Neural Network Reconstruction of Non-Gaussian Initial Conditions from Dark Matter Halos

Abstract

We develop a machine learning approach to reconstructing the cosmological initial conditions from late-time dark matter halo number density fields in redshift space, with the goal of improving sensitivity to cosmological parameters, and in particular primordial non-Gaussianity. Using an U-Net architecture, our model achieves a cross-correlation accuracy of 44% for scales out to $k=0.4\text{ h}/\text{Mpc}$ between reconstructed and true initial conditions of Quijote 1 Gpc${}^3$ simulation boxes with an average halo number density of $\bar{n}=4\times 10^{-4}$ (h/Mpc)${}^3$ in the tracer field at $z=0$ . We demonstrate that our reconstruction is likely to be optimal for this setup and that it is highly effective at reducing redshift-space distortions. Using a Fisher analysis, we show that reconstruction improves cosmological parameter constraints derived from the power spectrum and bispectrum. By combining the power spectrum monopole, quadrupole, and bispectrum monopole up to $k_{\max }=0.52\text{ h}/\text{Mpc}$, our joint analysis of pre- and post-reconstructed fields from the Quijote simulation suite finds improved marginalized errors on all cosmological parameters. In particular, reconstruction improves constraints on $f_{\mathrm{NL}}$ by factors of 1.33, 1.88, and 1.57 for local, equilateral, and orthogonal shapes. Our findings demonstrate the effectiveness of reconstruction in decoupling modes, mitigating redshift-space distortions and maximizing information on cosmology. The results provide important insights into the amount of cosmological information that can be extracted from small scales, and can potentially be used to complement standard analysis of observational data, upon further development.

29 pages, 12 figures

References in corpus (17)

• The Effective Field Theory of Inflation
• Observational Signatures and Non-Gaussianities of General Single Field Inflation
• Why your model parameter confidences might be too optimistic -- unbiased estimation of the inverse covariance matrix
• The imprints of primordial non-gaussianities on large-scale structure: scale dependent bias and abundance of virialized objects
• The Effective Field Theory of Cosmological Large Scale Structures
• Improving Cosmological Distance Measurements by Reconstruction of the Baryon Acoustic Peak
• The Local Dark Matter Density
• Primordial perturbations and non-Gaussianities in DBI and general multi-field inflation
• Primordial fluctuations and non-Gaussianities in multi-field DBI inflation
• Line-Intensity Mapping: 2017 Status Report
• Constraints on primordial non-Gaussianity from 800,000 photometric quasars
• On the assumption of Gaussianity for cosmological two-point statistics and parameter dependent covariance matrices
• Effective cosmic density field reconstruction with convolutional neural network
• Improving constraints on primordial non-Gaussianity using neural network based reconstruction
• Bayesian Simulation-based Inference for Cosmological Initial Conditions
• Predicting the Initial Conditions of the Universe using a Deterministic Neural Network
• Probing primordial non-Gaussianity by reconstructing the initial conditions

Cited by in corpus (2)

• Learning the Universe: Learning to Optimize Cosmic Initial Conditions with Non-Differentiable Structure Formation Models
• Xenon Signal Denoising via Supervised, Semi-Supervised, and Unsupervised Models