Energy simply vanishes, converted to heat, wasted—until now.
Within the microscopic architecture of every electric motor, energy has long vanished into heat through mechanisms too intricate for conventional science to fully explain. A team of Japanese researchers has now built an AI model grounded in the language of physics itself — one capable of mapping the hidden magnetic labyrinths inside motor materials and naming, for the first time, the precise forces responsible for that invisible loss. Their work, emerging from Tokyo University of Science and three partner institutions, suggests that the long-standing boundary between what engineers could observe and what they could understand may finally be dissolving.
- Electric motors silently hemorrhage energy as heat through microscopic magnetic structures so complex they have resisted decades of scientific explanation.
- The maze-like domain patterns inside motor materials shift unpredictably with temperature, creating a tangle of interacting forces that conventional tools could measure but never fully decode.
- Researchers combined persistent homology — a mathematics of hidden shape — with machine learning to construct a 'digital free-energy landscape' that makes the invisible terrain of magnetic loss visible for the first time.
- The model identified four distinct energy barriers governing magnetization reversal, and pinpointed the specific roles of entropy, exchange interactions, and demagnetizing effects in driving energy waste.
- Because the framework is built on universal thermodynamic principles rather than narrow pattern-matching, it carries the potential to optimize magnetic systems far beyond electric motors.
Inside every electric motor lives a problem engineers have long struggled to see: magnetic energy quietly converts to heat and disappears. The culprit hides in the microscopic structure of the motor's magnetic materials — in regions called domains, where atoms align their fields in the same direction. In certain soft magnetic materials, these domains form patterns so tangled they resemble a maze, with walls that branch and zigzag in ways that seem almost random. Scientists could observe these structures and measure the energy they wasted, but they could not explain the connection between the two. The problem was simply too complex.
Professor Masato Kotsugi and Dr. Ken Masuzawa at Tokyo University of Science, alongside colleagues from three other Japanese institutions, set out to bridge that gap. Their tool — the eX-GL model — fuses persistent homology, a mathematical technique for detecting hidden structural patterns, with machine learning that identifies which patterns carry the most physical meaning. The team photographed magnetic domains in a rare-earth iron garnet material across a range of temperatures and fed those images into the model.
What emerged was a kind of topographic map of invisible terrain: a digital free-energy landscape showing how the magnetic microstructure evolves as energy shifts. By linking this landscape to the moment when a magnetic field reverses direction, the researchers identified four major energy barriers governing the process. They could see, for the first time, not merely that energy was lost, but exactly which physical mechanisms — entropy, exchange interactions, demagnetizing effects — were responsible, and how their tug-of-war drives the maze to grow more complex and wasteful.
The implications reach beyond motors. Because free energy is a universal thermodynamic measure, the eX-GL framework — built on physics rather than pure pattern recognition — can potentially be adapted to other complex material systems. For electric vehicles, the stakes are immediate: every fraction of efficiency recovered in a motor means longer range or lighter batteries. Published in Scientific Reports and supported by Japanese government grants, this work suggests that pairing physics-based reasoning with artificial intelligence may be the key to unlocking efficiency gains that have remained hidden for decades.
Inside every electric motor humming in a car or appliance lives a problem that engineers have long struggled to see clearly: magnetic energy simply vanishes, converted to heat, wasted. The culprit hides in the microscopic architecture of the motor's magnetic materials—in structures so small and so intricately tangled that scientists have lacked the tools to understand exactly how and why the energy escapes. Now a team of researchers in Japan has built something new: an artificial intelligence model trained in the language of physics itself, capable of peering into these hidden magnetic labyrinths and revealing the mechanisms that steal efficiency from electric motors.
The heart of the problem lies in magnetic domains—tiny regions within materials where atoms align their magnetic fields in the same direction. How these domains arrange themselves, how they shift and reorganize in response to temperature changes, determines how much energy a motor wastes. In some soft magnetic materials used in motors, these domains form patterns so complex they resemble a maze, with walls that zigzag and branch in ways that seem almost random. Scientists call them maze domains. For years, researchers could observe these structures under a microscope and measure the energy loss they caused, but they could not fully explain the connection between what they saw and what they measured. The problem was too tangled, too many factors interacting at once.
Professor Masato Kotsugi and Dr. Ken Masuzawa at Tokyo University of Science, working with colleagues from three other Japanese universities, set out to bridge that gap. They developed what they call the entropy-feature-eXtended Ginzburg-Landau model—the eX-GL model for short. The approach combines two powerful tools: persistent homology, a mathematical technique that finds hidden structural patterns in data, and machine learning, which identifies which patterns matter most. The researchers photographed magnetic domains in a rare-earth iron garnet material at different temperatures, then fed those images into their model.
What emerged was a kind of map. The persistent homology step detected the intricate topological features buried in the domain images—the actual shape and arrangement of the magnetic maze. Machine learning then distilled this information down to its essential components, producing what the researchers call a digital free-energy landscape. This landscape shows how the magnetic microstructure evolves as energy changes, like a topographic map of an invisible terrain. By connecting this landscape to the actual magnetization reversal process—the moment when the magnetic field flips—the team identified four major energy barriers that control how easily the domains can reverse.
The analysis revealed something previously hidden: as domain walls grow longer and more convoluted, the maze becomes more complex, driven by a tug-of-war between entropy and exchange forces. These are fundamental physical properties, and understanding their interplay explained why certain magnetic structures waste more energy than others. The researchers measured the specific contributions of exchange interactions, demagnetizing effects, and entropy to the overall energy loss. For the first time, they could see not just that energy was lost, but exactly which physical mechanisms were responsible.
What makes this work significant is not just what it reveals about motors. Kotsugi emphasizes that free energy is a universal thermodynamic measure—the same principle that governs magnetic domains also governs other complex physical systems. The eX-GL model, because it is built on physics rather than pure pattern-matching, can potentially be adapted to study energy landscapes in other materials and technologies. The framework automates what once required painstaking manual interpretation, making it possible to tackle complexity that conventional simulation and experiment alone cannot fully explain.
For electric vehicles, the implications are practical. Every percentage point of efficiency recovered in a motor translates to longer range or smaller batteries. The hidden energy losses in magnetic domains represent one of the largest untapped opportunities for improvement. By revealing the mechanisms behind these losses, this work opens a path toward designing better motor materials and operating strategies. The research, published in Scientific Reports and supported by Japanese government grants, suggests that the marriage of physics-based reasoning with artificial intelligence may be the key to unlocking efficiency gains that have remained invisible for decades.
Citas Notables
Conventional simulations oversimplify real materials, while experiments reveal complexity without a clear way to quantify cause and effect. Our physics-based explainable AI framework addresses these limitations.— Professor Masato Kotsugi, Tokyo University of Science
Our eX-GL approach effectively automates the interpretation of complex magnetization reversal and enables identification of hidden mechanisms difficult to discern using conventional methods.— Professor Masato Kotsugi
La Conversación del Hearth Otra perspectiva de la historia
Why has this particular problem—magnetic energy loss—been so hard to solve until now?
Because you're dealing with structures too small to see clearly and too complex to simulate accurately. You could photograph the maze domains and measure the heat loss, but you couldn't connect the two. There were too many interacting factors—temperature, material structure, entropy, magnetic forces—all happening at once.
So the AI model is solving a translation problem?
Exactly. It's translating between what you can observe—the actual shape of the magnetic maze—and what you can measure—the energy loss. Persistent homology finds the hidden patterns in the images, machine learning figures out which patterns matter, and then physics connects those patterns to the energy barriers.
Why does this matter for electric vehicles specifically?
Because motors in EVs waste energy as heat through these magnetic losses. It's one of the biggest hidden drains. If you can understand and reduce it, you get longer range or smaller, cheaper batteries. Every percentage point counts.
Can this approach work on other problems?
That's the real promise. The researchers built this on fundamental physics—free energy, thermodynamics—not just pattern recognition. Those principles apply to lots of systems. You could potentially use the same framework to optimize other magnetic materials, superconductors, even phase transitions in other materials.
What's the practical next step?
Designing motor materials with fewer of these energy barriers, or operating strategies that avoid the worst of them. But first, engineers need to validate these findings in real motors and see if the efficiency gains match the theory.