91直播

A new physics-informed machine-learning method could help researchers find two-dimensional materials with unusual electronic properties more quickly and with fewer calculations. 

Researchers at The University of Manchester have developed a new computational approach to help identify two-dimensional materials that may host unusual quantum behaviour. The work, published in focuses on materials with 鈥渇lat bands鈥�, electronic states where electrons have very little kinetic energy. In these materials, interactions between electrons can become much more important, creating conditions linked to phenomena such as magnetism, unconventional superconductivity and topological electronic behaviour.  

Finding real materials with flat bands from large dataset is difficult. Conventional searches often rely on density functional theory calculations, which can reveal a material鈥檚 electronic structure but are time-consuming when applied across thousands of possible candidates. The 91直播 team took a different route. They developed a physics-informed scoring system that captures two signatures of flat-band behaviour, low band dispersion and a strong peak in the density of states, then trained a model to estimate that score directly from atomic structure. 

鈥淔lat bands are not only a feature we see in electronic calculations. They are often connected to the geometry of atoms in a material.鈥� said Dr Xiangwen Wang, leading author of the study. 鈥淥ur approach learns from that structure, which means we can search much larger materials spaces in a more targeted and interpretable way.鈥� 

The framework was trained using known two-dimensional materials and then applied to more than 10,000 unlabelled 2D materials. Among high-scoring candidates with kagome-like structural motifs, follow-up quantum calculations confirmed flat-band behaviour with 98.2% accuracy. The study also identified several materials predicted to host fragile topological flat bands, a form of electronic topology associated with strongly correlated quantum phases. These results suggest that the method can do more than sort large datasets, it can help reveal which structural features make certain materials promising for further study. 

, Senior Research Fellow in the  at The University of Manchester, said: 鈥淭he exciting part is not only that we found new candidate materials, but that the method changes how we search. Rather than calculating everything first and looking afterwards, we can now use physical intuition and structural learning to guide the search from the beginning. That makes discovery more scalable and more interpretable.鈥� 

The approach remains computational, so experimental work will be needed to test the most promising candidates in the laboratory. However, the researchers say the same strategy could be adapted to search for other classes of quantum materials, provided the target property can be expressed as a meaningful physics-based score. By connecting physical insight with structure-based learning, the study offers a more efficient way to move from large materials databases to shortlists of candidates for detailed quantum calculations and experimental validation.