Emulating Recombination with Neural Networks using Universal Differential Equations

TL;DR

This paper introduces a neural network-based emulator for cosmic recombination history using universal differential equations, enabling fast, flexible, and model-agnostic calculations crucial for analyzing precision cosmological data.

Contribution

The authors develop a differentiable neural network model employing universal differential equations to automatically emulate recombination physics without manual tuning.

Findings

Successfully models recombination history with high accuracy.

Enables rapid computation suitable for cosmological data analysis.

Supports exploration of new physics beyond standard models.

Abstract

With an aim towards modeling cosmologies beyond the $\Lambda$CDM paradigm, we demonstrate the automatic construction of recombination history emulators while enforcing a prior of causal dynamics. These methods are particularly useful in the current era of precision cosmology, where extremely constraining datasets provide insights into a cosmological model dominated by unknown contents. Cosmic Microwave Background (CMB) data in particular provide a clean glimpse into the interaction of dark matter, baryons, and radiation in the early Universe, but interpretation of this data requires knowledge of the Universe's ionization history. The exploration of new physics with new CMB data will require fast and flexible calculation of this ionization history. We develop a differentiable machine learning model for recombination physics using a neural network ordinary differential equation architecture (Universal Differential Equations, UDEs), building towards automatic dimensionality reduction and the avoidance of manual tuning based on cosmological model.