The CAMELS-CROCODILE Simulation Suite: A New Cosmology--Astrophysics Playground for Machine Learning
Abstract
We present CAMELS-CROCODILE, a suite of cosmological hydrodynamic simulations that extends the CAMELS framework with the \gadget{} smoothed particle hydrodynamics code and the Osaka feedback model. As a companion to the IllustrisTNG-based second-generation CAMELS suite, it provides an independent implementation of the baryonic physics that shapes galaxies, the intergalactic medium, and large-scale structure, broadening the range of subgrid models over which machine-learning (ML) inference can be marginalized and tested. The suite comprises $25$ and 50Mpc/h boxes organized into one-parameter, cosmic-variance, and Sobol-sequence sets that vary $26$ cosmological and astrophysical parameters around a fiducial Osaka model. We describe the simulation design and validate the suite against benchmarks and other CAMELS suites. Compared with IllustrisTNG, the fiducial model suppresses the matter power spectrum by only a few per cent, against up to ∼27% at z=0, and forms stars with similar efficiency below 1012.2 M⊙/h but up to $2.5$ times more efficiently in group-scale halos, both consistent with weaker AGN feedback. We also present the stellar and black hole bias and the Lyα absorption around galaxies. As a first ML application, we apply a graph neural network trained on IllustrisTNG galaxy catalogs, with frozen weights, to CAMELS-CROCODILE. It fails to recover Ωm and σ8 and returns overconfident posteriors, mainly for simulations containing more galaxies than any in its training set; within that range Ωm is recovered with an error only $1.5$ times the in-distribution value. These results underline the need to train and test ML inference across physically distinct galaxy formation models.
Related
- AI's Capability in Assisting Scientific Research in Physics, Astrophysics, and Cosmology I: Literature Review
- AI's Capability in Assisting Scientific Research in Physics, Astrophysics, and Cosmology II: Project Planning and Proposal Evaluation
- Cosmology & Astrophysics of the Dark Dimension with Bulk Majorana Neutrinos
- Distinguishing cosmology, astrophysics and systematics: the Bayesian search for a primordial gravitational-wave background in next-generation observatories
- Graph Neural Networks for Astrophysics and Cosmology
BibTeX
@misc{nagamine2026camelscrocodile,
title = {{The CAMELS-CROCODILE Simulation Suite: A New Cosmology--Astrophysics Playground for Machine Learning}},
author = {Kentaro Nagamine and Yuri Oku and Atsushi J. Nishizawa and Jun-Young Lee and Francisco Villaescusa-Navarro and Shy Genel and Daniel Anglés-Alcázar},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.23982},
eprint = {2609.23982},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.23982},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - The CAMELS-CROCODILE Simulation Suite: A New Cosmology--Astrophysics Playground for Machine Learning
AU - Kentaro Nagamine
AU - Yuri Oku
AU - Atsushi J. Nishizawa
AU - Jun-Young Lee
AU - Francisco Villaescusa-Navarro
AU - Shy Genel
AU - Daniel Anglés-Alcázar
PY - 2026
DA - 2026/09/21
PB - arXiv
DO - 10.48550/arxiv.2609.23982
UR - https://arxiv.org/abs/2609.23982
AB - We present CAMELS-CROCODILE, a suite of cosmological hydrodynamic simulations that extends the CAMELS framework with the \gadget{} smoothed particle hydrodynamics code and the Osaka feedback model. As a companion to the IllustrisTNG-based second-generation CAMELS suite, it provides an independent implementation of the baryonic physics that shapes galaxies, the intergalactic medium, and large-scale structure, broadening the range of subgrid models over which machine-learning (ML) inference can be marginalized and tested. The suite comprises $25$ and 50Mpc/h boxes organized into one-parameter, cosmic-variance, and Sobol-sequence sets that vary $26$ cosmological and astrophysical parameters around a fiducial Osaka model. We describe the simulation design and validate the suite against benchmarks and other CAMELS suites. Compared with IllustrisTNG, the fiducial model suppresses the matter power spectrum by only a few per cent, against up to ∼27% at z=0, and forms stars with similar efficiency below 10^12.2 M_⊙/h but up to $2.5$ times more efficiently in group-scale halos, both consistent with weaker AGN feedback. We also present the stellar and black hole bias and the Lyα absorption around galaxies. As a first ML application, we apply a graph neural network trained on IllustrisTNG galaxy catalogs, with frozen weights, to CAMELS-CROCODILE. It fails to recover Ω_m and σ_8 and returns overconfident posteriors, mainly for simulations containing more galaxies than any in its training set; within that range Ω_m is recovered with an error only $1.5$ times the in-distribution value. These results underline the need to train and test ML inference across physically distinct galaxy formation models.
ER -