Lila Sciences' AI finds palladium catalyst a 12-year expert rejected, running 1,000+ hours
AndrewLBeam · x · 2026-09-25
Lila Sciences shares how its autonomous science platform produced an unexpected breakthrough in clean hydrogen catalysts.
- Context: The oxygen evolution reaction (OER) in PEM electrolyzers must run under strong acid and high voltage; the industry-standard iridium oxide relies on one of Earth's rarest metals.
- The surprise: John Gregoire, a former Caltech research professor who spent 12 years on this exact reaction, thought the AI's palladium-based oxide suggestions had gone astray — combinations he wouldn't have let grad students test. The measurements were promising, so the team kept going.
- Results: Across 2,942 catalysts spanning 53 material systems and 26 elements, the platform screened 240 catalysts per week, roughly 10x faster than a standard lab. The best candidate adds tiny amounts of indium and manganese (0.09% of metal atoms) and held overpotential below 0.5 V at 10 mA/cm² in 1 M sulfuric acid for over 1,000 hours while retaining 96% of its palladium; unmodified palladium oxide crossed the threshold after only 200 hours.
- Microscopy revealed a needle-like structure formed during operation that may explain the durability; the mechanism is still under study.
More from Research
- UMAP 0.6 docs: how to embed million-scale datasets, tuned via 3M arXiv papers — leland_mcinnes · 2026-11-11
- UMAP 0.6 now installable via PyPI and conda-forge — leland_mcinnes · 2026-11-11
- UMAP 0.6 optimizers lift attraction-repulsion restrictions, boost kNN recall — leland_mcinnes · 2026-11-11
- UMAP 0.6 recursive init converges to within 3% of final layout in far fewer epochs — leland_mcinnes · 2026-11-11
- UMAP 0.6 fidelity gains at scale driven by efficient hard-negative mining — leland_mcinnes · 2026-11-11
- UMAP 0.6 recursive init: UMAP initializing itself, 32s vs 4min for spectral — leland_mcinnes · 2026-11-11