Clustering Confuses Spectro-photometry: An Investigation of 2D and 3D Forced Profile Matching for Stacking Line-intensity Mapping Data on Source Catalogues

TL;DR

This study addresses improving stacking analyses in line-intensity mapping (LIM) by introducing forced photometry through 2D and 3D point-response-function (PRF) models and sub-voxel centering. Using toy and realistic COMAP Pathfinder simulations, the authors show that matched PRF stacks yield $L'_{ m CO}$ recoveries with up to 14–25% higher SNR than the traditional stack, particularly when the angular profile and spectral tails reflect the underlying redshift-space clustering. The results reveal that the optimal stacking profile is broader than the instrument beam and exhibits spectral tails consistent with Fingers-of-God and other redshift-space effects, highlighting the importance of modeling clustering in LIM analyses. Collectively, the work demonstrates that PRF-based matched-filter stacking can enhance sensitivity and provides guidance for applying clustering-aware profiles to future LIM datasets and two-point function analyses.

Abstract

Line-intensity mapping (LIM) is an emerging observational technique that is used to observe the universe on large scales at low resolution through spectral line emission. Stacking analyses coadd cutouts of LIM data on positions of known signal emitters, robustly detecting signal otherwise hidden in a noisy map. In this article, we present two augmentations of a stacking pipeline, both aiming to refine the sensitivity of the stack by assuming a specific observed signal shape in 2D spatial axes or 3D spatial and spectral axes, as well as stacking on source coordinates more precise than the resolution of the LIM data cube. We test these methods on a series of simplistic and complex simulations mimicking observations with the CO Mapping Array Project (COMAP) Pathfinder. We find that these fitting methods provide up to a 25% advantage in detection significance over the original stack method in realistic COMAP-like simulations. We also find that the optimal fitting profile, given our CO line model and galaxy model, is larger than the 5' width of the COMAP beam and takes on a Lorentzian shape in the spectral dimension. Our findings suggest a nuanced dependence of the optimal profile size and shape on the LIM signal itself, including redshift-space clustering and fingers-of-God effects that depend on the tracer luminosity function and bias.

Figures (8)

• Figure 1: A depiction of the stacking analysis and its three different stacking methods.
• Figure 2: Average $L'_{\rm CO}$ recovered by PRF and original stacks from the toy model simulations of \ref{['sec:sanitycheck']}, with a simulated noise rms of $\sigma=10\,\mu$K.
• Figure 3: Ratio of SNR recovered by PRF stacking methods with different widths to the SNR of the original ($3\times5\times5$) stack across noise levels, using the toy model simulations of \ref{['sec:sanitycheck']} using a grid-like arrangement of sources with non-overlapping Gaussian emission profiles. The curves and shaded areas show median and 68% interval values across 50 simulations of ratios of SNR for the stack method indicated by the line style and panel title. A horizontal line (black, dash-dotted) marks a ratio of 1.
• Figure 4: Average $L'_{\rm CO}$ recovered by PRF and original stacks from the realistic simulations of \ref{['sec:realisticsims']}, with a simulated noise rms of $\sigma=10\,\mu$K.
• Figure 5: Ratio of SNR recovered by PRF stacking methods with different widths to the SNR of the original (3 × 5 × 5) stack across noise levels (similar to \ref{['fig:PlaceholderGrid5']}), using the realistic simulations of \ref{['sec:realisticsims']}. The curves and shaded areas show median and 68% interval values across 100 simulations of ratios of SNR for the stack method indicated by the line style and panel title. A horizontal line (black, dash-dotted) marks a ratio of 1.