Flat-band materials hold the key to unlocking fascinating quantum phenomena, from unconventional superconductivity, to nontrivial topology, and strong electron correlations, to name but a few. But in a vast materials space, how do we efficiently find the most promising candidates?

🔍 Meet Elf - an unsupervised AI model that learns electronic fingerprints directly from band structures. Elf autonomously clusters flat-band materials by electronic properties, uncovering hidden patterns and identifying promising candidates for experiments.

💡 Why does this matter?

• AI-driven electronic fingerprinting allows the classification and prediction of new materials beyond traditional paradigms.
• The method bridges the gap between high-throughput computation and materials discovery.
• Our approach can help accelerate the search for exotic quantum phases in materials.

Here's how it works:

🖼 Band structures as input - Elf analyses electronic band structures obtained from DFT calculations. These are processed as images, allowing the model to learn directly from electronic properties without bias toward crystal structures.

🧠 Feature extraction with AI - a convolutional autoencoder (ResNet18-based) compresses band structures into low-dimensional electronic fingerprints. This transformation captures key features of the band dispersion.

📊 Clustering in fingerprint space - using unsupervised learning techniques like HDBSCAN and UMAP, Elf groups materials based on their electronic fingerprints. This mapping helps reveal hidden patterns, chemical trends, and novel material classes that wouldn’t be obvious from conventional analysis.

🔬 Identifying promising candidates - by comparing newly discovered materials to well-known reference compounds, Elf highlights candidates with similar electronic properties. This accelerates the search for materials with desirable quantum behaviour and helps bridge the gap between computation and experimental validation.

🤗 Try it on Hugging Face:  https://huggingface.co/2Dmatters/Elf_encoder 

We’d love to hear your thoughts! Could AI-powered discovery transform the way we explore quantum materials? Let’s discuss! 👇