Machine learning unlocks a new class of magnetic materials

A team of researchers at the Tokyo University of Science has achieved something that materials scientists have been seeking for decades: the successful creation of bulk ferromagnetic quasicrystals that can be annealed and studied in detail. The breakthrough, reported in the Journal of the American Chemical Society, could open a new frontier in magnetic materials research and eventually influence next-generation sensing, computing, and quantum technologies.

The significance of the achievement lies not only in the creation of a novel material but also in the advanced technology used to discover it. By combining machine learning with experimental metallurgy, the researchers have demonstrated how artificial intelligence is becoming a powerful tool for materials discovery.

What are quasicrystals?

Quasicrystals occupy an unusual position in material science. Unlike conventional crystals, which display repeating atomic patterns, quasicrystals exhibit long-range order without periodic repetition. This gives them unique symmetries that are impossible in ordinary crystals, including five-fold and icosahedral arrangements.  

Since their discovery in the 1980s, quasicrystals have fascinated scientists because they challenge traditional ideas about how matter organizes itself. Their unusual atomic structures produce uncommon mechanical, thermal, and electronic properties. Yet understanding their magnetic behaviour has remained difficult.

While ferromagnetism, which is the phenomenon that allows materials to act as permanent magnets, has been extensively studied in conventional crystalline solids and amorphous materials, quasicrystals have largely remained outside this field. Ferromagnetic quasicrystals had been produced previously, but only through rapid quenching techniques that generated metastable materials with structural imperfections. These limitations prevented researchers from systematically studying their intrinsic magnetic properties.

The Tokyo team addressed this problem with a technology-driven approach. Rather than relying solely on trial-and-error experimentation, the researchers employed a machine-learning-based phase classifier to predict promising alloy compositions. Drawing on data from the HYPOD-X quasicrystal database and other materials datasets, the algorithm evaluated hundreds of potential candidates.

The model generated 675 potential quinary alloy systems and identified gold-copper-aluminium-indium alloys containing rare-earth elements as the most promising route toward stable ferromagnetic quasicrystals.

Using this guidance, the researchers synthesized three new materials containing gadolinium (Gd), terbium (Tb), and dysprosium (Dy). Importantly, the materials were produced using standard arc-melting techniques followed by controlled annealing rather than rapid quenching. This resulted in structurally coherent quasicrystals with unprecedented thermal stability.

As Professor Ryuji Tamura explained, the machine-learning-assisted strategy enabled the development of ferromagnetic icosahedral quasicrystals with exceptional structural quality, making systematic investigation of their intrinsic magnetic behaviour possible for the first time.

Revealing hidden magnetic behaviour

The ability to anneal and stabilize these materials transformed them from scientific curiosities into genuine research platforms. Testing revealed that all three quasicrystals displayed long-range ferromagnetic order at temperatures ranging from 9.7 to 28.3 Kelvin. This provides clear evidence that stable ferromagnetism can exist in highly ordered quasiperiodic structures.

More interestingly, the team observed two distinctly different forms of magnetic critical behaviour. The terbium- and dysprosium-based materials behaved in a manner consistent with mean-field ferromagnetism, a model characterized by effectively long-range magnetic interactions. In contrast, the gadolinium-containing material exhibited behaviour associated with shorter-range magnetic interactions and stronger spin fluctuations.

These findings suggest that magnetic criticality in quasicrystals is governed by a combination of quasiperiodic atomic order and magnetic spin symmetry. This relationship has not previously been explored in such detail because suitable materials did not exist.

Why technology companies should pay attention

At first glance, materials that operate below 30 Kelvin may appear to have limited practical relevance. However, history shows that fundamental breakthroughs in magnetism often underpin future technological revolutions.

Advanced magnetic materials are essential for a wide range of technologies, including high-density data storage, spintronics, quantum computing, magnetic sensing systems, and energy-conversion devices. Understanding how quasiperiodic structures influence magnetic interactions could lead to entirely new classes of functional materials with tunable properties.

The research also highlights a broader trend transforming materials science: the integration of artificial intelligence into discovery pipelines. Machine-learning systems are increasingly being used to identify promising compounds, predict material properties, and reduce the time required to move from theoretical concepts to experimental validation.

In traditional materials development, discovering a new alloy can require years of screening and testing. AI-guided approaches dramatically narrow the search space, allowing researchers to focus on the most promising candidates. The success of the Tokyo University of Science team demonstrates how computational intelligence and laboratory experimentation can work together to accelerate innovation.