The Atacama Cosmology Telescope: Cross-correlation of kSZ and continuity equation velocity reconstruction with photometric DESI LRGs

TL;DR

This work demonstrates a high-significance two-dimensional tomographic cross-correlation of kSZ-derived velocities from ACT DR6 with continuity-equation velocities inferred from DESI LRGs, achieving an $11\sigma$ combined detection when including the auto spectrum. The analysis jointly models large-scale velocity and galaxy bias, introducing a scale-independent optical-depth velocity bias $b_v^\alpha$ per redshift bin and a transfer-function calibration from simulations to account for reconstruction biases. The main results include a cross-correlation amplitude corresponding to $b_v = 0.339 \pm 0.034$ (11σ with auto), a null result for foreground contamination in the velocity cross-correlation, and constraints on local primordial non-Gaussianity with $f_{\mathrm{NL}}^{\mathrm{loc}} = -180^{+61}_{-86}$ (67% CL), consistent with zero at 95% CL. This framework demonstrates the viability of kSZ tomography for precision cosmology and foreground control, while highlighting calibration needs and potential improvements for tighter $f_{\mathrm{NL}}$ constraints in future work.

Abstract

Over the last year, kinematic Sunyaev--Zel'dovich (kSZ) velocity reconstruction -- the measurement of the large-scale velocity field using the anisotropic statistics of the small-scale kSZ-galaxy overdensity correlation -- has emerged as a statistically significant probe of the large-scale Universe. In this work, we perform a 2-dimensional tomographic reconstruction using ACT DR6 CMB data and DESI legacy luminous red galaxies (LRGs). We measure the cross-correlation of the kSZ-reconstructed velocity $v^{\mathrm{kSZ}}$ with the velocity inferred from the continuity equation applied to the DESI LRGs $v^{\mathrm{cont}}$ at the $\sim 10 σ$ level, detecting the signal with an amplitude with respect to our theory of $b_v = 0.339\pm 0.034$. We fit a scale-dependent galaxy bias model to our measurement in order to constrain local primordial non-Gaussianity $f_{\mathrm{NL}}^{\mathrm{loc}}$, finding {$f_{\mathrm{NL}}^{\mathrm{loc}}=-180^{+61}_{-86}$} at 67\% confidence, with $f_{\mathrm{NL}}^{\mathrm{loc}}$ consistent with zero at 95\% confidence. We also measure an auto spectrum at $2.1σ$ significance which provides a constraint on $b_v$ of $b_v=0.26_{-0.05}^{+0.11}$, which is consistent with the measurement from the cross spectrum. Our combined measurement is $b_v=0.33\pm0.03$, an $11σ$ measurement. We find a good fit of our model to the data in all cases. Finally, we use different ACT frequency combinations to explore foreground contamination, finding no evidence for foreground contamination in our velocity cross correlation. We compare to a similar measurement where $v^{\mathrm{kSZ}}$ is directly cross correlated with the large-scale galaxy field, and find signs of foreground contamination which is contained in the equal-redshift spectra.

Figures (34)

• Figure 1: Masks used in this analysis. The analysis mask used for the CMB dataset is shown on the top left; this is the ACT mask combined with the cluster mask created from the ACT DR5 cluster catalogue. The DESI DR9 mask that we use for the small-scale galaxies is shown on the top right. The product of this with the ACT mask + cluster mask, which we use to perform kSZ velocity reconstruction at $N_{\mathrm{side}}=4096$, is shown on middle left. We downgrade this to an $N_{\mathrm{side}}=64$ mask and apodize with a 2 degree apodization scale to perform analysis on the reconstructed kSZ velocity; this is shown on the middle right. The mask used for the large-scale DR10 DESI galaxies is shown on the bottom left. The overlap of the two large-scale maps is shown on the bottom right to indicate the overlap in sky area between both large-scale measurements, although note that we never use this mask in practice.
• Figure 2: An estimation of the power spectra of the ACT temperature maps we use, on the full ACT region (" ACT mask") as well as the ACT region with clusters masked ("ACT + clusters masked"). We show this both for minimum-variance NILC map ("MV NILC") as well as for the tSZ-deprojected NILC map ("y-deproj NILC"), and the single-frequency maps. These are pseudo-$C_\ell$ estimates, estimated by measuring the raw power spectra of the masked map and dividing by the sky area available. The impact of deprojecting the tSZ is large; the impact of masking the tSZ clusters is much smaller. The variance on the NGC area is noticably lower than on the SGC area. In all cases we have divided by an estimate of the beam. Unless otherwise indicated, all measurements are on the full (NGC+SGC) area. We include an estimate of the kSZ signal from the Websky simulations 2020JCAP...10..012S.
• Figure 3: The distribution of the photometric redshifts $z^{\mathrm{photo}}$ of the DR9 DESI LRGs that we use for the small-scale galaxy sample for kSZ velocity reconstruction. There are 27,253,833 objects in total, over 44.1% of the sky.
• Figure 4: The redshift distributions. The $\frac{dN}{dz^{\mathrm{photo}}}$, which we estimate just by histogramming the $z^{\mathrm{photo}}$ provided for each object, is in a faded line; the Gaussian-estimated $\frac{dN}{dz^{\mathrm{true}}}$ is shown in the solid line. The spectroscopically calibrated $\frac{dN}{dz^{\mathrm{true}}}$, provided by 2023JCAP...11..097Z, is also shown in the dashed lines. We use the spectroscopic estimates in our $N=4$ analysis, and the Gaussian estimates elsewhere. The coherence length for the velocity is $\sim$200 Mpc; we indicate the moving distance on the upper $x$ axis in order to compare to this number. In our 16-bin case, we approach this, but with broad redshift bins.
• Figure 5: The auto power spectra of the binned galaxy overdensity maps for the $N=4,16$ cases. We also show an estimate of the shot noise, which is simply $\frac{1}{\bar{n}}$ with $\bar{n}$ the mean number density, and the measured power spectra with this subtracted, to indicate on what scales the measurements are shot-noise dominated. We have binned the spectra in bins of width $\Delta\ell=40$ order to smooth them. Note that in the 16-bin case we define each group of 4 bins by dividing each of the bins in the 4-bin case into bins with an equal number of galaxies, so that each of these have equal galaxy number density (and hence shot noise estimates).