Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation
An LLM agent produced AutoSella, two optimizer variants that cut force-call counts against Sella on held-out molecular tests.
The paper says the agent rewrote the optimizer under gates meant to block early stopping and non-generalizing gains. Starting from Sella, it found variants that reduced force calls across benchmarks and potentials outside the search set. At the r2SCAN-3c DFT level, the best variant used 40.2% to 77.2% of Sella’s force calls for the same energy reduction, despite not using DFT gradients. Source: HF Daily Papers' note.
The paper says the agent rewrote the optimizer under gates meant to block early stopping and non-generalizing gains. Starting from Sella, it found variants that reduced force calls across benchmarks and potentials outside the search set. At the r2SCAN-3c DFT level, the best variant used 40.2% to 77.2% of Sella’s force calls for the same energy reduction, despite not using DFT gradients. Source: HF Daily Papers' note.
score 5