On the Use of WGANs for Super-Resolution in Dark-Matter Simulations

TL;DR

This paper explores using WGANs to enhance the resolution of dark-matter simulations, successfully reproducing key statistical properties but facing challenges in capturing detailed structural features.

Contribution

It demonstrates the application of WGANs for super-resolution in dark-matter simulations and assesses their effectiveness and limitations.

Findings

WGANs can generate high-resolution data with accurate summary statistics.

The model reproduces the power spectrum and halo mass function closely.

Generated data shows smeared features, indicating limitations in structural detail capture.

Abstract

Super-resolution techniques have the potential to reduce the computational cost of cosmological and astrophysical simulations. This can be achieved by enabling traditional simulation methods to run at lower resolution and then efficiently computing high-resolution data corresponding to the simulated low-resolution data. In this work, we investigate the application of a Wasserstein Generative Adversarial Network (WGAN) model, previously proposed in the literature, to increase the particle resolution of dark-matter-only simulations. We reproduce prior results, showing the WGAN model successfully generates high-resolution data with summary statistics, including the power spectrum and halo mass function, that closely match those of true high-resolution simulations. However, we also identify a limitation of the WGAN model in the form of smeared features in generated high-resolution data, particularly in the shapes of dark-matter halos and filaments. This limitation points to a potential weakness of the proposed WGAN-based super-resolution method in capturing the detailed structure of halos, and underscores the need for further development in applying such models to cosmological data.