmap for material scientists seeking to refine composition and processing techniques for next-generation power electronics.

Subject of Research: Physics-interpretable AI analysis of magnetic domain-wall contributions to high-frequency energy loss

Article Title: How AI can identify the magnetic structures behind high-frequency energy loss

Article References: How AI can identify the magnetic structures behind high-frequency energy loss. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: explainable AI, magnetic domain walls, high-frequency energy loss, soft magnetic materials, micromagnetic simulations, persistent homology, power electronics, energy efficiency, nanocrystalline soft magnet, Fugaku supercomputer, magnetization reversal, exchange energy, topological machine learning, Tokyo University of Science, NPG Asia Materials, power conversion, heat generation, materials informatics, domain-wall networks, coercive region