Discovery of flat-band 2D materials via physics-informed scoring and structure-based learning

Area of Science:

• Condensed Matter Physics
• Materials Science
• Quantum Chemistry

Background:

• Flat electronic bands are crucial for exotic quantum phenomena like superconductivity and fractional topology.
• Identifying flat bands typically relies on computationally intensive density functional theory (DFT) calculations, limiting large-scale material exploration.

Purpose of the Study:

• To develop a structure-informed, data-driven framework for discovering novel flat-band two-dimensional materials.
• To enable systematic and large-scale screening of materials for flat-band properties, bypassing the need for precomputed band structures.

Main Methods:

• Introduced a physically motivated flatness score combining band dispersion and density-of-states features.
• Developed a multimodal deep learning model trained on atomic structure to predict flat-band propensity.
• Applied the framework to screen over 10,000 unlabeled materials.

Main Results:

• Identified multiple previously unrecognized flat-band two-dimensional materials.
• Achieved 98% accuracy in predicting flat bands in kagome-lattice materials with a flatness score > 0.9.
• Demonstrated the framework's ability to discover topologically nontrivial systems.

Conclusions:

• The developed framework enables large-scale, interpretable screening for flat-band materials.
• This approach expands the design space for correlated quantum phases in low-dimensional systems.
• Facilitates the discovery of materials with potential for unconventional superconductivity and fractional topology.