Electric vehicles, renewable energy and data centers depend on efficient energy conversion. This is provided by magnetic components - transformers and inductors. But magnetic materials lose part of the energy, heat up and reduce efficiency. The higher the operating frequency, the more important it is to understand where losses come from at the micro level.
In soft magnetic materials, magnetization reversal occurs mainly through the motion of magnetic domain boundaries - regions with different magnetization directions. Conventional models estimate total losses but do not show which structures most strongly resist magnetization reversal.
A group led by Professor Masato Kotsugi of the Tokyo University of Science (TUS, Japan) developed an approach based on explainable AI: it not only gives a prediction but explains it through physical principles. Kosuke Nishioka was a co-author, and the paper was published in NPG Asia Materials.
The researchers conducted large-scale micromagnetic modeling on the Fugaku supercomputer. They studied an idealized iron nanocrystalline soft magnet at frequencies of 100 kHz, 1 MHz and 10 MHz. The shape and connectivity of changing domain boundary networks were described using persistent homology - a mathematical tool that quantitatively characterizes structures in images. Principal component analysis then compressed the features into a two-dimensional representation, linked it to a measure of exchange energy and built an energy landscape with inverse mapping.
The analysis showed that as frequency increases, domains fragment more and more. Near the coercive region, where magnetization reversal occurs, an energy barrier arose that increased with frequency. Inverse mapping showed that the largest contribution comes not from all boundaries uniformly, but from regions around propagating tips of reverse domains and individual thin boundary segments. Similar boundaries can play different roles during magnetization reversal.
The authors emphasize that the barrier found reflects the structural costs of exchange energy, not a direct measurement of heat released. It is not a ready-made loss indicator but an indication of structures for further study.
The work offers a new way to search for microscopic sources of losses. In the future, this may help develop low-loss materials for electric vehicles and energy converters in data centers. Understanding which structures most strongly resist magnetization reversal will help clarify how material composition and processing affect the behavior of domain boundaries.
"Our research may contribute to energy saving and CO2 emission reduction by improving energy conversion efficiency and reducing heating," Kotsugi notes. The combination of supercomputer calculations, topology, magnetism physics and explainable AI links complex magnetic structures to their contribution to energy barriers.