In EDS ansehen

Robust Field-level Likelihood-free Inference with Galaxies

Gespeichert in:
Bibliographische Detailangaben
Titel: Robust Field-level Likelihood-free Inference with Galaxies
Autoren: Natalí S. M. de Santi, Helen Shao, Francisco Villaescusa-Navarro, L. Raul Abramo, Romain Teyssier, Pablo Villanueva-Domingo, Yueying Ni, Daniel Anglés-Alcázar, Shy Genel, Elena Hernández-Martínez, Ulrich P. Steinwandel, Christopher C. Lovell, Klaus Dolag, Tiago Castro, Mark Vogelsberger
Quelle: The Astrophysical Journal, Vol 952, Iss 1, p 69 (2023)
Publication Status: Preprint
Verlagsinformationen: American Astronomical Society, 2023.
Publikationsjahr: 2023
Schlagwörter: FOS: Computer and information sciences, Computer Science - Machine Learning, Tensor Decompositions and Applications in Multilinear Algebra, Cosmology and Nongalactic Astrophysics (astro-ph.CO), Astrophysical Simulations, Cosmological parameters, Hydrodynamical simulations, FOS: Physical sciences, Astrophysics, 01 natural sciences, Machine Learning (cs.LG), Machine Learning, Astrostatistics, Artificial Intelligence, Field (mathematics), 0103 physical sciences, FOS: Mathematics, 0101 mathematics, Galaxy formation and evolution, Galaxy Formation and Evolution in the Universe, Statistical Modeling, Physics, Magnetohydrodynamical simulations, Pure mathematics, Astronomy and Astrophysics, Scientific Computing, Astrophysics - Astrophysics of Galaxies, Computer science, Cosmology, QB460-466, Algorithm, Computational Mathematics, Galaxy, Physics and Astronomy, Astrophysics of Galaxies (astro-ph.GA), Physical Sciences, Computer Science, Nonnegative Tensor Factorization, Scientific Computing and Data Analysis with Python, Mathematics, Astrophysics - Cosmology and Nongalactic Astrophysics
Beschreibung: We train graph neural networks to perform field-level likelihood-free inference using galaxy catalogs from state-of-the-art hydrodynamic simulations of the CAMELS project. Our models are rotational, translational, and permutation invariant and do not impose any cut on scale. From galaxy catalogs that only contain 3D positions and radial velocities of ∼1000 galaxies in tiny ( 25 h − 1 Mpc ) 3 volumes our models can infer the value of Ωm with approximately 12% precision. More importantly, by testing the models on galaxy catalogs from thousands of hydrodynamic simulations, each having a different efficiency of supernova and active galactic nucleus feedback, run with five different codes and subgrid models—IllustrisTNG, SIMBA, Astrid, Magneticum, SWIFT-EAGLE—we find that our models are robust to changes in astrophysics, subgrid physics, and subhalo/galaxy finder. Furthermore, we test our models on 1024 simulations that cover a vast region in parameter space—variations in five cosmological and 23 astrophysical parameters—finding that the model extrapolates really well. Our results indicate that the key to building a robust model is the use of both galaxy positions and velocities, suggesting that the network has likely learned an underlying physical relation that does not depend on galaxy formation and is valid on scales larger than ∼10 h −1 kpc.
Publikationsart: Article
Other literature type
ISSN: 1538-4357
0004-637X
DOI: 10.3847/1538-4357/acd1e2
DOI: 10.48550/arxiv.2302.14101
DOI: 10.60692/ttsb1-ey030
DOI: 10.60692/hw7re-n7w12
Zugangs-URL: http://arxiv.org/abs/2302.14101
https://doaj.org/article/aa15004170704853a946acac4e26dabb
Rights: CC BY
arXiv Non-Exclusive Distribution
Dokumentencode: edsair.doi.dedup.....ddaa8094e2bc269f1b70d88323de7c6e
Datenbank: OpenAIRE
Beschreibung
Abstract:We train graph neural networks to perform field-level likelihood-free inference using galaxy catalogs from state-of-the-art hydrodynamic simulations of the CAMELS project. Our models are rotational, translational, and permutation invariant and do not impose any cut on scale. From galaxy catalogs that only contain 3D positions and radial velocities of ∼1000 galaxies in tiny ( 25 h − 1 Mpc ) 3 volumes our models can infer the value of Ωm with approximately 12% precision. More importantly, by testing the models on galaxy catalogs from thousands of hydrodynamic simulations, each having a different efficiency of supernova and active galactic nucleus feedback, run with five different codes and subgrid models—IllustrisTNG, SIMBA, Astrid, Magneticum, SWIFT-EAGLE—we find that our models are robust to changes in astrophysics, subgrid physics, and subhalo/galaxy finder. Furthermore, we test our models on 1024 simulations that cover a vast region in parameter space—variations in five cosmological and 23 astrophysical parameters—finding that the model extrapolates really well. Our results indicate that the key to building a robust model is the use of both galaxy positions and velocities, suggesting that the network has likely learned an underlying physical relation that does not depend on galaxy formation and is valid on scales larger than ∼10 h −1 kpc.
ISSN:15384357
0004637X
DOI:10.3847/1538-4357/acd1e2