Emulation of Cosmological Mass Maps with Conditional Generative Adversarial Networks

Area of Science:

• Cosmology
• Astrophysics
• Computational Science

Background:

• Weak gravitational lensing mass maps are vital for understanding cosmic structure evolution and constraining cosmological models.
• N-body simulations, while accurate, are computationally expensive, posing a bottleneck for cosmological analyses.
• Existing simulation-based emulators often focus on map statistics, not the maps themselves, limiting their applicability.

Purpose of the Study:

• To develop a novel conditional Generative Adversarial Network (GAN) capable of generating cosmological weak gravitational lensing mass maps.
• To enable the generation of mass maps for a continuous range of cosmological parameters (Ωm, σ8) and source galaxy redshift distributions n(z).
• To overcome the limitations of existing GANs that are restricted to fixed cosmological parameters.

Main Methods:

• Developed a conditional Generative Adversarial Network (GAN) model.
• Trained the GAN on N-body simulation data.
• Generated mass maps for various cosmological parameters (Ωm, σ8) and redshift distributions n(z).
• Quantitatively compared GAN-generated maps against N-body simulations using metrics like power spectra, bispectra, and MS-SSIM.

Main Results:

• The conditional GAN successfully interpolates within the parameter space of simulated cosmologies.
• Generated mass maps exhibit good visual quality and high statistical accuracy across a range of cosmological parameters.
• Quantitative comparisons show typical differences <5% for most metrics, with bispectrum agreement <20%.

Conclusions:

• The developed conditional GAN is an effective tool for generating cosmological mass maps, capturing cosmological signals and variability.
• This work represents a significant step towards building direct emulators for mass maps, reducing computational costs.
• The code and data are publicly available, facilitating further research in cosmology and simulation-based inference.