CosmoAgent: AI Agent for Theoretical Cosmology

Abstract

We present CosmoAgent, an agentic AI framework using large language models (LLMs) to search for dark-energy and modified-gravity models. We define a covariant grammar that allows LLM agents to explore the space of modified-gravity Lagrangians. An in-house computational pipeline derives the equations of motion, solves the background dynamics, and evaluates the likelihood using late-time expansion data, including the DES 5-year type Ia supernova sample and the DESI Data Release 2 BAO measurements. Theories are ranked by the Akaike Information Criterion (AIC) to balance simplicity and likelihood. We benchmarked 8 frontier LLMs from ChatGPT, Claude, Gemini, Kimi, DeepSeek, and Qwen, and all LLMs are able to find theories with a lower nominal AIC than flat ΛCDM, with a best improvement of Δ AIC=-3.328. Seven recover quadratic quintessence. Exploratory searches also return induced-gravity and scalar-Gauss-Bonnet structures. The modest AIC improvements do not establish an observational preference for these models over flat ΛCDM. The code is publicly available at https://github.com/wsunamoutlook-prog/AI-Modified-Gravity

Theoretical consistency and observational constraints on entropic cosmology
Yunliang Ren, Xiaofeng Yang, Xuwei Zhang +3
Physics Letters B30 Oct 2026Peer-reviewedCC BY 4.0Full textCosmologyCrossref
Theoretical Consistency and Observational Constraints on Entropic Cosmology
Yunliang Ren, Xiaofeng Yang, Xuwei Zhang +3
arXiv29 Oct 2026PreprintCC BY 4.0Full textCosmology Gravitation & general relativityarXiv
Cosmology with weak lensing surveys
Dipak Munshi, Patrick Valageas, Ludovic van Waerbeke +1
Physics Reports22 Sep 2026Peer-reviewedLicence not statedFull text299 citedCosmologyHAL
On entropic cosmology, late-time acceleration and the quantum bounce
Luis Rey Díaz-Barrón, G.E. Pérez-Cuéllar, M. Sabido
Physics Letters B30 Oct 2026Peer-reviewedCC BY 4.0Full textCosmologyCrossref
The Breath Cannot Close: Why a Bounded Dark-Energy Sinusoid Forecloses a Cyclic Cosmology
Nicholas Archer Sanders
Zenodo (CERN European Organization for Nuclear Research)16 Jul 2026PreprintCC BY34 citedCosmologyOpenAlex

Cite

BibTeX RIS

BibTeX

@misc{alexander2026cosmoagent,
  title = {{CosmoAgent: AI Agent for Theoretical Cosmology}},
  author = {Stephon Alexander and Loukas Gouskos and Savvas M. Koushiappas and Cooper Niu and Wenrong Sun},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.16536},
  eprint = {2610.16536},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.16536},
}

RIS

TY  - UNPB
TI  - CosmoAgent: AI Agent for Theoretical Cosmology
AU  - Stephon Alexander
AU  - Loukas Gouskos
AU  - Savvas M. Koushiappas
AU  - Cooper Niu
AU  - Wenrong Sun
PY  - 2026
DA  - 2026/10/23
PB  - arXiv
DO  - 10.48550/arxiv.2610.16536
UR  - https://arxiv.org/abs/2610.16536
AB  - We present CosmoAgent, an agentic AI framework using large language models (LLMs) to search for dark-energy and modified-gravity models. We define a covariant grammar that allows LLM agents to explore the space of modified-gravity Lagrangians. An in-house computational pipeline derives the equations of motion, solves the background dynamics, and evaluates the likelihood using late-time expansion data, including the DES 5-year type Ia supernova sample and the DESI Data Release 2 BAO measurements. Theories are ranked by the Akaike Information Criterion (AIC) to balance simplicity and likelihood. We benchmarked 8 frontier LLMs from ChatGPT, Claude, Gemini, Kimi, DeepSeek, and Qwen, and all LLMs are able to find theories with a lower nominal AIC than flat ΛCDM, with a best improvement of Δ AIC=-3.328. Seven recover quadratic quintessence. Exploratory searches also return induced-gravity and scalar-Gauss-Bonnet structures. The modest AIC improvements do not establish an observational preference for these models over flat ΛCDM. The code is publicly available at https://github.com/wsunamoutlook-prog/AI-Modified-Gravity
ER  -

Abstract

Related

Cite