LLM autoresearch builds AutoSella, cutting DFT force calls to 40-77% of Sella's
Artem Tsypin · hf · 2026-10-06
Geometry optimization dominates quantum-chemistry workloads: each step costs a force evaluation, which at DFT level dominates wall time. The authors set an autoresearch agent to rewrite the optimizer itself, minimizing force-call counts under two admission gates that reject premature stopping and non-generalizing gains.
- Starting from Sella (the fastest open-source optimizer), the search yields AutoSella, a family of two optimizers
- Both consistently reduce force calls on held-out benchmarks and potentials not seen during search
- At the r2SCAN-3c DFT level, the best variant needs only 40.2-77.2% of Sella's force calls for the same energy reduction, despite the agent never using DFT gradients
More from Research
- O(n²) matrix multiplication is almost certainly false even if ω = 2, says basedjensen — basedjensen · 2026-10-06
- Cohere Labs Releases Tiny Aya, a Family of Small Models Covering 70+ Languages — Cohere_Labs · 2026-10-06
- Will LLMs wreck elegant math? O(n²) limits are already crumbling — teortaxesTex · 2026-10-06
- ETH's AME-2 legged locomotion paper accepted at TRO, training code open-sourced — ChongZzZhang · 2026-10-06
- 0.8B model beats 2B on ARC-Challenge (42.15%) via closed-form weight surgery with zero backprop — AdventurousTwo6445 · 2026-10-06
- MIT team's Science Task Taxonomy maps 232 subfields and 208,202 scientific tasks — JMateosGarcia · 2026-10-06