Super-Resolution Emulation of Large Cosmological Fields with a 3D Conditional Diffusion Model

TL;DR

This paper introduces a 3D conditional diffusion model that efficiently enhances low-resolution cosmological simulations to high-resolution, enabling large-scale, detailed cosmological volume generation with reduced computational costs.

Contribution

The authors develop a novel 3D diffusion-based super-resolution emulator capable of upscaling large cosmological volumes through iterative outpainting, significantly reducing computational resources needed for high-resolution simulations.

Findings

Successfully upscaled a cosmological volume 8 times larger than training data.

Achieved high accuracy in generated simulations using various summary statistics.

Demonstrated the model's ability to handle very large volumes with over 9000 outpainting iterations.

Abstract

High-resolution (HR) simulations in cosmology, in particular when including baryons, can take millions of CPU hours. On the other hand, low-resolution (LR) dark matter simulations of the same cosmological volume use minimal computing resources. We develop a denoising diffusion super-resolution emulator for large cosmological simulation volumes. Our approach is based on the image-to-image Palette diffusion model, which we modify to 3 dimensions. Our super-resolution emulator is trained to perform outpainting, and can thus upgrade very large cosmological volumes from LR to HR using an iterative outpainting procedure. As an application, we generate a simulation box with 8 times the volume of the Illustris TNG300 training data, constructed with over 9000 outpainting iterations, and quantify its accuracy using various summary statistics.