Deep Learning Enhances Cosmological Parameter Estimation

A recent paper titled "Deep Learning for Cosmological Parameter Inference from Dark Matter Halo Density Field" by Zhiwei Min and 14 co-authors presents a novel approach to estimating cosmological parameters using deep learning techniques. The authors developed a lightweight deep convolutional neural network (lCNN) to analyze simulated three-dimensional dark matter (DM) halo distributions.

The study utilized a training dataset comprising 2000 realizations of a cubic simulation box, each measuring 1000 $h^{-1}{\rm Mpc}$, and interpolated over a cubic grid of $300^3$ voxels. Each simulation was generated using $512^3$ DM particles and $512^3$ neutrinos. The research focused on six standard cosmological parameters under the flat $\Lambda$CDM model, including matter density ($\Omega_m$), baryon density ($\Omega_b$), Hubble parameter ($h$), scalar spectral index ($n_s$), amplitude of the matter power spectrum ($\sigma_8$), and the equation of state parameter for dark energy ($w$), along with the sum of neutrino masses ($M_\nu$).

Key findings from the study include:

• The lCNN model was found to be more efficient in extracting large-scale structure information from the halo density field compared to traditional statistical methods, such as the power spectrum and two-point correlation function.
• Combining the halo density field with its Fourier transformed version improved prediction accuracy.
• The model achieved high accuracy in inferring parameters like $\Omega_m$, $h$, $n_s$, and $\sigma_8$, while it was less effective for $\Omega_b$, $M_\nu$, and $w$.
• Compared to a simple random forest network, the lCNN provided unbiased estimations with reduced statistical errors, achieving approximately 33.3% reduction for $\Omega_m$ and 20.0% for $h$.

This research emphasizes the potential of deep learning techniques in cosmology, particularly in enhancing the accuracy of cosmological parameter estimation, which could have significant implications for our understanding of the universe's structure and evolution. The findings are detailed in the paper available on arXiv: arXiv:2404.09483.