Opus 5.5 agents discover two room-temperature magnetic semiconductor candidates

What it claims

DFT simulations run by AI agents working with one human author produced two candidate Luttinger-compensated (LC) semiconductors. These materials have zero net spin moment, but the electron states at the band edges all carry one spin.

• YBaMnFeO₅ is a new design. As far as the authors could find, it has never been made or proposed as this kind of magnet.
• Predicted band gap: 2.35 eV.
• Spin-sorted windows: 1.0 eV for holes and 1.4 eV for electrons.
• Magnetic ordering: about 420 K in the raw simulation, or about 490 K after calibrating against a known magnet.
• KV[Cr(CN)₆] is a Prussian-blue analogue first made in 1999.
• Predicted band gap: about 2.1 eV.
• Spin-sorted windows: 2.6 eV for holes and 1.6 eV for electrons.

The 376 K ordering temperature for KV[Cr(CN)₆] was measured by the 1999 chemists. It is not an agent prediction, as the summary implies. That sample measured 365 K after it had been heated.

The novelty claim for KV[Cr(CN)₆] is narrow:

• Its zero net moment was designed in deliberately in 1999.
• A 2008 hybrid-functional study already plotted spin-resolved states showing both band edges with the same spin, but never commented on it.

The new contribution is recognising the compound as LC, putting numbers on the windows, and testing how robust they are.

How it works

• Two DFT levels. PBE+U is the faster method and HSE06 the slower, usually more accurate one. All reported gaps and spin windows come from HSE06.
• Figure of merit. The spin window is compared against room-temperature thermal energy of about 26 meV. A window much larger than that keeps electrons sorted.
• Why the cobalt-free cyanide works. Chromium bonds to the carbon end of each cyanide and vanadium to the nitrogen end. This locks each metal into its own site, which is the site ordering YBaMnFeO₅ lacks.

The page does not describe the agent setup: no prompts, tooling, number of agents or how much was autonomous. The work is framed as "a team of AI agents and I."

How to use it

The public repo is github.com/spicylemonade/compensated-magnet-ledger. According to the page, it contains:

• input files
• raw outputs
• analysis code
• a one-command checker
• independent re-runs
• a list of known caveats

The page does not give the checker command itself.

Limits

YBaMnFeO₅ is probably not synthesisable in its useful form.

• Its spin sorting needs Mn and Fe in a perfect checkerboard.
• The agents' simulations show that ordering dissolves into a random mix at about 950 K.
• Standard synthesis runs at 900–1300 °C, and the atoms barely move at lower temperatures, so a scrambled crystal is likely.
• A scrambled crystal loses the spin sorting.

KV[Cr(CN)₆] predictions assume a perfect, dry crystal. The only sample, from 1999, differs from that:

• It is a powder with water in its pores.
• It showed a residual moment of 0.125 μB per formula unit, where the ideal crystal has zero.
• The two methods disagree on what the water does. HSE06 says the spin sorting survives; PBE+U says the hole window shrinks by more than half.

Nothing has been measured experimentally yet. Neither the band gap nor the spin sorting has been checked; the stated next step is to re-synthesise the compound and measure them.

Zero net moment holds only approximately. Luttinger's theorem guarantees it only for a perfect crystal near absolute zero. Spin–orbit coupling and heat can leave a small imbalance.

Context. A 2025 study predicted two other LC semiconductors, and both lose magnetic order below room temperature. That study named a room-temperature LC semiconductor as the open goal, which is the gap this work targets.