Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation Algorithms
Abstract
Geometry optimization is a major cost in many quantum-chemical workflows: each optimization step requires one force evaluation, and at the density-functional level that evaluation dominates the wall time. Research in this area has produced a broad range of optimization methods, and we ask whether a language model can improve on the best of them through autoresearch. An agent rewrites the optimizer itself to minimize force-call counts, restrained by two admission gates that reject premature stopping and improvements that do not generalize to unseen molecules. Starting from Sella, the fastest open-source optimizer available, the search produces AutoSella, a family of two optimizers. Both of them deliver consistent force-call reductions relative to Sella across held-out molecular benchmarks and potentials not used during the search. Most notably, at the r2SCAN-3c DFT level, the best variant requires only 40.2-77.2% of Sella's force calls while achieving the same energy reduction, even though agent used no DFT gradients.
Related
- Molecular Electrodynamic Relaxation in Dielectric and Plasmonic Cavities
- Resolving Spin-Phonon Relaxation Pathways in Molecular Qubits via Regularized Regression
- Angular Momentum Fluctuations Induced by Phonon-Rotation Coupling Govern Nuclear Spin Relaxation in a Molecular Rotator Phase
- Collective Microscopic Dynamics in Aqueous LiTFSI from Dilute to Water-in-Salt Electrolytes: Insights from Quadrupolar NMR Relaxation and Molecular Simulations
- Optimizing Affinity and Pharmacokinetics of a Cysteine-sulfoxide Derivative as Potential Inhibitor of BACE-1 using Fragment Growing and Molecular Dynamics
BibTeX
@misc{tsypin2026optimizing,
title = {{Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation Algorithms}},
author = {Artem Tsypin and Vladimir Deshchenya and Kuzma Khrabrov and Denis Potapov and Maxim Radchenko and Artur Kadurin and Michael G. Medvedev},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.06577},
eprint = {2610.06577},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.06577},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation Algorithms
AU - Artem Tsypin
AU - Vladimir Deshchenya
AU - Kuzma Khrabrov
AU - Denis Potapov
AU - Maxim Radchenko
AU - Artur Kadurin
AU - Michael G. Medvedev
PY - 2026
DA - 2026/10/07
PB - arXiv
DO - 10.48550/arxiv.2610.06577
UR - https://arxiv.org/abs/2610.06577
AB - Geometry optimization is a major cost in many quantum-chemical workflows: each optimization step requires one force evaluation, and at the density-functional level that evaluation dominates the wall time. Research in this area has produced a broad range of optimization methods, and we ask whether a language model can improve on the best of them through autoresearch. An agent rewrites the optimizer itself to minimize force-call counts, restrained by two admission gates that reject premature stopping and improvements that do not generalize to unseen molecules. Starting from Sella, the fastest open-source optimizer available, the search produces AutoSella, a family of two optimizers. Both of them deliver consistent force-call reductions relative to Sella across held-out molecular benchmarks and potentials not used during the search. Most notably, at the r2SCAN-3c DFT level, the best variant requires only 40.2-77.2% of Sella's force calls while achieving the same energy reduction, even though agent used no DFT gradients.
ER -