A cross-bispectrum estimator for CMB-HI intensity mapping correlations

TL;DR

This work introduces a cross-bispectrum estimator $B^{\bar{\kappa}\delta T_{21}\delta T_{21}}$ to recover HI IM–CMB lensing cross-correlations that are degraded by foreground removal of long radial HI modes. By exploiting the squeezed-limit coupling between small-scale HI power and a large-scale density mode modulated by CMB lensing, the authors demonstrate detectable signals with HIRAX and Advanced ACT, and forecast sub-percent constraints on the growth rate $f$ and the amplitude $\sigma_8$, as well as competitive dark energy and neutrino-mass bounds when combined with Planck priors. The method is shown to be robust to HI foreground removal since it relies on small-scale HI modes, and it provides a complementary cosmological handle that competes with next-generation galaxy surveys. These results highlight the HI–CMB lensing cross-bispectrum as a powerful, largely independent probe of structure growth, geometry, and neutrino physics in flat $\Lambda$CDM and $w_0w_a$CDM models.

Abstract

Intensity mapping of 21cm emission from neutral hydrogen promises to be a powerful probe of large-scale structure in the post-reionisation epoch. However, HI intensity mapping (IM) experiments will suffer the loss of long-wavelength line-of-sight HI modes in the foreground subtraction process. This significantly reduces HI IM cross-correlations with projected large-scale structure tracers, such as CMB secondary anisotropies. Here we propose a cross-bispectrum estimator, $B^{\bar κδT_{21} δT_{21}},$ to recover the cross-correlation of the HI IM field, $δT_{21},$ with the CMB lensing field, $κ,$ constructed by correlating the position-dependent HI power spectrum with the mean overdensity traced by CMB lensing. We study the cross-bispectrum estimator in the squeezed limit and forecast its detectability based on HI IM measurements from HIRAX and CMB lensing measurements from Advanced ACT. We find that $B^{\bar κδT_{21}δT_{21}},$ in combination with the HI IM and CMB lensing auto-spectra, can place sub-percent constraints on the growth rate of fluctuations, $f,$ and the small scale amplitude of fluctuations, $ σ_8.$ The cross-bispectrum, in combination with the auto-spectra and Planck priors, improves dark energy constraints to 0.025 on $w_0$ and 0.11 on $w_a$ for flat models. These results are robust to HI foreground removal because they derive from small-scale HI modes. The HI-CMB lensing cross-bispectrum thus provides a novel way to recover HI correlations with CMB lensing and constrain cosmological parameters at a level that is competitive with next-generation galaxy redshift surveys. As a striking example of this, we find a tight constraint of 27.8 meV (29.0 meV) on the sum of neutrino masses, while varying all redshift and standard cosmological parameters within a flat $Λ$CDM ($w_0w_a$CDM) model.

Figures (4)

• Figure 1: Binned signal-to noise ratio (SNR) in the $k_\perp$-$k_\parallel$ plane for the $z=0.95$ redshift bin. Top panel: SNR for the H i power spectrum measured by HIRAX. Bottom panel: SNR for the cross-bispectrum measured by HIRAX and AdvACT.
• Figure 2: Forecast 1$\sigma$ and 2$\sigma$ constraints on $f$ and $\sigma_8$ (top panel) from HIRAX H i and AdvACT lensing, in the redshift bin centred at $z=0.95,$ for different power spectrum and cross-bispectrum combinations. This is one of four redshift bins spanning the HIRAX redshift range. The bispectrum has a different degeneracy direction to the H i power spectrum in the $f$-$\sigma_8$ plane. The lensing power spectrum further constrains $\sigma_8$. We also show the $\Omega_m$ and $\sigma_8$ (bottom panel) constraints from HIRAX H i and AdvACT lensing for different power spectrum and cross-bispectrum combinations.
• Figure 3: Forecast 1$\sigma$ and 2$\sigma$ constraints on $w_0$ and $w_a$ from HIRAX H i and AdvACT lensing for different power spectrum and cross-bispectrum combinations.
• Figure 4: H i -CMB lensing cross-correlation signal (solid lines) as a function of angular wavenumber for three different values of $k_{\parallel,cut}$ computed in the $z_i=0.95$ redshift bin. The corresponding dashed lines show the signal-to-noise ratio (SNR) for each case. In the inset plot, we show the CMB lensing kernel in Fourier space, which rapidly falls off with increasing $k_\parallel,$ and the nominal value of $k_{\parallel,cut}$ that we use in this letter.