TY - JOUR
T1 - Robust Field-level Inference of Cosmological Parameters with Dark Matter Halos
AU - Shao, Helen
AU - Villaescusa-Navarro, Francisco
AU - Villanueva-Domingo, Pablo
AU - Teyssier, Romain
AU - Garrison, Lehman H.
AU - Gatti, Marco
AU - Inman, Derek
AU - Ni, Yueying
AU - Steinwandel, Ulrich P.
AU - Kulkarni, Mihir
AU - Visbal, Eli
AU - Bryan, Greg L.
AU - Anglés-Alcázar, Daniel
AU - Castro, Tiago
AU - Hernández-Martínez, Elena
AU - Dolag, Klaus
N1 - Funding Information:
We thank Douglas Potter for his help with the PKDGrav3 simulations. We also thank Tom Abel, Simeon Bird, Shy Genel, and David Spergel for enlightening discussions. The networks have been trained using the Tiger cluster at Princeton University. H.S. thanks the Flatiron Institute for the support during the preparation of this work. The work of F.V.N. has been supported by NSF grant AST-2108078. E.V. is supported by NSF grant AST-2009309 and NASA grant 80NSSC22K0629. D.A.A. acknowledges support by NSF grants AST-2009687 and AST-2108944, CXO grant TM2-23006X, and Simons Foundation award CCA-1018464. T.C. is supported by the INFN INDARK PD51 grant and by the FARE MIUR grant “ClustersXEuclid” R165SBKTMA. E.H.M. was supported by the grant agreements ANR-21-CE31-0019/490702358 from the French Agence Nationale de la Recherche/DFG for the LOCALIZATION project. K.D. acknowledges support through the COMPLEX project from the European Research Council (ERC) under the European Unions Horizon 2020 research and innovation program grant agreement ERC-2019-AdG 882679 as well as support by the Deutsche Forschungsgemeinschaft (DFG; German Research Foundation) under Germanys Excellence Strategy—EXC-2094-390783311. The CAMELS project is supported by NSF grants AST-2108944, AST-2108678, and AST-21080784. The Flatiron Institute is supported by the Simons Foundation. Kavli IPMU is supported by World Premier International Research Center Initiative (WPI), MEXT, Japan.
Publisher Copyright:
© 2023. The Author(s). Published by the American Astronomical Society.
PY - 2023/2/1
Y1 - 2023/2/1
N2 - We train graph neural networks on halo catalogs from Gadget N-body simulations to perform field-level likelihood-free inference of cosmological parameters. The catalogs contain ≲5000 halos with masses ≳1010 h −1 M ⊙ in a periodic volume of ( 25 h − 1 Mpc ) 3 ; every halo in the catalog is characterized by several properties such as position, mass, velocity, concentration, and maximum circular velocity. Our models, built to be permutationally, translationally, and rotationally invariant, do not impose a minimum scale on which to extract information and are able to infer the values of Ωm and σ 8 with a mean relative error of ∼6%, when using positions plus velocities and positions plus masses, respectively. More importantly, we find that our models are very robust: they can infer the value of Ωm and σ 8 when tested using halo catalogs from thousands of N-body simulations run with five different N-body codes: Abacus, CUBEP3M, Enzo, PKDGrav3, and Ramses. Surprisingly, the model trained to infer Ωm also works when tested on thousands of state-of-the-art CAMELS hydrodynamic simulations run with four different codes and subgrid physics implementations. Using halo properties such as concentration and maximum circular velocity allow our models to extract more information, at the expense of breaking the robustness of the models. This may happen because the different N-body codes are not converged on the relevant scales corresponding to these parameters.
AB - We train graph neural networks on halo catalogs from Gadget N-body simulations to perform field-level likelihood-free inference of cosmological parameters. The catalogs contain ≲5000 halos with masses ≳1010 h −1 M ⊙ in a periodic volume of ( 25 h − 1 Mpc ) 3 ; every halo in the catalog is characterized by several properties such as position, mass, velocity, concentration, and maximum circular velocity. Our models, built to be permutationally, translationally, and rotationally invariant, do not impose a minimum scale on which to extract information and are able to infer the values of Ωm and σ 8 with a mean relative error of ∼6%, when using positions plus velocities and positions plus masses, respectively. More importantly, we find that our models are very robust: they can infer the value of Ωm and σ 8 when tested using halo catalogs from thousands of N-body simulations run with five different N-body codes: Abacus, CUBEP3M, Enzo, PKDGrav3, and Ramses. Surprisingly, the model trained to infer Ωm also works when tested on thousands of state-of-the-art CAMELS hydrodynamic simulations run with four different codes and subgrid physics implementations. Using halo properties such as concentration and maximum circular velocity allow our models to extract more information, at the expense of breaking the robustness of the models. This may happen because the different N-body codes are not converged on the relevant scales corresponding to these parameters.
UR - http://www.scopus.com/inward/record.url?scp=85147803124&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85147803124&partnerID=8YFLogxK
U2 - 10.3847/1538-4357/acac7a
DO - 10.3847/1538-4357/acac7a
M3 - Article
AN - SCOPUS:85147803124
SN - 0004-637X
VL - 944
JO - Astrophysical Journal
JF - Astrophysical Journal
IS - 1
M1 - 27
ER -