Reconstructing the cosmological density and velocity fields from redshifted galaxy distributions using V-net

Abstract

The distribution of matter that is measured through galaxy redshift and peculiar velocity surveys can be harnessed to learn about the physics of dark matter, dark energy, and the nature of gravity. To improve our understanding of the matter of the Universe, we can reconstruct the full density and velocity fields from the galaxies that act as tracer particles. In this paper, we use the simulated halos as proxies for the galaxies. We use a convolutional neural network, a V-net, trained on numerical simulations of structure formation to reconstruct the density and velocity fields. We find that, with detailed tuning of the loss function, the V-net could produce better fits to the density field in the high-density and low-density regions, and improved predictions for the probability distribution of the amplitudes of the velocities. However, the weights will reduce the precision of the estimated β parameter. We also find that the redshift-space distortions of the halo catalogue do not significantly contaminate the reconstructed real-space density and velocity field. We estimate the velocity field β parameter by comparing the peculiar velocities of halo catalogues to the reconstructed velocity fields, and find the estimated β values agree with the fiducial value at the 68% confidence level.

Peculiar Velocity Reconstruction from Simulations and Observations Using Deep Learning Algorithms

“…In recent years, the utilization of machinelearning algorithms in astrophysics has rapidly increased, including cosmological value predictions (e.g., Wang et al 2021b;Chunduri & Mahesh 2023), galaxy morphological classification (e.g., Domínguez Sánchez et al 2022;, and more (see Baron 2019 for a recent overview). Many studies have employed deep learning algorithms in velocity reconstruction (e.g., Wu et al 2021;Qin et al 2023;Wu et al 2023;Ganeshaiah Veena et al 2023). Diverging from theorydriven algorithms, deep learning operates as a data-driven approach, with its effectiveness directly tied to the quality and quantity of the training data set and its performance evaluated by the employed loss function.…”

Cosmology inference at the field level from biased tracers in redshift-space

“…The recovery of initial conditions from simulated mock galaxies at fixed cosmology has been explored with various forward models, such as Lagrangian Perturbation Theory (LPT) [15][16][17], full particle-mesh simulations [18] or neural networks [19][20][21][22][23][24][25]. Field-level analyses have also been applied to infer initial conditions in the nearby universe [26] and from the BOSS SDSS-III catalog [27] at fixed cosmology.…”

Cosmological Parameter Constraints from the SDSS Density and Momentum Power Spectra

“…Previously intractable higher point statistics are increasingly studied due to rapid advances in numerical statistical modeling and larger data volumes (Gil-Marín et al 2017;Yankelevich & Porciani 2019;Philcox 2022). Alternative approaches to the standard N-point functions are also progressively being explored (Pisani et al 2015;Pan et al 2020;Uhlemann et al 2020;Villaescusa-Navarro et al 2021;Appleby et al 2022;Qin et al 2023).…”