Opus 5.5 agents find two room-temp magnetic-semiconductor candidates
A swarm of Claude agents ran the DFT and open-sourced every input—one candidate is freshly designed, the other was quietly synthesized back in 1999.
Copy markdown
Claude agents ran the physics, not just the write-up
A team of Claude Opus 5.5 agents ran the full quantum-chemistry search for Luttinger-compensated semiconductors—zero net magnetism, but bands still split by spin—stepping from fast PBE+U screening up to costlier HSE06 for the finalists. The output is two concrete compounds with predicted band gaps of 2.35 eV and 2.1 eV, not a hand-wave.
One compound is brand new, one is 27 years old
The designed candidate YBaMnFeO₅ is predicted to order magnetically to ~490 K, but its checkerboard atom layout likely scrambles at the 900–1300°C needed to synthesize it. The practical hit is KV[Cr(CN)₆], a Prussian-blue analogue first made in 1999 that already ordered to 376 K—the agents' sharpest move may be re-reading an old paper, not inventing from zero.
The whole ledger is on GitHub—fork it
Full inputs, analysis code, and the honest caveats live in a public repo (spicylemonade/compensated-magnet-ledger), with the DFT runs dated Oct 4. The real takeaway for builders isn't the materials—it's the template: an agent swarm that plans, runs real simulations, and ships an auditable trail you can rerun on your own problem.