New physics-based machine-learning method speeds search for 2D quantum materials

Imagine searching for a tiny, perfect grain of sand on a beach — but instead of looking grain by grain, you could spot the promising ones from above. That's roughly what researchers at the University of Manchester have done for a special class of quantum materials, and their new method is nearly perfect at picking winners.

A team led by Dr. Xiangwen Wang has built a machine learning tool that can identify 2D materials likely to host unusual quantum behavior. These are materials just a few atoms thick that could someday power better electronics, superconductors, or other technologies. The work was published in the journal Science Advances.

The key? Flat bands. That's what scientists call electronic states inside materials where electrons hardly move at all. When electrons are stuck in place, the push and pull between them becomes much more important — and that can lead to strange and useful behaviors like magnetism or a type of superconductivity that scientists still don't fully understand.

The old way of finding these materials was slow. Researchers would run heavy calculations on thousands of candidates one by one, which took lots of time and computer power. The Manchester team took a smarter route. They taught their model to look at a material's atomic structure — the way its atoms are arranged — and guess whether it might have flat bands, without doing all the heavy math upfront.

"Flat bands are not only a feature we see in electronic calculations," Dr. Wang said. "They are often connected to the geometry of atoms in a material. Our approach learns from that structure, which means we can search much larger materials spaces in a more targeted and interpretable way."

The team trained their tool using known 2D materials and then tested it on more than 10,000 materials nobody had looked at closely before. Among the most promising candidates — those with a specific atomic pattern called a kagome structure, which looks a bit like a woven basket — the method predicted flat-band behavior with 98.2% accuracy. That's nearly flawless.

The researchers also spotted several materials that appear to host what are called fragile topological flat bands, a special type of electronic behavior that scientists believe is connected to some of the most mysterious quantum phenomena.

Dr. Qian Yang, a senior researcher at the National Graphene Institute at Manchester, said the method flips the old search process on its head. "Rather than calculating everything first and looking afterward, we can now use physical intuition and structural learning to guide the search from the beginning," she said. "That makes discovery more scalable and more interpretable."

The work is still computational — the team hasn't yet tested their top candidates in a lab. But they say the same approach could be adapted to hunt for other kinds of quantum materials, making the whole process of discovery faster and smarter.