A statistical framework for recovering intensity mapping autocorrelations from crosscorrelations

TL;DR

This paper introduces a statistical framework that leverages crosscorrelations between different intensity mapping datasets to accurately recover the autocorrelation power spectrum, addressing challenges posed by foregrounds and systematics.

Contribution

It generalizes previous methods by providing a comprehensive approach to infer autocorrelations from multiple crosscorrelations within a Least Squares Estimator framework.

Findings

Effective recovery of autocorrelation spectra in certain noise regimes

Framework applicable to near-future line intensity mapping experiments

Derivation of alternative estimators and analysis of posterior distributions

Abstract

Intensity mapping experiments will soon have surveyed large swathes of the sky, providing information about the underlying matter distribution of the early universe. The resulting maps can be used to recover statistical information, such as the power spectrum, about the measured spectral lines (for example, HI, [CII], and [OIII]). However precise power spectrum measurements, such as the 21 cm autocorrelation, continue to be challenged by the presence of bright foregrounds and non-trivial systematics. By crosscorrelating different data sets, it may be possible to mitigate the effects of both foreground uncertainty and uncorrelated instrumental systematics. Beyond their own merit, crosscorrelations could also be used to recover autocorrelation information. Such a technique was proposed in Beane et al. (2019) for recovering the 21 cm power spectrum. Generalizing their result, we develop a statistical framework for combining multiple crosscorrelation signals in order to infer information about the corresponding autocorrelations. We do this first within the Least Squares Estimator (LSE) framework, and show how one can derive their estimator, along with several alternative estimators. We also investigate the posterior distribution of recovered autocorrelation and associated model parameters. We find that for certain noise regimes and cosmological signal modeling assumptions this procedure is effective at recovering autospectra from a set of crosscorrelations. Finally, we showcase our framework in the context of several near-future line intensity mapping experiments.