betaFeedback

WorksPhysics and Astronomy

Velocity Reconstruction from kSZ with SPT-3G and DESI-LS

Abstract

We reconstruct the large-scale radial velocity field from the kinetic Sunyaev-Zel'dovich effect for the first time with the South Pole Telescope, and measure the galaxy-velocity correlation and the velocity power spectrum. We cross-correlate the SPT-3G D1 maps at 95, 150, and 220 GHz with luminous red galaxies from the DESI Legacy Imaging Surveys over the SPT-3G Main field. For the extended galaxy sample, we detect the galaxy-velocity correlation at signal-to-noise ratio 11.211.2 and the velocity power spectrum at 8.18.1, combining the three frequencies. We detect the reconstructed velocities for the first time at 220 GHz with signal-to-noise ratios of 4.2 and 3.1 from the galaxy-velocity correlation dipole and monopole, and at 3.8 from the velocity power spectrum. We model and estimate foreground contributions to the monopole, detecting both the thermal Sunyaev-Zel'dovich effect and the cosmic infrared background contribution separately. On the around five percent of sky jointly covered by SPT-3G and DESI Legacy Imaging Surveys, reconstructed velocities give new constraints on local-type primordial non-Gaussianity at fNL=−38−61+61f_{\rm NL}=-38^{+61}_{-61}, and adding galaxy clustering gives fNL=−20−46+40f_{\rm NL}=-20^{+40}_{-46}. The galaxy-velocity correlation is consistent between the three frequencies.

Introduction

The large-scale peculiar-velocity field probes the matter fluctuations and traces the growth of cosmic structure, providing a direct test of cosmological models [see e.g. 1–3]. The kinetic Sunyaev–Zel’dovich (kSZ) effect is the Doppler shift that cosmic microwave background (CMB) photons acquire when they scatter off free electrons with a bulk peculiar velocity [4, 5]. The kSZ effect, therefore, probes the line-of-sight component of this velocity field. In recent years, velocity reconstruction from the kSZ effect has become a competitive probe of large-scale structure [6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16].

The kSZ effect has been measured through several complementary techniques. Pairwise estimators stack cluster pairs to detect their mean relative radial motion, first in Atacama Cosmology Telescope (ACT) and Sloan Digital Sky Survey (SDSS) [17], and subsequently in South Pole Telescope (SPT) and Dark Energy Survey (DES) [18], ACT [19], SPT-3G and DES [20], and ACT, Planck and Dark Energy Spectroscopic Instrument (DESI) [21]. Projected-field estimators cross-correlate the squared CMB temperature with a projected galaxy field [22, 23, 24]. The small-scale temperature power spectrum constrains the total kSZ power [25, 26, 27], and the kSZ trispectrum [28] sets an upper limit on the duration of reionization [29]. Velocity-weighted stacks of CMB maps on galaxies trace the gas density profiles around them [30, 31, 32, 33], and Fourier-space estimators have measured the galaxy–electron cross-power spectrum PgeP_{ge} [34, 35, 36].

This work uses kSZ velocity reconstruction, which recovers the three-dimensional radial peculiar-velocity field on large scales from the correlation between the small-scale CMB temperature and the small-scale galaxy density [6, 7, 37, 38, 39]. We cross-correlate the reconstructed velocity field with the galaxy density field, measuring a dipole, two galaxy modes and the CMB temperature (Smith et al. [6]), which PgeP_{ge} estimators probe on small scales [see e.g. 36] and velocity reconstruction on large scales.

Beyond detecting the kSZ signal, velocity reconstruction also helps us probe the initial conditions of the Universe through primordial non-Gaussianity (PNG). Local form of PNG induces a scale-dependent bias in the galaxy density field [40]. The galaxy–velocity cross-power PgvP_{gv} is sensitive to fNLf_{\mathrm{NL}} through this bias, which is largest on the largest scales. On these scales the reconstructed velocity is a second tracer of the matter density and, through sample-variance cancellation [41], can sharpen the galaxy-clustering constraint on fNLf_{\mathrm{NL}} [7, 40]. The tightest fNLf_{\mathrm{NL}} constraints come from CMB bispectra (e.g. Planck’s fNL=−0.9±5.1f_{\mathrm{NL}} = -0.9 \pm5.1 [42]) and the large-scale structure offers a complementary, low-redshift approach: the strongest bounds come from the scale-dependent clustering of spectroscopic DESI samples, with fNL=−3.6−9.1+9.0f_{\mathrm{NL}} = -3.6^{+9.0}_{-9.1} from the combined DESI DR1 luminous-red-galaxy (LRG) and quasar power spectra [43]. The current bounds on fNLf_{\mathrm{NL}} from velocity reconstruction are at ±34\pm34 (Chaussidon et al. [43]), where this gap is expected to be closed in the near future with Simons Observatory. The kSZ velocity reconstruction program has experienced a recent boom in the past two years. Bloch and Johnson [44] and Krywonos et al. [8] performed a light-cone reconstruction from Planck and unWISE data, placed an upper limit on the amplitude of the reconstructed signal, and constrained −220≲fNL≲136-220 \lesssim f_{\mathrm{NL}} \lesssim136 at 68% confidence. McCarthy et al. [10] first achieved a statistically significant velocity reconstruction from the ACT DR6 CMB maps and photometric DESI LRGs, detecting the signal at a signal-to-noise ratio SNR = 3.8 by cross-correlating with an external estimate of the velocity field from the Baryon Oscillation Spectroscopic Survey (BOSS). Laguë et al. [9] measured PgvP_{gv} at SNR = 7.2 with ACT DR5 and spectroscopic SDSS. Hotinli et al. [11] presented the first three-dimensional reconstruction of the large-scale velocity field using a photometric galaxy survey, from ACT DR5 and DESI Legacy Imaging Surveys (DESI-LS) LRGs reaching detection SNR = 11.7 with the extended sample and the day+night maps. Lai et al. [12] detected the signal at SNR ∼10\sim10 with ACT DR6 ×\times DESI-LS DR9 and a quadratic-maximum-likelihood power spectrum estimator. McCarthy et al. [13] cross-correlated a kSZ velocity reconstruction, from ACT DR6 and photometric DESI LRGs, with the velocity inferred from the continuity equation applied to the same galaxies, at SNR ∼10\sim10 (∼11\sim11 combined with the velocity auto-correlation). Embil Villagra et al. [15] modelled the thermal Sunyaev–Zel’dovich (tSZ) and cosmic infrared background (CIB) biases to the projected galaxy–velocity cross-correlation and introduced a parity-odd estimator, built from antisymmetric combinations of tomographic galaxy–velocity correlations, that cancels these foregrounds in the Limber approximation while preserving the signal. Most recently, Chaussidon et al. [16] combined DESI DR2 with ACT DR6, reporting a total SNR of 20.8 across LRG, emission-line galaxy and quasar samples (17.0, 8.3 and 6.8 per tracer), which is the highest significance PgvP_{gv} detection to date, and the first detection of the velocity auto-spectrum PvvP_{vv}, at SNR = 3.1.

In this work, we use the SPT-3G D1 maps and DESI-LS to measure the kSZ effect. Previously SPT and DES data have been combined in pairwise kSZ [18, 20] measurements, in cross correlations of CMB lensing and galaxy clustering [45, 46, 47, 48], cluster cosmology with weak-lensing mass calibration [49, 50], and CMB-cluster lensing [51, 52] measurements; but not yet in kSZ velocity reconstruction. Hotinli [36] measured PgeP_{ge} with the SPT-3G D1 maps and DESI-LS LRGs, weighting the galaxies by radial velocities reconstructed from the galaxy density rather than from the CMB.

We measure PgvP_{gv} and PvvP_{vv} with SPT-3G and DESI-LS, using the SPT-3G D1 maps at 95, 150, and 220 GHz [53] and the main and extended DESI-LS LRG samples of Ref. [54]. Using the main sample, we detect the Pgvℓ=1P_{gv}^{\ell=1} dipole at SNR 10.4 over the three frequencies, the velocity auto-spectrum at 7.2. The SNR values for the extended sample are 11.2 and 8.1, respectively. We measure the foreground-induced monopole at 16.8 (18.9) for the main (extended) sample. The velocity spectra give fNL=−38−61+61f_{\mathrm{NL}}=-38^{+61}_{-61}, and adding the galaxy auto-spectrum gives fNL=−20−46+40f_{\mathrm{NL}}=-20^{+40}_{-46}, both consistent with zero, as the previous ACT and DESI measurements [11, 16] that used a similar pipeline. The frequency differences of the dipole are consistent with zero, and the extended sample gives results that agree with the main sample.

The remainder of this paper is organized as follows. In §III, we describe the pipeline that produces the PgvP_{gv} dipole and monopole and the velocity auto-spectrum PvvP_{vv}. In §IV, we describe our data: the SPT-3G D1 maps and the DESI-LS LRG catalogs. In §V, we present the results, including the galaxy auto-spectrum, the galaxy-velocity correlations, and velocity spectra, fNLf_{\mathrm{NL}}, as well as the consistency tests; and in §VI, we conclude. Our pipeline, methods, and analysis follow the Refs. [11, 16] closely.

Preliminaries

Throughout this paper, x\boldsymbol{x} is a three-dimensional comoving position, the unit three-vector θ\boldsymbol{\theta} is a direction on the sky, and x^\hat{\boldsymbol{x}} is the unit three-vector along the line of sight, pointing away from the observer. We write the comoving distance to redshift zz as χ(z)\chi(z) and switch between χ\chi and zz as the radial variable where convenient. We use the following Fourier convention:

f(x)=∫d3k(2π)3 f~(k)eik⋅x,f~(k)=∫d3x f(x)e−ik⋅x,(1)f(\boldsymbol{x}) = \int\frac{\mathrm{d}^{3}\boldsymbol{k}}{(2\pi)^{3}}\,\tilde{f}(\boldsymbol{k})e^{i\boldsymbol{k}\cdot\boldsymbol{x}}, \qquad\tilde{f}(\boldsymbol{k}) = \int\mathrm{d}^{3}\boldsymbol{x}\,f(\boldsymbol{x})e^{-i\boldsymbol{k}\cdot\boldsymbol{x}}, \tag*{(1)}

where k\boldsymbol{k} is the three-dimensional Fourier wavevector, and we define the wavenumber as k=∣k∣k = \lvert\boldsymbol{k}\rvert in what follows. We write the kSZ contribution to the CMB temperature as a line-of-sight integral [4]:

TkSZ(θ)=∫dχ K(χ) vr(χθ) δe(χθ),(2)T_{\mathrm{kSZ}}(\boldsymbol{\theta}) = \int\mathrm{d}\chi\,K(\chi)\,v_{r}(\chi\boldsymbol{\theta})\,\delta_{e}(\chi\boldsymbol{\theta}), \tag*{(2)}

with δe(χθ)\delta_{e}(\chi\boldsymbol{\theta}) and vr(χθ)v_{r}(\chi\boldsymbol{\theta}) the electron overdensity and the radial velocity at comoving distance χ\chi in the direction θ\boldsymbol{\theta}.

The radial weight K(z)K(z) is given by:

K(z)≡−TCMBσTne,0xe(z)e−τ(z)(1+z)2(3)K(z) \equiv-T_{\mathrm{CMB}} \sigma_{\mathrm{T}} n_{\mathrm{e},0} x_{\mathrm{e}}(z) e^{-\tau(z)} (1+z)^2 \tag*{(3)}

where TCMBT_{\mathrm{CMB}} and ne,0n_{\mathrm{e},0} are the present-day mean CMB temperature and mean electron density, σT\sigma_{\mathrm{T}} is the Thomson cross-section, and xe(z)x_{\mathrm{e}}(z) and τ(z)\tau(z) are the ionization fraction and the optical depth to redshift zz. Throughout this work, we take a fixed redshift at the mean of our galaxy sample z∗=0.754z_* = 0.754, and set xe=1x_{\mathrm{e}} = 1 and e−τ=1e^{-\tau} = 1.

In the presence of primordial non-Gaussianity (PNG), the galaxy bias bgb_g gains an extra term on large scales [40, 55]:

b(k)=bg+2δc(bg−1)fNLα(k,z),α(k,z)=δm(k,z)Φ(k,z)=2k2T(k)D(z)3Ωm0H02(4)b(k) = b_g + 2\delta_c(b_g - 1)\frac{f_{\mathrm{NL}}}{\alpha(k,z)}, \qquad\alpha(k,z) = \frac{\delta_m(k,z)}{\Phi(k,z)} = \frac{2k^2T(k)D(z)}{3\Omega_{m0}H_0^2} \tag*{(4)}

where T(k)T(k) and D(z)D(z) are the matter transfer function and the linear growth factor, Ωm0\Omega_{m0} and H0H_0 the present-day matter density and the Hubble constant, and we adopt the spherical-collapse threshold δc=1.68\delta_c = 1.68. We evaluate α(k,z)\alpha(k,z) directly from the linear matter power spectrum. Since α(k,z)∝k2\alpha(k,z) \propto k^2 as k→0k \to0, the PNG correction is largest on the largest scales. The coefficient 2δc(bg−1)2\delta_c(b_g - 1) assumes a universal mass function and is taken to be equal to the so-called ‘local PNG bias’ bΦb_\Phi. Departures from this relation can shift the inferred fNLf_{\mathrm{NL}} [56, 57]. Recent works has shown, however, that bΦb_\Phi can be measured from how the number of galaxies changes with redshift [58, 59]. In what follows we will assume bΦb_\Phi fixed, and measure the fNLf_{\mathrm{NL}} from galaxy and galaxy-velocity correlations.

Pipeline

We begin in this section with a description of our velocity-reconstruction framework. The majority of our analysis pipeline was first introduced in Hotinli et al. [11], followed by improvements made in Chaussidon et al. [16].

kSZ velocity reconstruction estimator

The kSZ quadratic estimator reconstructs the large-scale radial-velocity field (up to a multiplicative bias) from a CMB temperature map and a galaxy catalogue. The output of this reconstruction is a 3-d velocity field v^r(x)\hat{v}_r(\boldsymbol{x}) on large scales, whereas the inputs are the 3-d galaxy field δg(x)\delta_g(\boldsymbol{x}) and the 2-d CMB T(θ)T(\boldsymbol{\theta}) on small scales.

The small-scale galaxy field is memory and compute intensive at high-resolution (at wavenumbers kS∼1 Mpc−1k_S \sim1\ \mathrm{Mpc}^{-1}), so we define our estimator as a weighted catalog:

v^r(x)=∑i∈galWivT~(θi)δ3(x−xi)(5)\hat{v}_r(\boldsymbol{x}) = \sum_{i\in\mathrm{gal}} W_i^v \widetilde{T}(\boldsymbol{\theta}_i)\delta^3(\boldsymbol{x}-\boldsymbol{x}_i) \tag*{(5)}

where θi\boldsymbol{\theta}_i is the angular location of the ii-th galaxy, WivW_i^v is a survey-specific per-galaxy weight, and T~(θ)\widetilde{T}(\boldsymbol{\theta}) is a high-pass filtered CMB map which we construct in harmonic space as:

T~ℓm=Fℓν∫d2θ WCMB(θ)Tν(θ)Yℓm∗(θ)(6)\widetilde{T}_{\ell m} = F_\ell^\nu\int d^2\boldsymbol{\theta}\, W_{\mathrm{CMB}}(\boldsymbol{\theta}) T^\nu(\boldsymbol{\theta}) Y_{\ell m}^*(\boldsymbol{\theta}) \tag*{(6)}

where TνT^\nu is the temperature map at frequency ν\nu and WCMBW_{\mathrm{CMB}} is the CMB pixel weight. Here, both WivW_i^v and T~(θi)\widetilde{T}(\boldsymbol{\theta}_i) are weights that act on our galaxy catalogs. FℓνF_\ell^\nu is the ℓ\ell-weighting of the quadratic estimator.

For our purposes, the optimal choice for FℓνF_\ell^\nu down-weights CMB multipoles by the variance of the maps we input; the total galaxy power spectrum (including Poisson noise) PggtotP_{gg}^{\mathrm{tot}} and the total CMB power spectrum (including foregrounds and noise) Cℓtot,νC_\ell^{\mathrm{tot},\nu}; and up-weights them by galaxy-electron correlation PgeP_{ge}. Here, we define the ℓ\ell-weighting as:

Fℓν=Pgefid(ℓ/χ∗)Pggfid(ℓ/χ∗)bℓνCℓtot,ν(7)F_\ell^\nu= \frac{P_{ge}^{\mathrm{fid}}(\ell/\chi_*)}{P_{gg}^{\mathrm{fid}}(\ell/\chi_*)}\frac{b_\ell^\nu}{C_\ell^{\mathrm{tot},\nu}} \tag*{(7)}

where PgefidP_{ge}^{\mathrm{fid}} and PggfidP_{gg}^{\mathrm{fid}} are our fiducial choices for the galaxy power spectrum and galaxy-electron correlation, which we calculate from the halo-model at k=ℓ/χ∗k = \ell/\chi_*, with χ∗=χ(z∗)\chi_* = \chi(z_*), bℓνb_\ell^\nu is the CMB survey beam, and total observed CMB temperature power spectrum Cℓtot,νC_\ell^{\mathrm{tot},\nu} is fitted from the data.

The CMB contains other signals, such as those from the thermal Sunyaev-Zel’dovich (tSZ) effect [60] and the cosmic infrared background (CIB) [61], that are correlated with galaxies. These are foregrounds for our analysis and contribute to T~(θi)\widetilde{T}(\boldsymbol{\theta}_i). For the remainder of this section, we omit these contributions and return to the description of our modelling and mitigation of these in §III C.

The reconstructed velocity in Eq. (5) is biased relative to the true radial-velocity field [6]. The mean of the reconstructed velocity satisfies (see Hotinli et al. [11] for the derivation):

⟨v^r(x)⟩=bvB(x)nv(x)vrtrue(x),(8)\left\langle\hat{v}_{r}(\boldsymbol{x}) \right\rangle= b_{v} B(\boldsymbol{x}) n_{v}(\boldsymbol{x}) v_{r}^{\mathrm{true}}(\boldsymbol{x}), \tag*{(8)}

where nv(x)=⟨∑iWivδ3(x−xi)⟩n_{v}(\boldsymbol{x}) = \left\langle\sum_{i} W_{i}^{v} \delta^{3}(\boldsymbol{x}-\boldsymbol{x}_{i}) \right\rangle is the weighted 3-d galaxy number density, bvb_{v} is the ‘kSZ velocity bias’, well approximated as scale-independent on large scales (k≲0.04k \lesssim0.04) that we consider in this work, as shown in NN-body simulations in Ref. [62]. The function B(x)B(\boldsymbol{x}) is defined by:

B(x)≡WCMB(θ)B0(χ),whereB0(χ)≡K(χ)χ2∫d2ℓ(2π)2bℓvFℓvPgefid(ℓ/χ).(9)B(\boldsymbol{x}) \equiv W_{\mathrm{CMB}}(\theta) B_{0}(\chi), \qquad\text{where} \qquad B_{0}(\chi) \equiv\frac{K(\chi)}{\chi^{2}} \int\frac{d^{2}\ell}{(2\pi)^{2}} b_{\ell}^{v} F_{\ell}^{v} P_{ge}^{\mathrm{fid}}(\ell/\chi). \tag*{(9)}

The kSZ velocity bias is formally defined by:

bv(χ)=∫(d2ℓ/(2π)2)bℓvFℓvPgetrue(ℓ/χ)∫(d2ℓ/(2π)2)bℓvFℓvPgefid(ℓ/χ),(10)b_{v}(\chi) = \frac{\int(d^{2}\ell/(2\pi)^{2}) b_{\ell}^{v} F_{\ell}^{v} P_{ge}^{\mathrm{true}}(\ell/\chi)}{\int(d^{2}\ell/(2\pi)^{2}) b_{\ell}^{v} F_{\ell}^{v} P_{ge}^{\mathrm{fid}}(\ell/\chi)}, \tag*{(10)}

which is an overall prefactor that is roughly the ratio between the true galaxy-electron power spectrum PgetrueP_{ge}^{\mathrm{true}} and the fiducial galaxy-electron power spectrum we defined above. Both B0(χ)B_{0}(\chi) and bv(χ)b_{v}(\chi) vary slowly with redshift, and following Hotinli et al. [11] we fit bvb_{v} as a constant in redshift.

Power spectrum estimators

Once reconstructing radial velocities following §III A, we estimate the power spectra of the radial velocity and its cross-correlation with galaxies. We define multi-polar spin-ℓ\ell power spectra as in the galaxy clustering analyses. In our case, our main observable—the reconstructed radial velocity field—is a spin-1 field. In this section, we describe how we define its spectra, summarizing the formalism developed in Refs. [11, 16], which builds on the dipolar power-spectrum estimators of Refs. [63, 64].

For a field ff, the spin-ℓ\ell Fourier transform pair is

Fℓ[f](k)=∫d3x ϵ∗Lℓ(μ)f(x)e−ik⋅x,Fℓ[f](x)=∫d3k(2π)3ϵLℓ(μ)f(k)eik⋅x,(11)\begin{aligned} \mathcal{F}_{\ell}[f](\boldsymbol{k}) &= \int d^{3}x \, \epsilon^{*} L_{\ell}(\mu) f(\boldsymbol{x}) e^{-i\boldsymbol{k}\cdot\boldsymbol{x}},\\ \mathcal{F}_{\ell}[f](\boldsymbol{x}) &= \int\frac{d^{3}k}{(2\pi)^{3}} \epsilon L_{\ell}(\mu) f(\boldsymbol{k}) e^{i\boldsymbol{k}\cdot\boldsymbol{x}}, \tag*{(11)} \end{aligned}

where μ=k^⋅x^\mu= \hat{\boldsymbol{k}}\cdot\hat{\boldsymbol{x}}, LℓL_{\ell} is the Legendre polynomial of order ℓ\ell, and ϵ=i\epsilon= i if ℓ\ell is odd and ϵ=1\epsilon= 1 if ℓ\ell is even. We use Eq. (11) to define

v^1≡F1[v^r](k)=∫d3x iμv^r(x)e−ik⋅xandP^gvℓ=1(k)=3Ngv∫d3k(2π)3ρg∗(k)v^1(x),(12)\hat{v}_{1} \equiv\mathcal{F}_{1}[\hat{v}_{r}](\boldsymbol{k}) = \int d^{3}x \, i\mu\hat{v}_{r}(\boldsymbol{x}) e^{-i\boldsymbol{k}\cdot\boldsymbol{x}} \qquad\text{and} \qquad \hat{P}_{gv}^{\ell=1}(k) = \frac{3}{N_{gv}} \int\frac{d^{3}k}{(2\pi)^{3}} \rho_{g}^{*}(\boldsymbol{k})\hat{v}_{1}(\boldsymbol{x}), \tag*{(12)}

where NgvN_{gv} is a normalization factor (see Appendix A of Hotinli et al. [11] for our definitions of power spectra normalization we use throughout this paper) and ρg(x)\rho_{g}(\boldsymbol{x}) is the 3-d large-scale galaxy field, which we define through the ‘galaxies-minus-randoms’ prescription:

ρg(x)≡∑i∈galWigδ3(x−xi)−NgNr∑j∈randWjgδ3(x−xj),(13)\rho_{g}(\boldsymbol{x}) \equiv\sum_{i\in\mathrm{gal}} W_{i}^{g}\delta^{3}(\boldsymbol{x}-\boldsymbol{x}_{i}) - \frac{N_{g}}{N_{r}}\sum_{j\in\mathrm{rand}} W_{j}^{g}\delta^{3}(\boldsymbol{x}-\boldsymbol{x}_{j}), \tag*{(13)}

where WigW_{i}^{g} is a per-galaxy weight (such as the FKP or imaging weights), and NgN_{g} and NrN_{r} are the number of galaxies in the data catalog and randoms, respectively. The estimator P^gvℓ=1\hat{P}_{gv}^{\ell=1} of the galaxy-velocity correlation dipole is our main observable in this work, which we will compare to a few-parameter model for Pgv(k)P_{gv}(k):

Pgv(k)=bv(bg+fNL2δc(bg−1)α(k,z))faHkPlin(k)(14)P_{gv}(k) = b_{v}\left(b_{g}+f_{\mathrm{NL}}\frac{2\delta_{c}(b_{g}-1)}{\alpha(k,z)}\right)\frac{faH}{k}P_{\mathrm{lin}}(k) \tag*{(14)}

where Plin(k)P_{\mathrm{lin}}(k) is the linear matter power spectrum, ff is the linear growth rate, aa is the scale factor and HH is the Hubble expansion rate.

In addition to the galaxy-velocity correlation, we also define power-spectrum estimators for velocity-velocity auto-correlation,

P^vv(k)=9Nvv∫d3k(2π)3F1[v^r](k)F1[v^r](−k),(15)\widehat{P}_{vv}(k)=\frac{9}{N_{vv}}\int\frac{\mathrm{d}^{3}k}{(2\pi)^{3}}F_{1}[\widehat{v}_{r}](k)F_{1}[\widehat{v}_{r}](-k), \tag*{(15)}

and for reasons that will become clear in §III C, the galaxy-velocity cross-correlation monopole by:

P^gvℓ=0(k)=1Ngv∫d3k(2π)3F0[ρg](k)F0[v^r](−k).(16)\widehat{P}^{\ell=0}_{gv}(k)=\frac{1}{N_{gv}}\int\frac{\mathrm{d}^{3}k}{(2\pi)^{3}}F_{0}[\rho_{g}](k)F_{0}[\widehat{v}_{r}](-k). \tag*{(16)}

We define the galaxy power spectrum as:

P^gg(k)≡1Ngg∫d3k(2π)3ρg∗(k)ρg(k).(17)\widehat{P}_{gg}(k)\equiv\frac{1}{N_{gg}}\int\frac{\mathrm{d}^{3}k}{(2\pi)^{3}}\rho_{g}^{*}(k)\rho_{g}(k). \tag*{(17)}

A standard method in cosmology for comparing these spectra to their theoretical predictions is to compute and de-convolve the effect of survey geometry. In this paper, we use a different approach: we convolve the model for these spectra with the survey geometry and compute the response of the estimators to the model in a way that accounts for the survey geometry.

Foreground contributions

So far in constructing the estimator of §III A we treated the filtered CMB temperature T~\widetilde{T} of Eq. (6) as containing only the kSZ signal. In practice, other signals (foregrounds, for our purposes) in the CMB that are correlated with the galaxy survey will also contribute to our estimator. Here we describe how our pipeline accounts for these contributions.

First, we have found in Hotinli et al. [11] that we have to modify our estimator in order to mitigate correlated CMB foregrounds that on average make an additive contribution to the value of T~\widetilde{T} in a way that is slowly varying in redshift and independent of the radial velocity. In order to mitigate this systematic contribution, we divide our galaxies into redshift bins bb and subtract the mean weighted temperature T~b\widetilde{T}_{b} within each in. We define a ‘mean-subtracted’ velocity reconstruction v^r(x)\widehat{v}_{r}(x) by:

v^r(x)=∑b∑i∈bWiv(T~(θi)−T~b)δ3(x−xi).(18)\widehat{v}_{r}(x)=\sum_{b}\sum_{i\in b}W_{i}^{v}\left(\widetilde{T}(\theta_{i})-\widetilde{T}_{b}\right)\delta^{3}(x-x_{i}). \tag*{(18)}

This subtraction projects out the modes of v^r\widehat{v}_{r} that vary slowly in redshift (including those that are constant across the sky, the radial analog of the integral constraint in galaxy clustering [65]). In fact, in our analysis, we apply this mean subtraction to the galaxy fields as well. Sampling the redshift distribution of randoms from data is generally understood to alter the clustering measurements [65], and our treatment is equivalent to including a radial integral constraint that accounts for this effect. In what follows, we will use Nz=25N_{z}=25 redshift bins and apply the same mean subtraction to both our data and the surrogate fields of §III E, for both v^r\widehat{v}_{r} and galaxy fields.

We now model the remaining foreground contributions. At the small angular scales (ℓ>2000\ell>2000) the CMB contains a series of signals, including the primary CMB, the tSZ effect, the CIB:

Tobsν=Tprim+TkSZ+TtSZ+TCIBν+⋯ ,(19)T_{\mathrm{obs}}^{\nu}=T_{\mathrm{prim}}+T_{\mathrm{kSZ}}+T_{\mathrm{tSZ}}+T_{\mathrm{CIB}}^{\nu}+\cdots, \tag*{(19)}

where the tSZ, CIB terms depend on frequency. We approximate these contributions as biased tracers of the matter density field, bνfgδmb_{\nu}^{\mathrm{fg}}\delta_{m}, where bνfgb_{\nu}^{\mathrm{fg}} is the ‘foreground bias’ at frequency ν\nu. At each frequency, the mean of the velocity-reconstruction estimator of Eq. (5) is therefore:

⟨v^rν(x)⟩=bνB(x)nν(x)vrtrue(x)⏟kSZ+bνfgWCMB(θ)nν(x)δm(x)⏟foregrounds.(20)\left\langle\widehat{v}_{r}^{\nu}(x)\right\rangle=\underbrace{b_{\nu}B(x)n_{\nu}(x)v_{r}^{\mathrm{true}}(x)}_{\mathrm{kSZ}}+\underbrace{b_{\nu}^{\mathrm{fg}}W_{\mathrm{CMB}}(\theta)n_{\nu}(x)\delta_{m}(x)}_{\mathrm{foregrounds}}. \tag*{(20)}

In what follows, we focus on the tSZ and CIB contributions, the other contributions are sub-dominant compared to these on the scales we use in our analysis.2

The two terms in Eq. (20) depend differently on μ\mu, the cosine of the angle between the wavevector and the line of sight we defined in §III B, so they enter different multipoles of the galaxy-velocity power spectrum. The density term enters the even multipoles, mostly the monopole ℓ=0\ell= 0. Since vrtrue∝μδmv^{\mathrm{true}}_r \propto\mu\delta_m, the velocity term enters the odd multipoles, mostly the dipole ℓ=1\ell= 1. In an idealized geometry, the two contributions separate cleanly, and the foreground does not bias the kSZ signal. The survey selection function (footprint, angular mask, completeness, and radial selection), however, introduces the usual window-function convolution that mixes multipoles, so that foreground power leaks from the monopole into the dipole and can contaminate the velocity signal. We will model this contamination with the surrogate methodology of §III E.

We model the foreground bias as:

bfgν=AtSZftSZ(ν)+ACIBfCIB(ν),(21)b_{\mathrm{fg}}^\nu= A_{\mathrm{tSZ}} f_{\mathrm{tSZ}}(\nu) + A_{\mathrm{CIB}} f_{\mathrm{CIB}}(\nu), \tag*{(21)}

which is a more interpretable parameterization, rather than defining independent values for bfgνb_{\mathrm{fg}}^\nu at each CMB map frequency. We model ftSZ(ν)f_{\mathrm{tSZ}}(\nu) assuming the analytic non-relativistic tSZ frequency spectrum of [60], and we model fCIB(ν)f_{\mathrm{CIB}}(\nu) by a modified blackbody, for which we take a dust temperature Td=24T_{\mathrm{d}} = 24 K and an emissivity index β=1.6\beta= 1.6. We normalize these functions so that ftSZ(150)=fCIB(150)=1f_{\mathrm{tSZ}}(150) = f_{\mathrm{CIB}}(150) = 1. The amplitudes AtSZA_{\mathrm{tSZ}} and ACIBA_{\mathrm{CIB}} have a physical convention that is described in §III E. Throughout this work, we use the galaxy-velocity monopole Pgvℓ=0P_{gv}^{\ell=0} to fit these two amplitudes and to model their leakage into galaxy-velocity dipole and velocity-velocity correlations.

Galaxies and velocities in redshift space

We measure galaxies and velocities in redshift space. At linear and mildly non-linear scales, the galaxy density in redshift space is related to linear matter density δm\delta_m in real space by taking into account the Kaiser and Fingers-of-God effects. Following Chaussidon et al. [16], we define:

δgs(k)=(bg+fNL2δc(bg−1)α(k,z)+fμ2)Dg(k,μ,Σs)δm(k)(22)\delta_g^s(\boldsymbol{k}) = \left(b_g + f_{\mathrm{NL}} \frac{2\delta_c(b_g - 1)}{\alpha(k,z)} + f\mu^2\right) D_g(k,\mu,\Sigma_s)\delta_m(\boldsymbol{k}) \tag*{(22)}

where Dg(k,μ,Σs)=1/(1+(kμΣs)2/2)D_g(k,\mu,\Sigma_s) = 1/(1 + (k\mu\Sigma_s)^2/2) is a damping term that accounts for the Fingers-of-God effect, parametrized by the damping parameter Σs\Sigma_s. This parametrization absorbs the nonlinearities, extending the validity of our modeling to mildly nonlinear conditions [16], as well as capturing observational effects due to redshift errors, for example.

On large scales, the radial velocity field is directly related to the matter density through the continuity relation, vr(k,μ)=(ifaHμ/k)δm(k)v_r(k,\mu) = (if aH\mu/k)\delta_m(k), and it thus provides an unbiased probe of large scale density fluctuations. In what follows, we approximate the radial velocity field in redshift space with a damping term similar to that of the galaxy, as:

vrs(k)=ifaHμkDv(k,Σv)δm(k),(23)v_r^s(\boldsymbol{k}) = \frac{if aH\mu}{k}D_v(k,\Sigma_v)\delta_m(\boldsymbol{k}), \tag*{(23)}

where Dv(k,Σv)=sinc⁡(kΣv)D_v(k,\Sigma_v) = \operatorname{sinc}(k\Sigma_v), parameterized by Σv\Sigma_v. We will use these definitions when modeling the observed spectra using our surrogate fields framework.

Surrogate fields

In this section, we describe our ‘surrogate fields’ framework. We use surrogate fields to model the galaxy and velocity auto- and cross-spectra. A surrogate field is a Monte-Carlo representation of a simplified random field that preserves the exact field-level covariance of a more complicated random field that includes survey geometry, light-cone evolution, photometric uncertainties, and so on. The surrogate fields framework allows us to model and calculate the expectation values for our power spectrum estimators and their covariance accurately on large scales (k≳0.02 Mpc−1k \gtrsim0.02\,\mathrm{Mpc}^{-1}). A more complete introduction of the surrogate fields can be found in Hotinli et al. [11]. Here, we give a brief overview.

We start with a Gaussian field δlin(x)\delta_{\mathrm{lin}}(\boldsymbol{x}) in the bounding box. In order to model the galaxy field, we ‘paint’ the values of δlin(x)\delta_{\mathrm{lin}}(\mathbf{x}) onto randoms as follows:

Sgsig(x)≡NgNr∑j∈randWβδG(x)δ3(x−xj),(24)S_g^{\mathrm{sig}}(\mathbf{x}) \equiv\frac{N_g}{N_r} \sum_{j\in\mathrm{rand}} W_\beta\delta_G(\mathbf{x})\delta^3(\mathbf{x}-\mathbf{x}_j), \tag*{(24)}

where

δG(k)=(bg+fNL2δc(bg−1)α(k,z)+fμ2)Dg(k,μ,Σs)δlin(k).(25)\delta_G(\mathbf{k}) = \left(b_g+f_{\mathrm{NL}}\frac{2\delta_c(b_g-1)}{\alpha(k,z)}+f\mu^2\right)D_g(k,\mu,\Sigma_s)\delta_{\mathrm{lin}}(\mathbf{k}). \tag*{(25)}

We model the Poisson noise term by adding a Gaussian white noise; by defining an independent Gaussian random variable ηj\eta_j for each random jj as:

Sgnoise(x)≡snNgNr∑j∈randWjgηjδ3(x−xj),(26)S_g^{\mathrm{noise}}(\mathbf{x}) \equiv s_n\frac{N_g}{N_r}\sum_{j\in\mathrm{rand}}W_j^g\eta_j\delta^3(\mathbf{x}-\mathbf{x}_j), \tag*{(26)}

where η\eta satisfies:

⟨ηj2⟩=NrNg−⟨δG(zj)2⟩.(27)\langle\eta_j^2\rangle=\frac{N_r}{N_g}-\langle\delta_G(z_j)^2\rangle. \tag*{(27)}

The sum of the two terms gives our surrogate field Sg(x)=Sgsig(x)+Sgnoise(x)S_g(\mathbf{x})=S_g^{\mathrm{sig}}(\mathbf{x})+S_g^{\mathrm{noise}}(\mathbf{x}). In each Monte-Carlo realization, we produce Sg(x)S_g(\mathbf{x}) and store its more optimized reparameterization as three separate fields:

Sg(x)=bgSg(0)(x)+fNLSg(1)(x)+snSg(2)(x),(28)S_g(\mathbf{x})=b_gS_g^{(0)}(\mathbf{x})+f_{\mathrm{NL}}S_g^{(1)}(\mathbf{x})+s_nS_g^{(2)}(\mathbf{x}), \tag*{(28)}

as we fit bgb_g, fNLf_{\mathrm{NL}} and sns_n to data in §V. Modeling the radial velocity field follows a similar manner. We define the velocity surrogate field dipole as:

Sv(x)=Svsig(x)+Svfg(x)+Svnoise(x).(29)S_v(\mathbf{x})=S_v^{\mathrm{sig}}(\mathbf{x})+S_v^{\mathrm{fg}}(\mathbf{x})+S_v^{\mathrm{noise}}(\mathbf{x}). \tag*{(29)}

We model the signal term as:

Svsig≡bvNgNr∑j∈randWjvB(xj)vrsim(xj)δ3(x−xj),(30)S_v^{\mathrm{sig}}\equiv b_v\frac{N_g}{N_r}\sum_{j\in\mathrm{rand}}W_j^vB(\mathbf{x}_j)v_r^{\mathrm{sim}}(\mathbf{x}_j)\delta^3(\mathbf{x}-\mathbf{x}_j), \tag*{(30)}

where

vrsim(k)=ifaHμkDv(k,Σv)δlin(k);(31)v_r^{\mathrm{sim}}(\mathbf{k})=\frac{if aH\mu}{k}D_v(k,\Sigma_v)\delta_{\mathrm{lin}}(\mathbf{k}); \tag*{(31)}

and the foreground term as:

Svfg≡bfgNgNr∑j∈randWjvB(xj)δlin(x)δ3(x−xj).(32)S_v^{\mathrm{fg}}\equiv b_{\mathrm{fg}}\frac{N_g}{N_r}\sum_{j\in\mathrm{rand}}W_j^vB(\mathbf{x}_j)\delta_{\mathrm{lin}}(\mathbf{x})\delta^3(\mathbf{x}-\mathbf{x}_j). \tag*{(32)}

For the noise term, we use a bootstrapping procedure based on real CMB maps.3 In each Monte Carlo realization, we select a subset SS of the random catalog {1,…,Nr}\{1,\ldots,N_r\} of size NgN_g (number of galaxies in the galaxy catalog) and define:

Svnoise(x)≡snv∑j∈randMjWjvT~(θj)δ3(x−xj),whereMj≡{1if 1∈S0if 0∉S(33)S_v^{\mathrm{noise}}(\mathbf{x})\equiv s_{nv}\sum_{j\in\mathrm{rand}}M_jW_j^v\widetilde{T}(\theta_j)\delta^3(\mathbf{x}-\mathbf{x}_j),\qquad\text{where}\quad M_j\equiv \begin{cases} 1 & \text{if }1\in S\\ 0 & \text{if }0\notin S \end{cases} \tag*{(33)}

as we have done in Hotinli et al. [11].

We parameterize SvS_v in a way more optimized for our analysis:

Sv(x)≡bvvSv(0)(x)+bfgvSv(1)(x)+svvSv(2)(x),(34)S_v(\mathbf{x}) \equiv b_v^v S_v^{(0)}(\mathbf{x}) + b_{\mathrm{fg}}^v S_v^{(1)}(\mathbf{x}) + s_v^v S_v^{(2)}(\mathbf{x}), \tag*{(34)}

and store these fields separately for each frequency ν\nu. Note here that because we include B(x)B(\mathbf{x}) — which is a constant value for our analysis choices — in our definition of the surrogate field, our fitted foreground bias bfgfitb_{\mathrm{fg}}^{\mathrm{fit}} relates to the bfgvb_{\mathrm{fg}}^v introduced as bfg≡B×bfgfitb_{\mathrm{fg}} \equiv B \times b_{\mathrm{fg}}^{\mathrm{fit}}. We have maintained this choice made in Chaussidon et al. [16] for consistency. In following works, we will likely modify the definition for SfgvS_{\mathrm{fg}}^v to avoid this redundancy.

In what follows, in order to measure the foreground bias bfgvb_{\mathrm{fg}}^v, we calculate the galaxy-velocity monopole Pgvℓ=0P_{gv}^{\ell=0} for each of our surrogate realizations. The surrogate velocity monopole is defined similarly to the dipole, with Fourier transforms set to spin-0 rather than spin-1 for calculating the galaxy-velocity correlation. This gives us the model prediction for the velocity-galaxy cross-correlation on the left-hand side of Eq. (16) and allows us to measure and distinguish the tSZ and CIB contributions to our estimators.

For each Monte Carlo realization, we store the Fourier transforms of the SgS_g and SvS_v components, which we labeled with upper indices (0,1,2)(0,1,2) above. This also allows us to fit the damping amplitudes Σs\Sigma_s and Σv\Sigma_v, as we add the damping terms DgD_g and DvD_v directly to the multipoles after the window convolution. While we fit independent amplitudes for the biases bvb_v and bfgb_{\mathrm{fg}}, we use a common damping term Σv\Sigma_v for all frequencies, as Σv\Sigma_v should not depend on the choice of the CMB map.4

Our pipeline also stores a term proportional to fNLf_{\mathrm{NL}} for monopole. In this work, we will not use the monopole signal, i.e., Pgvℓ=0(k)P_{gv}^{\ell=0}(k), as a probe of PNG, since this term is not a direct measure of the kSZ signal. It probes the correlations between galaxies and foregrounds, such as tSZ and CIB, which we assume are biased tracers of galaxies; and may be subject to systematics or contributions that are beyond the scope of this paper for us to model. We nevertheless define Pgvℓ=0(k)P_{gv}^{\ell=0}(k) monopole with such a term for consistency and to account for the leakage of foregrounds into velocity-galaxy dipole and velocity-velocity correlation signals.

In what follows, we omit the use of the largest scales in our monopole measurements. We also do not fit fNLf_{\mathrm{NL}} to our monopoles in our analysis in §V, where we indeed see a subtle hint of a 1/k21/k^2–like contribution within 1–2 sigma significance, although it is subdominant compared to foreground contributions we are fitting. We refrain from interpreting this as either systematics or PNG-like.

Lastly, in this work, we analyze a photometric survey, so we must extend the expressions we introduced above to incorporate the photometric redshift errors. We do so by defining for each galaxy the observed (photometric) and true redshifts, zobsz_{\mathrm{obs}} and ztruez_{\mathrm{true}}, respectively, and σz\sigma_z, with which we denote the photo-zz errors. The galaxy catalogs generally contain (zobs,σz)(z_{\mathrm{obs}},\sigma_z) values for each object, but ztruez_{\mathrm{true}} is unknown. Here, we deconvolve the observed (zobs,σz)(z_{\mathrm{obs}},\sigma_z) values from the galaxy catalog in order to infer the underlying triple (zobs,ztrue,σz)(z_{\mathrm{obs}},z_{\mathrm{true}},\sigma_z) distribution. We define the comoving positions of galaxies at observed and true redshifts as xobs=(θ,zobs)\mathbf{x}_{\mathrm{obs}}=(\boldsymbol{\theta},z_{\mathrm{obs}}) and xtrue=(θ,ztrue)\mathbf{x}_{\mathrm{true}}=(\boldsymbol{\theta},z_{\mathrm{true}}), respectively. We then modify the definitions we made in this section by placing the delta functions at observed locations: δ3(x−xi)→δ3(x−xiobs)\delta^3(\mathbf{x}-\mathbf{x}_i) \to\delta^3(\mathbf{x}-\mathbf{x}_i^{\mathrm{obs}}). We place the δG\delta_G and vsimv^{\mathrm{sim}} fields at true locations. We also let the weights WigW_i^g and WivW_i^v be functions of both zobsz_{\mathrm{obs}} and σz\sigma_z. A more detailed description of our framework with photo-zz errors can be found in Hotinli et al. [11].

Data

In this section, we describe the CMB maps and the galaxy catalogs we use in this analysis. One novel aspect of this work is that for CMB, we use the maps and ancillary products of the SPT-3G D1 release [53] (§IV A). This is the first time SPT maps are used for kSZ velocity reconstruction analysis. Overall, we have found these products to be straight-forward to work with, and the maps we use to be of the highest quality for our purposes. For the galaxy catalog, we follow Hotinli et al. [11] and use DESI-LS DR9. We will transition to using the more recent DR11 dataset of the same survey in upcoming works. Figure 1 shows various footprints of related surveys, DESI-LS DR9, DES, and SPT-3G.

SPT-3G D1 maps and ancillary products

In this analysis we use the full-depth coadded temperature maps at 95, 150, and 220 GHz from the SPT-3G D1 release [53, 74], made from two years (2019–2020) of observations of the SPT-3G Main field with the SPT-3G receiver [75] on the South Pole Telescope [76]. The Main field spans declination −42∘-42^\circ to −70∘-70^\circ and right ascension −50∘-50^\circ

Filtered CMB temperature, projected LRG overdensity, and reconstructed velocities over the SPT-3G footprint

Figure 1. Filtered CMB temperature, projected LRG overdensity and reconstructed velocities over the SPT-3G footprint, shown in equatorial coordinates. (a) The filtered 150-GHz temperature evaluated at the main-sample galaxy positions, with the beam equalisation and filtering of Eq. (36). For display, we smooth it with a Gaussian of 0.8∘0.8^\circ full width at half maximum (FWHM), subtract its in-footprint median, and scale it by its standard deviation. (b): The projected LRG overdensity over 0.4≤zobs≤1.10.4 \le z_{\mathrm{obs}} \le1.1 (using our galaxies-minus-randoms prescription) and smoothed with the same Gaussian. (c) Projected 150-GHz radial-velocity estimator of the main sample over same footprint. We show the estimator in the shell 0.65≤zobs<0.750.65 \le z_{\mathrm{obs}} < 0.75, calculated from the filtered temperatures as described in §III A. The pixel are smoothed with a 2.5∘2.5^\circ FWHM Gaussian. We then subtract the median of the estimator and scale the result by its standard deviation inside the footprint. The projected estimator contains reconstruction noise and residual foregrounds, and we assign no significance to individual structures. In all panels, the grey shading shows the Planck GNILC 353-GHz dust intensity [71]. Green shading shows the DESI-LS DR9 imaging footprint [72], traced by its randoms, and the orange curve marks the DES boundary [73]; neither used as a footprint in our analysis.

to 50°. The maps are stored in HEALPix [77] format at Nside=8192N_{\rm side}=8192, and we downgrade them to Nside=4096N_{\rm side}=4096 for processing.

For the analysis mask, we adopt the apodised borders-and-objects mask provided in the SPT-3G D1 release, which apodises both the survey boundary and the disks covering the high-signal-to-noise objects (point sources and clusters) that are masked in the SPT timestream high-pass filter. We build the CMB pixel weight by thresholding the per-frequency inverse-noise weight maps wν(θ)w^\nu(\theta) released with the data, each normalized to its maximum.

WCMB(θ)=Wapo(θ)∏ν1[wν(θ)>0.05],(35)W_{\rm CMB}(\theta)=W_{\rm apo}(\theta)\prod_\nu\mathbb{1}\left[w^\nu(\theta)>0.05\right], \tag*{(35)}

where WapoW_{\rm apo} is the apodised borders-and-objects mask above, 1[⋅]\mathbb{1}[\cdot] is the indicator function and the product runs over the three frequencies. This threshold removes the noisy non-uniform-coverage boundary of the field. We use the resulting WCMBW_{\rm CMB} at all three frequencies (the footprint corresponds to fsky=0.043f_{\rm sky}=0.043, 1780 deg2^2).

For the beams bℓνb_\ell^\nu we defined in §III, we use SPT-3G effective beams that are part of the auxiliary data products as described in §5.2 of Ref. [78] and in Refs. [27, 53]. These are defined as a product of the filter transfer function, a pixel window, and the SPT-3G D1 beam. As in Refs. [16, 70] we equalize the 150 and 220 GHz beams to the 95 GHz beam, which is the broadest. We evaluate Eq. (7) at 95 GHz with Cℓtot,95C_\ell^{\rm tot,95} fitted to the 95 GHz CMB map; and rescale the filter at 150 and 220 GHz by the beam ratio so that all three reconstructions share the same effective ℓ\ell-weighting:

Fℓν=bℓ95bℓνFℓ95,ν∈{150,220}.(36)F_\ell^\nu=\frac{b_\ell^{95}}{b_\ell^\nu}F_\ell^{95},\qquad\nu\in\{150,220\}. \tag*{(36)}

While this choice makes us lose some SNR, it nevertheless simplifies our analysis by making the velocity bias of Eq. (10) the same at all three frequencies and sets the null maps (90 − 150, 90 − 220, and 150 − 220) we describe in §V E free of the kSZ signal. We show the filters used in this analysis in Figure 2.

Figure 2. The CMB filters FℓF_\ell at each frequency, divided by their maximum Fℓmax⁡F_\ell^{\max}, for ACT DR5 at 90 and 150 GHz from Hotinli et al. [11], and SPT-3G D1 at 95, 150 and 220 GHz used here. SPT-3G D1 filters peak at a higher ℓ\ell around ∼5100\sim5100 compared to ACT DR5 around ℓ∼3600\ell\sim3600.

Below in §V E, we use the two half-depth coadds released with the SPT-3G D1 maps, formed by randomly partitioning the 2019–2020 dataset into two equal-depth halves for null tests. The release additionally provides 500 sign-flip noise coadds and 500 sets of mock-observation input and output maps. We do not use these in the present analysis since the surrogate framework provides our covariance estimate.

DESI-LS LRG catalogue

As we stated above, for the galaxy survey, we use the DESI-LS DR9 LRGs [54, 72, 79]. DESI-LS combines data from three optical surveys (DECaLS, BASS, and MzLS) covering approximately 14000 deg2^2 of the extragalactic sky. We use the main and extended LRG samples of [54]. The main sample is the DESI LRG target sample [79]. The extended sample uses looser selection cuts, which give 2–3 times the density and slightly larger photometric-redshift errors. The main sample is nested within the extended sample, with identical per-object photometric redshifts.

Whereas Hotinli et al. [11] restricted its ACT analysis to the North Galactic Cap (NGC), here we work in the Galactic South, where the SPT-3G Main field overlaps with DESI-LS. The southern LRG catalog is built from both the DECam imaging of DECaLS and the photometry of DES [54, 79].

We restrict to the photometric redshift range 0.4≤zobs≤1.10.4\leq z_{\rm obs}\leq1.1, apply the veto masks and quality cuts described in Ref. [80] and §3.3 of Ref. [54], and restrict to sky regions where WCMB(θ)>0W_{\mathrm{CMB}}(\theta)>0. We summarize the properties of our galaxy samples after these cuts in Table I. Throughout, number densities are measured over the approximately 1400 deg2^2 that the galaxies actually occupy, as traced by the surviving randoms, rather than over the full 1780 deg2^2 where WCMB>0W_{\mathrm{CMB}}>0.

PropertyMainExtended
NgN_g690,1351,742,115
nˉg\bar{n}_g [Mpc−3^{-3}]1.39×10−41.39 \times10^{-4}3.50×10−43.50 \times10^{-4}
zmin⁡,zmax⁡,z∗z_{\min}, z_{\max}, z_*0.4, 1.1, 0.7540.4, 1.1, 0.754
effective area [deg2^2]≈1780\approx1780≈1780\approx1780
fskyf_{\mathrm{sky}}0.0430.043
photo-zz scatter σz/(1+z)\sigma_z/(1+z)0.0240.027

Table 1. DESI-LS LRG sample properties in the SPT-3G D1 footprint. The main selection is nested within the extended one. The number densities are measured over the galaxy-occupied area traced by the randoms. The effective area and fskyf_{\mathrm{sky}} are evaluated where WCMB>0W_{\mathrm{CMB}}>0. We list the samples after the photometric-redshift-error cut described in §IV B. The photo-zz entries are the constant errors we adopt from [54].

Our analysis in this papers differs from Hotinli et al. [11] two notable ways. First, we remove the 5% of galaxies with the largest fractional photometric-redshift uncertainty σz/(1+z)\sigma_z/(1+z). This sets a threshold of ∼0.056\sim0.056 for the extended sample and ∼0.055\sim0.055 for the main sample. The removed galaxies have a median σz/(1+z)\sigma_z/(1+z) of 0.071 in both samples, compared to around 0.027 (extended) and 0.024 (main) for the galaxies we keep. Second, we adopt a constant σz\sigma_z value for our galaxies and randoms in our analysis, which we set using the spectroscopically confirmed normalized median absolute deviation of the scatter of LRG photometric redshifts from Ref. [54]. We use σz=0.027(1+z)\sigma_z=0.027(1+z) (extended) and 0.024(1+z)(1+z) (main).

We weight every galaxy equally in the galaxy density and reconstructed velocity (i.e. Wg=WvW_g=W_v), and with our choice of constant photo-zz errors, these weights are the same for every galaxy. As in Ref. [11], we do not include an FKP weight [81]. As described in §III E, we assign each random object a (zobs,ztrue)(z_{\mathrm{obs}},z_{\mathrm{true}}) pair drawn from a deconvolution of the observed redshift distribution of the galaxy sample after with the constant scatter above. We apply the same redshift and quality cuts to the randoms as to the galaxy catalog.

Results: SPT-3G ×\times DESI-LS

We now describe our analysis and results in this section. Our results contain a number of first detections: We detect the velocity-galaxy correlation dipole, monopole, and velocity-velocity autocorrelation for the first time with SPT-3G. We extend the kSZ analysis up to date to 220 GHz CMB maps for the first time, as we find that the SPT-3G maps at this frequency are more than adequate for this analysis, where we detect the galaxy-velocity dipole at high significance for both the main and extended catalogs, and we detect the galaxy-velocity monopole and velocity-auto at a significance larger than 3σ3\sigma for the extended catalog. For the first time, we analyze the galaxy-velocity correlation together with the galaxy-galaxy spectrum, and provide joint constraints that improve fNLf_{\mathrm{NL}} measurement from galaxies alone. We start by introducing our statistical analysis choices in §V A, and describe our results in the rest of this section.

Statistical analysis

Parameter constraints

In §III we introduced our pipeline, where we also defined the parameters we use to model our observables: galaxy-velocity correlation dipole and monopole, as well as velocity-velocity and galaxy-galaxy auto-correlations. These are the galaxy bias bgb_g, three velocity bias parameters (bv95b_v^{95}, bv150b_v^{150}, bv220b_v^{220}) for each CMB frequency, the amplitudes that parametrize the foreground biases (AtSZA_{\mathrm{tSZ}}, ACIBA_{\mathrm{CIB}}), the local PNG amplitude fNLf_{\mathrm{NL}}, a parameter that allows us to fit for the galaxy Poisson noise sns_n, and three additional parameters for the velocity reconstruction noise (snv95s_{nv}^{95}, snv150s_{nv}^{150}, snv220s_{nv}^{220}). We also defined the damping parameters (ΣgΣv\Sigma_g \Sigma_v) for redshift-space distortions.

We infer these parameters by Monte Carlo Markov chain sampling of a Gaussian bandpower likelihood L\mathcal{L} that we define by:

log⁡L(θ)=−12log⁡det⁡C(θ)−12(t(θ)−d)TC−1(t(θ)−d)+const.,(37)\log\mathcal{L}(\theta)=-\frac{1}{2}\log\det C(\theta)-\frac{1}{2}(t(\theta)-d)^{T}C^{-1}(t(\theta)-d)+\mathrm{const.}, \tag*{(37)}

where d\mathbf{d} is a ‘data’ vector that can contain all or some of the bandpowers, Pgvℓ=1(k)P_{gv}^{\ell=1}(k), Pgvℓ=0(k)P_{gv}^{\ell=0}(k), and Pvv(k)P_{vv}(k) at each frequency, and Pgg(k)P_{gg}(k). Here, t(θ)\mathbf{t}(\theta) is our ‘theory’ predictions given (including all parameters for clarity) θ∈{bg,sn,fNL,bv95,bv150,bv220,AtSZ,ACIBsnv95,snv150,snv220,Σs,Σv}\theta\in\{b_g, s_n, f_{\mathrm{NL}}, b_v^{95}, b_v^{150}, b_v^{220}, A_{\mathrm{tSZ}}, A_{\mathrm{CIB}}s_{nv}^{95}, s_{nv}^{150}, s_{nv}^{220}, \Sigma_s, \Sigma_v\}. The matrix C(θ)C(\theta) is the bandpower-covariance we model using surrogate fields. Our surrogate field machinery computes all entries of CC, including correlations of bandpowers between different frequencies.

Later in this section, we define and use four likelihoods with different combinations of the spectra we measure and model:

  1. In §V B, we use the galaxy auto-power spectrum only, which is used to fit bgb_g and sns_n, and these are then held fixed when we measure the galaxy–velocity and velocity-auto signals separately.

  1. In §V C 1, we fit foreground parameters using a likelihood including only the galaxy monopoles Pgvℓ=0P_{gv}^{\ell=0} in each frequency (also see below).

  1. In Sections §V C 2, §V C 3, and §V D 1 we use the galaxy–velocity dipole Pgvℓ=1(k)P_{gv}^{\ell=1}(k) together with the velocity auto-power monopole Pvvℓ=0(k)P_{vv}^{\ell=0}(k).

  1. Finally in §V D 2 we add the galaxy auto-power spectrum to the third likelihood above, and use this for our final fNLf_{\mathrm{NL}} constraints.

We make a few simplifying analysis choices:

  • We measure the galaxy-galaxy correlation separately to fit bgb_g and sns_n, which we use as fixed parameters (after marginalizing over fNLf_{\mathrm{NL}}) for the analysis of the velocity reconstruction separately from PggP_{gg}. We fit these parameters together with the others only for our joint likelihood in §V D.

  • We fit the galaxy-velocity monopole by itself to infer the values of AtSZA_{\mathrm{tSZ}} and ACIBA_{\mathrm{CIB}} from the frequency dependence of (21). Here, we omit using the four largest-scale bandpowers to avoid interpreting the likely-systematic contribution at these scales, and set fNL=Σv=0f_{\mathrm{NL}}=\Sigma_v=0. We fix bgb_g, sns_n and Σs\Sigma_s to their measured values from the PggP_{gg} measurement.

  • Although the surrogate framework lets us evaluate C(θ)C(\theta) at any parameter point, we hold CC fixed in both the likelihood minimization and the sampling, following [16]. We evaluate C(θ)C(\theta) at the fitted values of all parameters, and repeat the MCMC fits with the updated values until no value changes by more than 0.05σ0.05\sigma.

For all parameters, we use uniform priors. In the velocity fits we use bv∈(0,2)b_v \in(0,2), snv∈(0.5,3)s_{nv} \in(0.5,3), fNL∈(−500,500)f_{\mathrm{NL}} \in(-500,500), and Σv∈(0,50)\Sigma_v \in(0,50) Mpc. In the monopole fits we sample the two foreground amplitudes as coefficients of the surrogate templates, with priors (−0.01,0.01)(-0.01,0.01). In the fits that include the galaxy auto-spectrum, we use bg∈(1,4)b_g \in(1,4) and Σs∈(0,50)\Sigma_s \in(0,50) Mpc.

We sample the likelihood with emcee [82], seeding the walkers from a multi-start minimization of the likelihood. We report constraints as marginalized posterior medians with 1σ1\sigma uncertainties. For the goodness of fit, the signal-to-noise ratios, and the model predictions in the figures, we use the maximum-likelihood point of each fit.

The χ2\chi^2 statistics, detection SNR and null tests

We assess whether our models are a good fit by computing the χ2\chi^2 statistic separately for our observables:

χ2=(d−m)TC−1(d−m) ,(38)\chi^2 = (\mathbf{d}-\mathbf{m})^{T}C^{-1}(\mathbf{d}-\mathbf{m}) \, , \tag*{(38)}

where m\mathbf{m} is the best-fit model and the covariance matrix CC here is calculated at the best-fit model.

We quantify a detection signal-to-noise (SNR) value by:

SNR=dTC0−1ms(msTC0−1ms)1/2 ,(39)\mathrm{SNR} = \frac{\mathbf{d}^{T}C_0^{-1}\mathbf{m}_s}{(\mathbf{m}_s^{T}C_0^{-1}\mathbf{m}_s)^{1/2}} \, , \tag*{(39)}

where ms\mathbf{m}_s indicates the difference between our best-fit model and the model at the ‘null point’, where the detected component is set to zero. For example, when showing the SNR for the dipole and monopole measurements below, we subtract our best-fit monopole and dipole leakage contamination from the signals, respectively, and we subtract both the reconstruction noise and the foreground contamination from the velocity auto when calculating its SNR.

Figure 3. Galaxy auto-spectrum Pgg(k)P_{gg}(k) measured within the SPT-3G D1 footprint for the DESI-LS LRG main and extended samples. Top sub-panels show the data with the best-fit model overlaid in black where we set fNLf_{\mathrm{NL}} free (red where we fix fNL=0f_{\mathrm{NL}} = 0); bottom sub-panels show residuals normalised by the diagonal covariance error. Insets show the best-fit bgb_g, sns_n, Σg\Sigma_g values and χ2/Ndof\chi^2/N_{\mathrm{dof}} where fNLf_{\mathrm{NL}} is set free and marginalized. Fixing fNL=0f_{\mathrm{NL}} = 0 gives best-fit values bg=2.156b_g = 2.156 (1.956) and Σg=3.3\Sigma_g = 3.3 (4.8) for the main (extended) sample, with χ2\chi^2 value increased by 0.22 (0.78). We use the model predictions with fNLf_{\mathrm{NL}} set free throughout the paper. The displayed goodness of fit is evaluated at these values and is acceptable in both samples, with PTEgof=0.67\mathrm{PTE}_{\mathrm{gof}} = 0.67 (main) and 0.93 (extended).

Similarly, the covariance C0C_0 in our SNR is the covariance at the null point, where we subtract the best-fit signal from the covariance. For the velocity signal, for example, we set every bvνb_v^\nu to zero and keep the foreground and noise terms in C0C_0; for the foreground-induced monopole, we set AtSZ=ACIB=0A_{\mathrm{tSZ}} = A_{\mathrm{CIB}} = 0. We call Eq. (39) the detection SNRs, as in Hotinli et al. [11], since C0C_0 is the covariance under the null hypothesis; the SNRs we present in this work are these ‘detection’ SNRs.

Joint SNRs over the three frequencies use the full covariance between them, and, for the dipole and the velocity auto-spectra, lie below the quadrature sum of the per-frequency SNRs mainly because the reconstruction noise is correlated between frequencies. With C0C_0 replaced by the covariance at the best fit, which includes the detected component, Eq. (39) gives the ‘amplitude’ SNR. The amplitude SNR measures how precisely the amplitude of ms\boldsymbol{m}_s is determined, not how strongly the null hypothesis is rejected. Note the SNR values shown in Chaussidon et al. [16] for velocity auto are this ‘amplitude’ SNR. Table II lists the detection and amplitude SNRs of the dipole, the monopole, and the velocity auto-spectra.

We report the goodness of fit as the reduced χ2/Ndof\chi^2/N_{\mathrm{dof}} and the probability-to-exceed PTEgof=Pr⁡(χ2>χobs2)\mathrm{PTE}_{\mathrm{gof}} = \Pr(\chi^2 > \chi^2_{\mathrm{obs}}). We quantify each null test by the probability p0p_0 of obtaining a larger χ2\chi^2 than zero by chance and consider a fit or a test to pass if PTEgof\mathrm{PTE}_{\mathrm{gof}} or p0p_0 exceeds 0.05.

We use frequency differences to test consistency between frequencies. The kSZ signal has the blackbody spectrum of the CMB, and the beam equalization of Eq. (36) gives the three frequencies the same response to it, so the kSZ dipole cancels in the difference between two frequencies. A contribution that differs between frequencies, such as foreground leakage, would not cancel. For each pair (νa,νb)(\nu_a,\nu_b) we test Pgvνa,ℓ=1−Pgvνb,ℓ=1P_{\mathrm{gv}}^{\nu_a,\ell=1} - P_{\mathrm{gv}}^{\nu_b,\ell=1} against zero, with the covariance of the difference from the surrogate ensemble at the best fit values, and quote p0p_0.

We also use half-splits from the SPT-3G half-depth coadds of S1V. We make two velocity reconstructions from these and test their difference against zero, the sky signal cancels in the difference, so a dipole there would come from the map noise.

Galaxy auto-spectrum

We begin our analysis by measuring galaxy auto-spectrum. We fit our prediction of P^gg(k)\widehat{P}_{gg}(k) from our surrogate fields with bgb_g, fNLf_{\mathrm{NL}} and the damping scale Σ~g\widetilde{\Sigma}_g set free, and with the shot-noise amplitude sns_n at the Poisson level of the galaxies-minus-randoms field. For the main sample, the best fit is:

bg=2.186−0.087+0.099fNL=−33−47+42,(40)b_g = 2.186^{+0.099}_{-0.087} \qquad f_{\mathrm{NL}} = -33^{+42}_{-47}, \tag*{(40)}

sn=1.025s_n = 1.025, and Σg=4.6 Mpc\Sigma_g = 4.6\ \mathrm{Mpc}, where χ2/Ndof=17.7/21\chi^2/N_{\mathrm{dof}} = 17.7/21 and PTEgof=0.67\mathrm{PTE}_{\mathrm{gof}} = 0.67. The extended sample gives bg=2.009−0.083+0.089b_g = 2.009^{+0.089}_{-0.083}, sn=1.082s_n = 1.082 and Σg=6.6 Mpc\Sigma_g = 6.6\ \mathrm{Mpc} (χ2/Ndof=12.3/21\chi^2/N_{\mathrm{dof}} = 12.3/21, PTEgof=0.93\mathrm{PTE}_{\mathrm{gof}} = 0.93) and fNL=−44−45+43f_{\mathrm{NL}} = -44^{+43}_{-45}. We hold bgb_g, sns_n and Σg\Sigma_g fixed at these values in our analysis of PgvP_{gv} and PvvP_{vv} in what that follows, except in the joint fits of §V D. Figure 3 shows the bandpowers and the best-fit model. Table II collects the parameters for both samples.

It is worth briefly discussing the fNLf_{\mathrm{NL}} constraints we obtain from the galaxy auto-spectrum alone, which are worse but comparable to the DESI-LS analysis of Rezaie et al. [83], which uses the entire footprint of this survey. Here, we only use a fraction of the DESI-LS galaxies on the SPT-3G D1 footprint. Our analysis differs from such clustering analyzes in multiple ways, which may help explain our results. First, we are using a footprint (the SPT-3G footprint in Figure 1) that avoids the mixing of different galaxy and survey samples, so that we avoid using image-systematics treatments that cost sensitivity to these analyzes. Next, we do a 3-d analysis instead of a 2-d (purely angular) analysis in Ref. [83], which gains us more modes. Our pipeline also includes a mean-subtraction for the galaxy field (in addition to our velocity estimates) for both data and surrogate fields, which is essentially equivalent to including a radial integral constraint (RIC) that accounts for the corresponding loss of fNLf_{\mathrm{NL}} sensitivity.

Galaxy–velocity and velocity power spectra

Next, we measure three spectra: the galaxy–velocity dipole, which carries the kSZ signal; the foreground-induced galaxy–velocity monopole; and the velocity auto-spectrum, which contains the kSZ signal, the reconstruction noise, and a foreground term. In this section, we discuss our detection of these signals and our measurement of the velocity and foreground biases. We describe our constraints on fNLf_{\mathrm{NL}} in §V D.

Galaxy–velocity monopole Pgvℓ=0P_{gv}^{\ell=0} and foreground decomposition.

We start by measuring the galaxy-velocity monopole Pgvℓ=0P_{gv}^{\ell=0} in order to quantify the foreground bias and its contamination of the galaxy-velocity dipole, which we will measure next. We detect the monopole signal at SNR 16.8 over the three frequencies for the main sample, the monopole fit gives:

AtSZ=(−4.18−0.28+0.28)×103 μK−1,ACIB=(+0.51−0.34+0.34)×103 μK−1,(41)A_{\mathrm{tSZ}} = \left(-4.18^{+0.28}_{-0.28}\right) \times10^3\ \mu\mathrm{K}^{-1}, \qquad A_{\mathrm{CIB}} = \left(+0.51^{+0.34}_{-0.34}\right) \times10^3\ \mu\mathrm{K}^{-1}, \tag*{(41)}

which set the per-frequency foreground biases through Eq. (21), with posterior medians for the main sample:

bfg={−6.6at 95 GHz−3.7at 150 GHz+1.7at 220 GHz×103 μK−1.(42)b_{\mathrm{fg}} = \begin{cases} -6.6 & \text{at 95 GHz}\\ -3.7 & \text{at 150 GHz}\\ +1.7 & \text{at 220 GHz} \end{cases} \times10^3\ \mu\mathrm{K}^{-1}. \tag*{(42)}

Figure 4 shows the galaxy-velocity monopole bandpowers for the main sample, and surrogate fields model predictions at the best-fit values. The insets shows the per-frequency SNRs. While measurement significance for the CIB contribution is low for the main sample, we detect the CIB amplitude at an SNR exceeding 2 with the extended sample, for which we detect the monopole contribution at SNR value of 18.9. We defer the majority of our results for the extended sample to Appendix A for brevity.

The foreground biases are negative at 95 and 150 GHz, where the tSZ effect is a decrement, and positive at 220 GHz, near the frequency where tSZ effect vanishes and the CIB dominates. The two-component model of Eq. (21) describes the monopole at every frequency, as can be seen from Figure 4, so the measured frequency dependence agrees with that of the tSZ and CIB effects. We also show the subdominant contribution to the monopole from the kSZ effect with gray solid lines, induced by the window convolution.

Galaxy–velocity dipole Pgvℓ=1P_{gv}^{\ell=1}.

Our main observable is the kSZ galaxy–velocity correlation dipole. We detect the kSZ dipole of the main sample at total SNR 10.4. For each frequency, we measure velocity–galaxy dipole at:

SNR={7.7at 95 GHz,9.0at 150 GHz,3.6at 220 GHz.(43)\mathrm{SNR} = \begin{cases} 7.7 & \text{at 95 GHz},\\ 9.0 & \text{at 150 GHz},\\ 3.6 & \text{at 220 GHz}. \end{cases} \tag*{(43)}

This is the first detection of the galaxy-velocity correlation from CMB at 220 GHz. Figure 5 and Figure 6 show the bandpowers and the cumulative signal-to-noise of the velocity-galaxy dipole.

Figure 5. Per-frequency Pgvℓ=1P_{gv}^{\ell=1} dipole, which carries the kSZ signal, within the SPT-3G D1 footprint for the DESI-LS LRG main sample; columns are 95, 150, and 220 GHz. Top sub-panels show kPgvℓ=1(k)kP_{gv}^{\ell=1}(k) with the model predictions. Bottom sub-panels show residuals normalised by the diagonal covariance error. Insets show bvb_v best-fits (posterior median with the offsets to the 84th and 16th percentiles), χ2/Nbins\chi^2/N_{\mathrm{bins}} and the SNR of Eq. (39). The joint SNR beside the row label is evaluated over all 3x24 dipole bandpowers. The 95 GHz bandpower at 0.024<k<0.026 Mpc−10.024 < k < 0.026\,\mathrm{Mpc}^{-1} drives the low goodness of fit discussed in §V C.

Figure 6. Cumulative detection SNR of the main kSZ galaxy-velocity dipole Pgvℓ=1P_{gv}^{\ell=1} for main and extended catalogs. The joint curve uses all three frequencies with the full covariance.

Having set the beam equalization as in Eq. (36), we expect the blackbody kSZ signal to give the same velocity bias at all three frequencies. We find that the data agrees, as the velocity fit with a per-frequency bvb_v gives:

bv={0.290−0.046+0.043at 95 GHz,0.248−0.036+0.033at 150 GHz,0.320−0.087+0.086at 220 GHz,(44)b_v = \begin{cases} 0.290^{+0.043}_{-0.046} & \text{at 95 GHz},\\ 0.248^{+0.033}_{-0.036} & \text{at 150 GHz},\\ 0.320^{+0.086}_{-0.087} & \text{at 220 GHz}, \end{cases} \tag*{(44)}

for the main catalog.

We also define a single bvb_v value shared between all frequencies, and parameterize our surrogate field model predictions accordingly. Setting a single bvb_v raises χ2\chi^2 by 3.0 for two fewer parameters (probability 0.22) and gives

bv=0.255−0.035+0.033.(45)b_v = 0.255^{+0.033}_{-0.035}. \tag*{(45)}

As we described above, the value of bvb_v depends on the ratio between the true galaxy-electron power spectrum Pgetrue(k)P_{ge}^{\mathrm{true}}(k); and the fiducial Pgefid(k)P_{ge}^{\mathrm{fid}}(k) we choose in our analysis. In this analysis, we use the halo model prescription in the hmvec code [6] and compute the Pge(k,z)P_{ge}(k,z) at our central redshift z⋆=0.754z_\star= 0.754. We used the Battaglia gas profile normalized in such a way that the enclosed gas mass equals the cosmic baryon fraction at the virial radius, as in Refs. [11, 16]. This returns a gas profile that do not capture the baryonic feedback that is understood to reduce the ionized gas content of galaxy groups. Measurements find that ionized gas residing within the virial radius of these halos amounts to around only 25-30 percent of the cosmic baryon fraction [see e.g. 34, 36, 84]. Our measurement naturally returns a value lower than unity, in agreement with these observations.

Velocity auto-spectrum Pvvν×νP_{vv}^{\nu\times\nu}.

In addition to the galaxy-velocity correlation, we also detect the kSZ signal in the three auto-spectra of reconstructed velocities. Our detection SNR for the three spectra is at 7.2 (8.1) for the main (extended) sample, the highest detection of the auto-correlations to date. The ‘amplitude’ SNR (which we defined in §V A) is ∼3.3\sim3.3 for both samples.

The velocity auto-spectrum contains the kSZ signal, the reconstruction noise, and the foregrounds. For the detection SNR, we fit and subtract foreground and noise contributions from the velocity signal. We use the best-fit signal component only in the auto-spectra and include the foregrounds and noise in the covariance C0C_0.

The SNR on the amplitude is noticeably lower than the detection SNR because the 95 GHz and 150 GHz auto-spectra are signal dominated on the largest scales, where most of the detection SNR arises. Figure 7 shows the bandpowers with the noise and foreground terms subtracted, and the corresponding best-fit signal we model.

Figure 7. Per-frequency velocity auto-spectrum within the SPT-3G D1 footprint for the DESI-LS LRG main sample. Top sub-panels show PvvP_{vv} at each frequency minus the terms of Eq. (29) that do not depend on bvb_v (i.e. the reconstruction noise and foreground contribution). Solid lines are our model prediction for the best-fit kSZ signal. Bottom sub-panels show normalised residuals. Insets show bvb_v values (posterior median with the offsets to the 84th and 16th percentiles), the posterior median of snvνs_{nv}^{\nu}, χ2/Nbins\chi^2/N_{\rm bins} and the SNR of Eq. (39). The three auto-spectra give PTEgof=0.39\mathrm{PTE}_{\rm gof}=0.39, 0.49 and 0.26 at 95, 150 and 220 GHz. Jointly, the auto-spectra reach kSZ SNR 7.2.

The fitted reconstruction-noise amplitudes are:

snv={1.11 at 95 GHz,1.06 at 150 GHz,1.03 at 220 GHz.(46)s_{nv} = \begin{cases} 1.11\ \text{at 95 GHz},\\ 1.06\ \text{at 150 GHz},\\ 1.03\ \text{at 220 GHz}. \end{cases} \tag*{(46)}

Each has uncertainty 0.01, so the bootstrap slightly underestimates the reconstruction noise. We evaluate the covariance at the fitted amplitudes, which carries this calibration into the uncertainties.

Goodness of fit.

Our fits describe the data well, with one exception. All our fits pass the goodness-of-fit test (Table II), except for the overall velocity fit of the main sample (PTEgof=0.02\mathrm{PTE}_{\rm gof}=0.02). We find that this is mainly sourced by a single 95 GHz galaxy-velocity dipole bandpower, at 0.024<k<0.026 Mpc−10.024 < k < 0.026\ \mathrm{Mpc}^{-1}, visible in Figure 5, which lies 3.8σ3.8\sigma above our best-fit model. This excess appears at no other frequency and in no other spectrum. We do not explore this systematic contribution further in this paper, but will do so in upcoming works.

Constraints on fNLf_{\mathrm{NL}}

Last on our parameter fits, we discuss our fNLf_{\mathrm{NL}} constraints. Local PNG enters our model only through the scale-dependent galaxy bias. The fNLf_{\mathrm{NL}} enters the galaxy auto-spectrum through the square of this bias and the galaxy-velocity correlation linearly.

Constraints on fNLf_{\mathrm{NL}} from velocity-galaxy correlations

We first discuss our constraints on fNLf_{\mathrm{NL}} from velocity reconstruction alone, without the galaxy auto-spectrum. While the velocity auto-spectrum does not depend on fNLf_{\mathrm{NL}}, we find that jointly fitting the galaxy-velocity dipole and velocity-velocity autocorrelation improves fNLf_{\mathrm{NL}} constraints compared to the galaxy-velocity dipole alone, as was seen in Chaussidon et al. [16], by reducing the degeneracy between bvb_v and fNLf_{\mathrm{NL}}. Throughout this work, we will quote constraints on fNLf_{\mathrm{NL}} from the joint analysis of the galaxy-velocity dipole and velocity auto-spectrum.

We sample fNLf_{\mathrm{NL}} with the other free parameters of the velocity. Using the galaxy-velocity dipole and velocity-velocity correlation, we find

fNL=−38−61+61,(47)f_{\mathrm{NL}} = -38^{+61}_{-61}, \tag*{(47)}

for the main sample. Setting bvb_v to be shared gives a similar constraint at fNL=−35−63+62f_{\mathrm{NL}} = -35^{+62}_{-63} with a single bvb_v. The extended sample gives a similar constraint at fNL=−50−63+63f_{\mathrm{NL}} = -50^{+63}_{-63}, all consistent with zero and with the galaxy auto-spectrum at Eq. (40).

We test our fNLf_{\mathrm{NL}} constraints a range of analysis choices: dropping the 95 GHz outlier bandpower; keeping only the dipole at k<0.02 Mpc−1k < 0.02\ \mathrm{Mpc}^{-1}; or fitting the monopole together with the dipole and velocity auto (recall that we fit the monopole separately in our baseline analysis). These choices move fNLf_{\mathrm{NL}} by less than 0.4σ0.4\sigma in both main and extended samples and do not change our constraints.

These are the first measurements of PNG from kSZ using SPT-3G maps, on a relatively small footprint (fsky∼0.05f_{\mathrm{sky}} \sim0.05) in the sky, and from data collected between 2019–2020. We find SPT-3G maps give a constraint on fNLf_{\mathrm{NL}} comparable to our analysis on ACT DR6×\timesDESI-LS in Chaussidon et al. [16], which found the LRG sample gives σ(fNL)≃±44\sigma(f_{\mathrm{NL}}) \simeq\pm44 on a larger sky area. While our fNLf_{\mathrm{NL}} constraints are noticeably weaker than those from the CMB bispectrum and from DESI galaxy clustering [42, 43] upcoming CMB and large-scale structure surveys will promptly close this gap. Our measurements show how largely insensitive the galaxy-velocity cross-correlation is to imaging systematics, which can add spurious power to galaxy clustering on the large scales where the PNG signal is largest [83, 85], adding further value to kSZ velocity reconstruction and fNLf_{\mathrm{NL}} measurements from these signatures.

Joint measurements of fNLf_{\mathrm{NL}} from velocity-galaxy and galaxy-galaxy auto

Next, we add the galaxy auto-spectrum to the velocity measurements. Combining galaxy auto with galaxy-velocity dipole, in principle, allows sample-variance cancellation between the two tracers [7, 41] and measuring fNLf_{\mathrm{NL}} without the limitation of cosmic variance. Our measurement significance here is not yet sufficient for us to be in this regime.

We free bgb_g and Σg\Sigma_g and append the 24 galaxy auto-spectrum bandpowers to the velocity data vector, with the Pgg×PgvP_{gg} \times P_{gv} and Pgg×PvvP_{gg} \times P_{vv} cross-covariances included. We hold bvgb_v^g fixed at its best-fit value from the monopole measurements, and make all other choices as in §V A. The full fit gives

fNL=−20−46+40,(48)f_{\mathrm{NL}} = -20^{+40}_{-46}, \tag*{(48)}

for the main catalog, where setting a single bvb_v for all frequencies gives a similar fNL=−17−46+40f_{\mathrm{NL}} = -17^{+40}_{-46}, both consistent with zero. Adding the velocity spectra to the galaxy auto-spectrum moves fNLf_{\mathrm{NL}} from −33−47+42-33^{+42}_{-47} and narrows its interval by a few percent. The full fit gives bg=2.20−0.06+0.08b_g = 2.20^{+0.08}_{-0.06} with bvb_v values consistent with the PggP_{gg} and the velocity reconstruction best-fit values. The velocity biases, noise amplitudes, and velocity damping scale move by at most 0.18σ0.18\sigma from the velocity fits. Our constraints on these parameters are shown in Figure 8.

Figure 8. The marginalised posteriors of fNLf_{\mathrm{NL}}, bgb_g and bvb_v values (the other parameters are marginalised) and AtSZA_{\mathrm{tSZ}} and ACIBA_{\mathrm{CIB}} are held at the maximum-likelihood point of the monopole fit in both fits. The velocity fit of PvvP_{vv} and the PgvP_{gv} dipole (blue) in §V D 1 and the full fit in §V D 2, which adds PggP_{gg} (orange), are overlaid. The fNLf_{\mathrm{NL}} constraints of the two fits are −38−61+61-38^{+61}_{-61} and −20−46+40-20^{+40}_{-46}, respectively, both consistent with zero. The green contour shows the constraints from PggP_{gg} alone. In the velocity fit in blue, the bgb_g value is fixed at the PggP_{gg} best fit of §V B. The contours enclose 39% and 86% of the two-dimensional posterior mass; stars and dashed lines mark the posterior medians.

It is valuable to note that the Fisher forecast of Ref. [7], evaluated for our survey (volume, kk range, galaxy density, reconstruction noise, and photometric-redshift errors) with our measured bgb_g and bvb_v, predicts a ∼10\sim10–$20%$ smaller σ(fNL)\sigma(f_{\mathrm{NL}}) than the galaxy auto-spectrum alone. In this first measurement, adding the velocities narrows the fNLf_{\mathrm{NL}} interval by ∼3%\sim3\% in the full fit. We trace this shortfall to the correlation between the galaxy and velocity signals in our bandpowers. Fisher forecasts treat each Fourier mode as measured on its own, so on linear scales, the two signals are perfectly correlated. In a survey of our size, the window and the photometric-redshift errors mix the largest scale modes, and at k≲0.01 Mpc−1k \lesssim0.01\ \mathrm{Mpc}^{-1}, where most of our fNLf_{\mathrm{NL}} information lies, the correlation falls from 0.9 to 0.4 toward the largest scales. With the measured correlation and velocity amplitude, the same forecast predicts a $\sim 4% improvement, in line with what we find. The gap, therefore, reflects how the largest scales are treated at the current level of precision. In upcoming works, we will aim to improve our treatment of the largest scales to achieve better agreement between the estimator and a Fisher forecast.

Consistency and null tests

Finally, we discuss our consistency checks and null tests in this section. Our null tests consist of differencing our galaxy-velocity dipole correlation measurements at different frequencies. We also include half-split null tests, using SPT-3G half-depth coadds.

Frequency-difference nulls

The kSZ signal has a black-body frequency dependence. Our treatment of the velocity-reconstruction estimator also assumes this: our measured velocity-galaxy correlations should be frequency independent. Our frequency-difference null tests this directly.

We define three frequency pairs with (να,νβ)(\nu_\alpha, \nu_\beta) and measure the difference Pgvℓ=1,να−Pgvℓ=1,νβP_{gv}^{\ell=1,\nu_\alpha}-P_{gv}^{\ell=1,\nu_\beta} using the corresponding

Half-split null.
Observable QuantityMain per-frequency bvνb_v^\nuMain single bvb_vExtended per-frequency bvνb_v^\nuExtended single bvb_v
Galaxy auto-spectrum PggP_{gg}: best fit; bgb_g, sns_n and Σg\Sigma_g held fixed in the monopole and velocity fits
bgb_g2.1862.009
sns_n (Poisson level)1.0251.082
Σg\Sigma_g [Mpc]4.596.57
fNLf_{\mathrm{NL}} (marginalised)−33−47+42-33^{+42}_{-47}−44−45+43-44^{+43}_{-45}
χ2/Ndof\chi^2/N_{\mathrm{dof}} (PTE)17.7/21 (0.67)12.3/21 (0.93)
Monopole fits (Pgvℓ=0P_{gv}^{\ell=0}): foreground amplitudes [103 μK−110^3\,\mu\mathrm{K}^{-1}] and goodness of fit
AtSZA_{\mathrm{tSZ}}−4.18−0.28+0.28-4.18^{+0.28}_{-0.28}−4.18−0.28+0.28-4.18^{+0.28}_{-0.28}−4.01−0.24+0.23-4.01^{+0.23}_{-0.24}−4.02−0.24+0.24-4.02^{+0.24}_{-0.24}
ACIBA_{\mathrm{CIB}}+0.51−0.34+0.34+0.51^{+0.34}_{-0.34}+0.51−0.33+0.34+0.51^{+0.34}_{-0.33}+0.72−0.29+0.30+0.72^{+0.30}_{-0.29}+0.74−0.30+0.30+0.74^{+0.30}_{-0.30}
SNR, detection16.816.818.918.8
SNR, amplitude15.515.517.017.0
χ2/Ndof\chi^2/N_{\mathrm{dof}} (PTE)38.6/58 (0.98)38.6/58 (0.98)51.6/58 (0.71)51.4/58 (0.72)
Velocity fits (Pgvℓ=1P_{gv}^{\ell=1} dipole and PvvP_{vv} auto-spectra, bfgνb_{\mathrm{fg}}^\nu held fixed)
bv95b_v^{95}0.290−0.046+0.0430.290^{+0.043}_{-0.046}—0.240−0.042+0.0390.240^{+0.039}_{-0.042}—
bv150b_v^{150}0.248−0.036+0.0330.248^{+0.033}_{-0.036}—0.266−0.035+0.0320.266^{+0.032}_{-0.035}—
bv220b_v^{220}0.320−0.087+0.0860.320^{+0.086}_{-0.087}—0.427−0.119+0.1070.427^{+0.107}_{-0.119}—
single bvb_v—0.255−0.035+0.0330.255^{+0.033}_{-0.035}—0.251−0.032+0.0300.251^{+0.030}_{-0.032}
Σv\Sigma_v [Mpc]5.5−3.9+5.85.5^{+5.8}_{-3.9}5.0−3.5+5.65.0^{+5.6}_{-3.5}5.5−3.8+5.75.5^{+5.7}_{-3.8}5.4−3.8+5.85.4^{+5.8}_{-3.8}
fNLf_{\mathrm{NL}}−38−61+61-38^{+61}_{-61}−35−63+62-35^{+62}_{-63}−58−72+69-58^{+69}_{-72}−50−63+63-50^{+63}_{-63}
Pgvℓ=1P_{gv}^{\ell=1} SNR, detection10.410.411.211.3
PvvP_{vv} SNR, detection7.26.98.17.9
Pgvℓ=1P_{gv}^{\ell=1} SNR, amplitude7.37.47.87.7
PvvP_{vv} SNR, amplitude3.33.33.33.2
χ2/Ndof\chi^2/N_{\mathrm{dof}} (PTE)172.3/136 (0.019)175.3/138 (0.017)154.1/136 (0.14)164.9/138 (0.059)
Full fit (Pgg+Pgvℓ=1+PvvP_{gg}+P_{gv}^{\ell=1}+P_{vv})
fNLf_{\mathrm{NL}}−20−46+40-20^{+40}_{-46}−17−46+40-17^{+40}_{-46}−40−41+41-40^{+41}_{-41}−38−39+40-38^{+40}_{-39}
bgb_g2.20−0.06+0.082.20^{+0.08}_{-0.06}2.20−0.06+0.082.20^{+0.08}_{-0.06}1.99−0.06+0.071.99^{+0.07}_{-0.06}2.00−0.05+0.072.00^{+0.07}_{-0.05}
χ2/Ndof\chi^2/N_{\mathrm{dof}} (PTE)185.5/158 (0.067)188.9/160 (0.059)165.5/158 (0.32)176.3/160 (0.18)

Table 2. Parameter constraints from SPT-3G and DESI-LS LRG main and extended samples, as described in the main text. Rows correspond to our results from galaxy auto-spectrum, the galaxy-velocity monopole, galaxy-velocity dipole and velocity-velocity correlations. The last set of rows correspond to our joint fit of galaxy auto and galaxy-velocity dipole and velocity-velocity correlation. Columns correspond to main and extended DESI-LS samples and two different parametrization of kSZ velocity bias: separate fits for each frequency, and a single bvb_v for all frequency maps. We give the details and figures of the extended sample in Appendix A.

covariances calculated with the surrogate fields, which share the density realization and reconstruction noise between the three frequencies.

We find the χ2\chi^2 and null probabilities p0p_0 of the three frequency pairs (Table III) to be

PairSampleχ2/Ndof\chi^2/N_{\mathrm{dof}}p0p_0 (single bvb_v)
95 − 150Main33.8/240.08 (0.09)
Extended22.2/240.57 (0.57)
95 − 220Main23.2/240.51 (0.51)
Extended30.7/240.16 (0.22)
150 − 220Main19.1/240.75 (0.75)
Extended29.0/240.22 (0.29)

Table III. SPT-3G × DESI-LS frequency-difference null tests on the Pgvℓ=1P_{gv}^{\ell=1} dipole.

(95−150) GHz:χ2/Ndof=1.41,p0=0.09,(95-150)\ \mathrm{GHz}: \quad\chi^2/N_{\mathrm{dof}}=1.41,\quad p_0=0.09,
(95−220) GHz:χ2/Ndof=0.97,p0=0.51,(95-220)\ \mathrm{GHz}: \quad\chi^2/N_{\mathrm{dof}}=0.97,\quad p_0=0.51,
(150−220) GHz:χ2/Ndof=0.79,p0=0.75.(49)(150-220)\ \mathrm{GHz}: \quad\chi^2/N_{\mathrm{dof}}=0.79,\quad p_0=0.75. \tag*{(49)}

All 24 bandpowers enter every difference, with the covariance evaluated at the velocity fit with a single bvb_v. Every difference passes p0>0.05p_0>0.05 for both main and extended samples, and the largest excess is 1.3σ1.3\sigma, from the 95–150 GHz pair. Figure 9 shows the bandpower differences.

Frequency-difference tests of the $P_{gv}^{\ell=1}$ dipole for the DESI-LS LRG main sample

Figure 9. Frequency-difference tests of the Pgvℓ=1P_{gv}^{\ell=1} dipole for the DESI-LS LRG main sample. Columns show 95–150, 95–220, and 150–220 GHz; bottom sub-panels show the corresponding normalised residuals. Error bars are the square roots of the diagonal of the difference covariance. Insets shows χ2/Ndof\chi^2/N_{\mathrm{dof}} and the probability p0p_0 of Table III. All three tests pass.

Half-split null

In order to test that the map noise adds no dipole, we also measure the Pgvℓ=1P_{gv}^{\ell=1} dipole of the extended sample, which contains every main-sample galaxy, on the half-difference of the two half-depth coadds of SPT-3G D1. We use the weights and filter of the full-depth maps. We estimate the covariance from 2000 surrogate realizations built on the half-difference maps, with no kSZ or foreground term and with the noise amplitudes snvs_{nv} of the extended-sample velocity fit. The dipole is consistent with zero at 95, 150 and 220 GHz (χ2/Ndof=1.21\chi^2/N_{\mathrm{dof}} = 1.21, 0.81 and 0.93 for 24 bandpowers; p0=0.22p_0 = 0.22, 0.72 and 0.56).

Bandpower correlations

Finally, Figure 10 shows the correlation matrix of the covariance of the main-sample galaxy-velocity and velocity-velocity correlations. Below k=0.01k = 0.01 Mpc−1^{-1}, the dipoles of two frequencies are correlated by up to 0.89 because they share the velocity signal. Above k=0.02k = 0.02 Mpc−1^{-1}, the correlation between frequencies, in the dipoles and in the velocity auto-spectra, follows the correlation of the reconstruction noise, which the three maps share through the primary CMB and foreground anisotropies.

Figure 10. Correlation coefficients Rij=Cij/(CiiCjj)1/2R_{ij}=C_{ij}/(C_{ii}C_{jj})^{1/2} of the covariance CC of the main-sample galaxy-galaxy, galaxy-velocity dipole and velocity-velocity surrogates at best-fit values at ν=95\nu=95, 150 and 220 GHz. Within each block, kk increases to the right and downward over the 24 bandpowers, centred at k=0.001k=0.001 to 0.047 Mpc−10.047\,\mathrm{Mpc}^{-1}.

Discussion

We have measured for the first time the galaxy–velocity cross-power spectrum and the velocity auto-spectrum using SPT-3G D1 maps at 95, 150, and 220 GHz and DESI-LS LRGs. Together with PgeP_{ge} from the same D1 maps and DESI-LS LRGs [36], both observables of the kSZ bispectrum [6] (the small-scale PgeP_{ge} and the large-scale PgvP_{gv}) have now been measured at high accuracy and in agreement with ACT results.

For the main sample, we detect the kSZ signal in the galaxy-velocity dipole at SNR 10.4 and in the velocity auto-spectrum at 7.2, combining the three frequencies. The velocity biases fitted at the three frequencies agree, and the frequency differences of the dipole are consistent with zero, as expected for the frequency-independent kSZ signal. We measure the foreground bias from the galaxy-velocity monopole, which is dominated by the tSZ effect at 95 and 150 GHz and by the CIB at 220 GHz.

We constrain fNLf_{\mathrm{NL}} to around ±60\pm60 using the velocity-galaxy correlation and at around ±40\pm40 combined with the galaxy auto-spectrum. Our fNLf_{\mathrm{NL}} measurements are consistent with zero, as in previous ACT ×\times DESI measurements [11, 16]. Across our alternative analysis choices, fNLf_{\mathrm{NL}} moves by less than 0.4σ0.4\sigma in both samples, and the extended sample agrees with the main sample.

The reconstruction recovers the radial velocity field up to the scale-independent factor bvb_v, which we marginalize over in our fNLf_{\mathrm{NL}} constraints. The value of bvb_v depends on the fiducial galaxy-electron cross-spectrum PgeP_{ge} we assume on small scales, and is therefore also a probe of astrophysics.

Our halo-model–based calculation of PgeP_{ge} uses the Battaglia profile [86], truncated at the virial radius and normalized so that the enclosed gas mass equals the cosmic baryon fraction Ωb/Ωm\Omega_b/\Omega_m of the halo mass; on large scales, the electrons then trace the matter. This construction largely fails to capture the reduction in PgeP_{ge} on small scales due to baryonic feedback seen in recent measurements [e.g. 32–34, 36, 84]. As a result, we find bv∼0.25b_v\sim0.25; previous ACT×\timesDESI measurements [11, 16] likewise found values below unity. The fit quality (SNRs, PTEs, and p0p_0 values) and our fNLf_{\mathrm{NL}} constraints do not depend on this choice.

Furthermore, since bvb_v is roughly a weighted average over multipoles of the ratio Pgetrue/PgefidP_{ge}^{\mathrm{true}}/P_{ge}^{\mathrm{fid}}, filters that weight scales differently can yield different values of bvb_v. We apply a uniform filter across our SPT-3G D1 maps, but because SPT-3G has lower noise on small scales, this filter is centered at higher multipoles than the ACT filter used in Hotinli et al. [11] (median ℓ≈5100\ell\approx5100 versus 3600). Baryonic feedback suppresses PgetrueP_{ge}^{\mathrm{true}} relative to PgefidP_{ge}^{\mathrm{fid}} on small scales, so the extra weight our filter gives these scales lowers bvb_v further. We argue that this partly explains the difference between the bvb_v measurements in Hotinli et al. [11] and in this work.

For the first time, we detect the velocity–galaxy cross-correlation monopole and dipole signals with the 220 GHz CMB maps. We use the 220 GHz monopole to separate and measure the tSZ and CIB contributions. We detect the CIB amplitude with monopole at over 2σ2\sigma significance with the extended sample.

Our frequency difference tests of the dipole are consistent with zero, as expected for the frequency-independent kSZ signal. Across alternative analysis choices we consider, we find fNLf_{\mathrm{NL}} moves by less than 0.4σ0.4\sigma, and the extended sample agrees with the main sample. As in previous works [11, 16], we find that the galaxy–velocity cross-correlation is largely insensitive to imaging systematics.

Beyond fNLf_{\mathrm{NL}} [7], reconstructed large-scale velocities can probe early-Universe physics including the primordial trispectrum amplitude τNL\tau_{\mathrm{NL}} [87], other primordial shapes [88], compensated isocurvature perturbations [89, 90], pre-inflationary relics [91], primordial vector fields [92], and CMB anomalies [93]. Reconstructed velocities also provide late-time tests of dark energy [94], modified gravity [95], neutrino mass [96] and relativistic effects [97], and probe the epoch of reionisation [98]. Velocity reconstruction has recently been used to bound an intrinsic dipole between the CMB and large-scale structure [8] and the cosmic bulk flow [14].

Our results use the observations from 2019 and 2020, the first two seasons of the SPT-3G Main survey. The full survey depth will improve the velocity reconstruction, and the growing spectroscopic overlap with DESI will improve the photometric-redshift calibration. In the southern sky, the Rubin Observatory LSST [99] and Euclid [100] will provide deeper galaxy samples for velocity reconstruction with SPT as the CMB counterpart; larger maps and catalogs over a wider area will strengthen the constraint on primordial non-Gaussianity [7] and beyond. A follow-up paper will combine these SPT-3G measurements of this paper with ACT×\timesDESI-LS measurements in the Galactic North.

Acknowledgments

We thank Gil Holder for very useful discussions. SCH was supported by the P. J. E. Peebles Fellowship at Perimeter Institute. SCH acknowledges support from grant NSF PHY-2309135 to the Kavli Institute for Theoretical Physics (KITP), where this work was initiated. SCH also acknowledges the Aspen Center for Physics, where this work was further developed and which is supported by National Science Foundation grant PHY-2210452. EC and SF are supported by Lawrence Berkeley National Laboratory and the Director, Office of Science, Office of High Energy Physics of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. KMS was supported by an NSERC Discovery Grant, by the Daniel Family Foundation, and by the Centre for the Universe at Perimeter Institute. Research at Perimeter Institute is supported in part by the Government of Canada through the Department of Innovation, Science and Economic Development and by the Province of Ontario through the Ministry of Colleges and Universities. The computations reported in this work were run on the high-performance computing cluster at Perimeter Institute.

The South Pole Telescope program is supported by the National Science Foundation (NSF) through awards OPP-1852617 and OPP-2332483. Partial support is also provided by the Kavli Institute of Cosmological Physics at the University of Chicago.

We acknowledge the use of public data from the SPT-3G D1 release and from the DESI Legacy Imaging Surveys, the latter made available through the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility operated under Contract No. DE-AC02-05CH11231.

The Legacy Surveys consist of three individual and complementary projects: the Dark Energy Camera Legacy Survey (DECaLS; Proposal ID 2014B-0404; PIs: David Schlegel and Arjun Dey), the Beijing-Arizona Sky Survey (BASS; NOAO Prop. ID 2015A-0801; PIs: Zhou Xu and Xiaohui Fan), and the Mayall z-band Legacy Survey (MzLS; Prop. ID 2016A-0453; PI: Arjun Dey). DECaLS, BASS and MzLS together include data obtained, respectively, at the Blanco telescope, Cerro Tololo Inter-American Observatory, NSF’s NOIRLab; the Bok telescope, Steward Observatory, University of Arizona; and the Mayall telescope, Kitt Peak National Observatory, NOIRLab. Pipeline processing and analyses of the data were supported by NOIRLab and the Lawrence Berkeley National Laboratory (LBNL). The Legacy Surveys project is honored to be permitted to conduct astronomical research on Iolkam Du’ag (Kitt Peak), a mountain with particular significance to the Tohono O’odham Nation. NOIRLab is operated by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation. LBNL is managed by the Regents of the University of California under contract to the U.S. Department of Energy.

This project used data obtained with the Dark Energy Camera (DECam), which was constructed by the Dark Energy Survey (DES) collaboration. Funding for the DES Projects has been provided by the U.S. Department of Energy, the U.S. National Science Foundation, the Ministry of Science and Education of Spain, the Science and Technology Facilities Council of the United Kingdom, the Higher Education Funding Council for England, the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, the Kavli Institute of Cosmological Physics at the University of Chicago, Center for Cosmology and Astro-Particle Physics at the Ohio State University, the Mitchell Institute for Fundamental Physics and Astronomy at Texas A&M University, Financiadora de Estudos e Projetos, Fundacao Carlos Chagas Filho de Amparo a Pesquisa do Estado do Rio de Janeiro, Conselho Nacional de Desenvolvimento Cientifico e Tecnologico and the Ministerio da Ciencia, Tecnologia e Inovacao, the Deutsche Forschungsgemeinschaft and the Collaborating Institutions in the Dark Energy Survey. The Collaborating Institutions are Argonne National Laboratory, the University of California at Santa Cruz, the University of Cambridge, Centro de Investigaciones Energeticas, Medioambientales y Tecnologicas-Madrid, the University of Chicago, University College London, the DES-Brazil Consortium, the University of Edinburgh, the Eidgenossische Technische Hochschule (ETH) Zurich, Fermi National Accelerator Laboratory, the University of Illinois at Urbana-Champaign, the Institut de Ciencies de l’Espai (IEEC/CSIC), the Institut de Fisica d’Altes Energies, Lawrence Berkeley National Laboratory, the Ludwig Maximilians Universitat Munchen and the associated Excellence Cluster Universe, the University of Michigan, NSF’s NOIRLab, the University of Nottingham, the Ohio State University, the University of Pennsylvania, the University of Portsmouth, SLAC National Accelerator Laboratory, Stanford University, the University of Sussex, and Texas A&M University.

BASS is a key project of the Telescope Access Program (TAP), which has been funded by the National Astronomical Observatories of China, the Chinese Academy of Sciences (the Strategic Priority Research Program “The Emergence of Cosmological Structures” Grant # XDB09000000), and the Special Fund for Astronomy from the Ministry of Finance. The BASS is also supported by the External Cooperation Program of Chinese Academy of Sciences (Grant # 114A11KYSB202160057), and Chinese National Natural Science Foundation (Grant # 12120101003, # 11433005). The Legacy Survey team makes use of data products from the Near-Earth Object Wide-field Infrared Survey Explorer (NEOWISE), which is a project of the Jet Propulsion Laboratory/California Institute of Technology. NEOWISE is funded by the National Aeronautics and Space Administration. The Legacy Surveys imaging of the DESI footprint is supported by the Director, Office of Science, Office of High Energy Physics of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231, by the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility under the same contract; and by the U.S. National Science Foundation, Division of Astronomical Sciences under Contract No. AST-0950945 to NOAO.

Data availability

The SPT-3G D1 maps and ancillary products are publicly available through the SPT-3G data-release Globus collection “SPT-3G Public Data Release” (0bf0cefa-568a-4b37-861d-32c9102ce7d6, documented at https://pole.uchicago.edu/public/data/quan26/index.html; see [53]), and the DESI Legacy Imaging Surveys DR9 LRG catalogues [54] and randoms through the DESI public data server (https://data.desi.lbl.gov/public/papers/c3/lrg_xcorr_2023/ and https://data.desi.lbl.gov/public/ets/target/catalogs/dr9/0.49.0/randoms/). The modified version of the kszx package (https://github.com/kmsmith137/kszx) used in this work, together with the analysis code that produces the numbers and figures of this paper, will be made public upon publication and is available from the authors on request before then.

The extended sample

In §V we focused on the main galaxy sample of DESI-LS LRGs; here, we briefly go over our results for the extended sample that we skipped in the main part of this pair. Overall, we find that the main and extended samples agree very well with each other.

We detect the galaxy-velocity correlation dipole for the extended sample at

SNR⁡(velocity dipole)={7.7at 95 GHz,10.7at 150 GHz,4.2at 220 GHz,(50)\operatorname{SNR}\left(\text{velocity dipole}\right) = \begin{cases} 7.7 & \text{at 95 GHz},\\ 10.7 & \text{at 150 GHz},\\ 4.2 & \text{at 220 GHz}, \end{cases} \tag*{(50)}

and we find the total SNR of 11.2. Fitting the velocity bias values for each frequency give

bv={0.240−0.042+0.039 at 95 GHz,0.266−0.035+0.032 at 150 GHz,0.427−0.119+0.107 at 220 GHz.(51)b_v = \begin{cases} 0.240^{+0.039}_{-0.042}\ \text{at 95 GHz},\\ 0.266^{+0.032}_{-0.035}\ \text{at 150 GHz},\\ 0.427^{+0.107}_{-0.119}\ \text{at 220 GHz}. \end{cases} \tag*{(51)}

A single bvb_v shared accross frequencies gives the same SNR, and:

bv=0.251−0.032+0.030,(52)b_v = 0.251^{+0.030}_{-0.032}, \tag*{(52)}

consistent with the per-frequency values. Figure 11 galaxy-velocity dipole bandpowers. For the foreground amplitudes, the galaxy-velocity monopole fit gives

Figure 11. As Figure 5, for the extended sample. The three dipoles give PTEgof=0.50\mathrm{PTE}_{\mathrm{gof}} = 0.50, 0.70 and 0.36 at 95, 150 and 220 GHz.

AtSZ=(−4.01−0.24+0.23)×103 μK−1,ACIB=(+0.72−0.29+0.30)×103 μK−1(53)A_{\mathrm{tSZ}} = (-4.01^{+0.23}_{-0.24}) \times10^3\ \mu\mathrm{K}^{-1}, \qquad A_{\mathrm{CIB}} = (+0.72^{+0.30}_{-0.29}) \times10^3\ \mu\mathrm{K}^{-1} \tag*{(53)}

and the corresponding foreground biases of the extended sample are

bfg={−6.3 at 95 GHz−3.3 at 150 GHz+2.3 at 220 GHz×103 μK−1.(54)b_{\mathrm{fg}} = \begin{cases} -6.3\ \text{at 95 GHz}\\ -3.3\ \text{at 150 GHz}\\ +2.3\ \text{at 220 GHz} \end{cases} \times10^3\ \mu\mathrm{K}^{-1}. \tag*{(54)}

We detect foreground-induced monopole is detected at

SNR (velocity monopole)={14.4 at 95 GHz,10.5 at 150 GHz,3.1 at 220 GHz,(55)\mathrm{SNR}\ (\text{velocity monopole}) = \begin{cases} 14.4\ \text{at 95 GHz},\\ 10.5\ \text{at 150 GHz},\\ 3.1\ \text{at 220 GHz}, \end{cases} \tag*{(55)}

and the total SNR of 18.9. Figure 12 shows the bandpowers with the model of the monopole fit. The reconstruction-noise amplitudes are

Figure 12. As Figure 4, for the extended sample. The three monopoles give PTEgof=0.19\mathrm{PTE}_{\mathrm{gof}} = 0.19, 0.98 and 0.97 at 95, 150 and 220 GHz.

Figure 4. Per-frequency foreground-induced Pgvℓ=0P_{gv}^{\ell=0} monopole within the SPT-3G D1 footprint for the DESI-LS LRG main sample. Top sub-panels show kPgvℓ=0(k)kP_{gv}^{\ell=0}(k) with the model of the monopole fit and its tSZ, CIB and kSZ-signal components (thick dashed and dotted for tSZ and CIB, and thin solid line for kSZ). Bottom sub-panels show normalised residuals. We fit the model omitting the first four bandpowers, which drawn as faded open markers. Insets shows bfgνb_{\mathrm{fg}}^{\nu} values in units of 103 μK10^3\,\mu\mathrm{K} (posterior median with the offsets to the 84th and 16th percentiles), χ2/Nbins\chi^2/N_{\mathrm{bins}} and the foreground-monopole SNR of Eq. (39) of that frequency. Over these 20 bandpowers, the three monopoles give PTEgof=0.83\mathrm{PTE}_{\mathrm{gof}}=0.83, 0.95 and 0.96 at 95, 150 and 220 GHz. The frequency dependence follows the tSZ+CIB model of Eq. (21).

snv={1.09 at 95 GHz,1.05 at 150 GHz,1.02 at 220 GHz,(56)s_{nv} = \begin{cases} 1.09\ \text{at 95 GHz},\\ 1.05\ \text{at 150 GHz},\\ 1.02\ \text{at 220 GHz}, \end{cases} \tag*{(56)}

each with uncertainty 0.01. The per-frequency kSZ SNRs of the velocity auto-spectra are

SNR (velocity auto)={6.1 at 95 GHz,7.6 at 95 GHz,3.8 at 95 GHz,(57)\mathrm{SNR}\ (\text{velocity auto}) = \begin{cases} 6.1\ \text{at 95 GHz},\\ 7.6\ \text{at 95 GHz},\\ 3.8\ \text{at 95 GHz}, \end{cases} \tag*{(57)}

and their joint kSZ SNR is 8.1. The auto-spectra give χ2={24.3,25.3,30.1}\chi^2 = \{24.3, 25.3, 30.1\} (24 bandpowers each) at the best fit, where PTEgof={0.44,0.39,0.18}\mathrm{PTE}_{\mathrm{gof}} = \{0.44, 0.39, 0.18\}. For the velocity damping scale, we find Σv=5.5−3.8+5.7 Mpc\Sigma_v = 5.5^{+5.7}_{-3.8}\ \mathrm{Mpc}. Figure 13 shows the noise- and foreground-subtracted bandpowers and the signal model.

Figure 13. As Figure 7, for the extended sample. Jointly, the auto-spectra reach kSZ SNR 8.1. The three auto-spectra give PTEgof=0.44\mathrm{PTE}_{\mathrm{gof}} = 0.44, 0.39 and 0.18 at 95, 150 and 220 GHz.

For the extended sample, we find that the Pgvℓ=1+PvvP_{gv}^{\ell=1} + P_{vv} fits gives:

fNL=−58−72+69,(58)f_{\mathrm{NL}} = -58^{+69}_{-72}, \tag*{(58)}

where for a single bvb_v shared between different frequencies, we find fNL=−50−63+63f_{\mathrm{NL}} = -50^{+63}_{-63}, both consistent with zero. The full fit (Pgg+Pgvℓ=1+PvvP_{gg} + P_{gv}^{\ell=1} + P_{vv}) gives:

bg=1.99−0.06+0.07andfNL=−40−41+41,(59)b_g = 1.99^{+0.07}_{-0.06} \qquad\text{and} \qquad f_{\mathrm{NL}} = -40^{+41}_{-41}, \tag*{(59)}

and fNL=−38−39+40f_{\mathrm{NL}} = -38^{+40}_{-39} with a single bvb_v. We give the fNLf_{\mathrm{NL}} values from PggP_{gg} alone in §V B.

The three frequency differences of the extended sample are consistent with zero. The reduced χ2\chi^2 and null proba

bilities p0p_0 (Table III) are:

(95−150) GHz:χ2/Ndof=0.93,p0=0.57,(95−220) GHz:χ2/Ndof=1.28,p0=0.16,(150−220) GHz:χ2/Ndof=1.21,p0=0.22,(60)\begin{aligned} (95-150)\ \mathrm{GHz}: \quad\chi^2/N_{\mathrm{dof}} &= 0.93,\quad p_0 = 0.57,\\ (95-220)\ \mathrm{GHz}: \quad\chi^2/N_{\mathrm{dof}} &= 1.28,\quad p_0 = 0.16,\\ (150-220)\ \mathrm{GHz}: \quad\chi^2/N_{\mathrm{dof}} &= 1.21,\quad p_0 = 0.22, \tag*{(60)} \end{aligned}

with the covariance evaluated at the velocity fit with a single bvb_v. Concatenating 95–150 and 150–220 GHz with their full cross-pair covariance gives χ2/Ndof=49.4/48\chi^2/N_{\mathrm{dof}} = 49.4/48 (p0=0.42p_0 = 0.42). Figure 14 shows the difference null test bandpowers.

Figure 14. As Figure 9, for the extended sample. All three tests pass.

Figure 15 compares the posteriors of the main and extended samples. The fNLf_{\mathrm{NL}} values of Eqs. (47) and (58) differ by less than the error of either. Every main-sample galaxy belongs to the extended sample, so the errors are correlated, and we quote no significance for the difference. The tSZ amplitudes of the two samples agree within their errors, and we find AtSZmain/AtSZext=1.04A_{\mathrm{tSZ}}^{\mathrm{main}}/A_{\mathrm{tSZ}}^{\mathrm{ext}} = 1.04.

Figure 15. Marginalized posteriors of the per-frequency velocity biases bvfb_v^f, fNLf_{\mathrm{NL}} and the velocity damping scale Σv\Sigma_v from the velocity fit, for the main (red, filled) and extended (grey, dashed) samples. The reconstruction-noise amplitudes snvs_{nv} are marginalized and not shown. The diagonal panels show the one-dimensional marginals; the contours enclose the 1σ1\sigma and 2σ2\sigma regions of a two-dimensional Gaussian (39% and 86% of the two-dimensional posterior mass). Stars and dashed lines mark the posterior medians. The Σv\Sigma_v marginals peak at the lower prior edge, Σv=0\Sigma_v = 0.

References

References

  1. [4]R. A. Sunyaev and Y. B. Zeldovich, The Velocity of clusters of galaxies relative to the microwave background. The Possibility of its measurement, Mon. Not. Roy. Astron. Soc. 190, 413 (1980).
  2. [5]J. P. Ostriker and E. T. Vishniac, Generation of microwave background fluctuations from nonlinear perturbations at the era of galaxy formation, Astrophys. J. Lett. 306, L51 (1986).DOI
  3. [6]K. M. Smith, M. S. Madhavacheril, M. Münchmeyer, S. Ferraro, U. Giri, and M. C. Johnson, Kinetic Sunyaev-Zeldovich tomography and the bispectrum, Phys. Rev. D 113, 083532 (2026), arXiv:1810.13423 [astro-ph.CO].
  4. [7]M. Münchmeyer, M. S. Madhavacheril, S. Ferraro, M. C. Johnson, and K. M. Smith, Constraining local non-Gaussianities with kinetic Sunyaev-Zel’dovich tomography, Phys. Rev. D 100, 083508 (2019), arXiv:1810.13424.
  5. [8]J. Krywonos, S. C. Hotinli, and M. C. Johnson, Constraints on cosmology beyond ΛCDM with kinetic Sunyaev Zel’dovich velocity reconstruction, arXiv e-prints (2024), arXiv:2408.05264 [astro-ph.CO].arxiv.org/abs/2408.05264
  6. [9]A. Laguë, M. S. Madhavacheril, K. M. Smith, S. Ferraro, and E. Schaan, Constraints on Local Primordial Non-Gaussianity with 3D Velocity Reconstruction from the Kinetic Sunyaev-Zeldovich Effect, Phys. Rev. Lett. 134, 151003 (2025), arXiv:2411.08240 [astro-ph.CO].arxiv.org/abs/2411.08240
  7. [10]F. McCarthy et al., The Atacama Cosmology Telescope: Large-scale velocity reconstruction with the kinematic Sunyaev–Zel’dovich effect and DESI LRGs, J. Cosmol. Astropart. Phys. 2025 (05), 057, arXiv:2410.06229 [astro-ph.CO].arxiv.org/abs/2410.06229
  8. [11]S. C. Hotinli, K. M. Smith, and S. Ferraro, Velocity Reconstruction from KSZ: Measuring f_NL with ACT and DESILS, J. Cosmol. Astropart. Phys. 2026 (10), 015, arXiv:2506.21657 [astro-ph.CO].arxiv.org/abs/2506.21657
  9. [12]A. C. M. Lai, Y. Kvasiuk, and M. Münchmeyer, KSZ Velocity Reconstruction with ACT and DESI-LS using a Tomographic QML Power Spectrum Estimator, arXiv e-prints (2025), arXiv:2506.21684 [astro-ph.CO].arxiv.org/abs/2506.21684
  10. [13]F. McCarthy et al., The Atacama Cosmology Telescope: Cross-correlation of kSZ and continuity equation velocity reconstruction with photometric DESI LRGs, J. Cosmol. Astropart. Phys. 2026 (09), 100, arXiv:2511.15701 [astro-ph.CO].arxiv.org/abs/2511.15701
  11. [14]S. S. Gandhi, M. C. Johnson, J. Krywonos, and M. J. Hudson, Measuring cosmic bulk flow with kinetic Sunyaev-Zel’dovich velocity reconstruction, arXiv e-prints (2026), arXiv:2605.12499 [astro-ph.CO].arxiv.org/abs/2605.12499
  12. [15]C. Embil Villagra, F. McCarthy, A. Baleato Lizancos, B. D. Sherwin, and A. Challinor, Estimation and mitigation of foregrounds in projected kSZ velocity reconstruction, arXiv e-prints (2026), arXiv:2603.28746 [astro-ph.CO].arxiv.org/abs/2603.28746
  13. [16]E. Chaussidon et al., Measurement of the galaxy-velocity power spectrum of DESI tracers with the kinematic Sunyaev-Zeldovich effect using DESI DR2 and ACT DR6, arXiv e-prints (2026), arXiv:2604.04867 [astro-ph.CO].arxiv.org/abs/2604.04867
  14. [17]N. Hand et al., Evidence of Galaxy Cluster Motions with the Kinematic Sunyaev-Zel’dovich Effect, Phys. Rev. Lett. 109, 041101 (2012), arXiv:1203.4219.DOI
  15. [18]B. Soergel et al. (DES, SPT), Detection of the kinematic Sunyaev–Zel’dovich effect with DES Year 1 and SPT, Mon. Not. Roy. Astron. Soc. 461, 3172 (2016), arXiv:1603.03904 [astro-ph.CO].DOI
  16. [19]V. Calafut et al., The Atacama Cosmology Telescope: Detection of the pairwise kinematic Sunyaev-Zel’dovich effect with SDSS DR15 galaxies, Phys. Rev. D 104, 043502 (2021), arXiv:2101.08374 [astro-ph.CO].DOI
  17. [20]E. Schiappucci et al. (SPT-3G, DES), Measurement of the mean central optical depth of galaxy clusters via the pairwise kinematic Sunyaev-Zel’dovich effect with SPT-3G and DES, Phys. Rev. D 107, 042004 (2023), arXiv:2207.11937 [astro-ph.CO].DOI
  18. [21]Y. Gong et al., Detection of the pairwise kinematic Sunyaev-Zel’dovich effect and pairwise velocity with DESI DR1 galaxies and ACT DR6 and Planck CMB data, Phys. Rev. D 113, 063538 (2026), arXiv:2511.23417 [astro-ph.CO].arxiv.org/abs/2511.23417
  19. [22]J. C. Hill, S. Ferraro, N. Battaglia, J. Liu, and D. N. Spergel, Kinematic Sunyaev–Zel’dovich Effect with Projected Fields: A Novel Probe of the Baryon Distribution with Planck, WMAP, and WISE Data, Phys. Rev. Lett. 117, 051301 (2016), arXiv:1603.01608 [astro-ph.CO].
  20. [23]S. Ferraro, J. C. Hill, N. Battaglia, J. Liu, and D. N. Spergel, Kinematic Sunyaev-Zel’dovich effect with projected fields. II. Prospects, challenges, and comparison with simulations, Phys. Rev. D 94, 123526 (2016), arXiv:1605.02722 [astro-ph.CO].DOI
  21. [24]A. Kusiak, B. Bolliet, S. Ferraro, J. C. Hill, and A. Krolewski, Constraining the baryon abundance with the kinematic Sunyaev-Zel’dovich effect: Projected-field detection using Planck, WMAP, and unWISE, Phys. Rev. D 104, 043518 (2021), arXiv:2102.01068 [astro-ph.CO].DOI
  22. [25]C. L. Reichardt et al. (SPT), An Improved Measurement of the Secondary Cosmic Microwave Background Anisotropies from the SPT-SZ + SPTpol Surveys, Astrophys. J. 908, 199 (2021), arXiv:2002.06197 [astro-ph.CO].DOI
  23. [26]S. K. Choi et al. (ACT), The Atacama Cosmology Telescope: a measurement of the Cosmic Microwave Background power spectra at 98 and 150 GHz, J. Cosmol. Astropart. Phys. 2020 (12), 045, arXiv:2007.07289 [astro-ph.CO].DOI
  24. [27]P. Chaubal et al. (SPT-3G), SPT-3G D1: A Measurement of Secondary Cosmic Microwave Background Anisotropy Power, arXiv e-prints (2026), arXiv:2601.20551 [astro-ph.CO].arxiv.org/abs/2601.20551
  25. [28]K. M. Smith and S. Ferraro, Detecting Patchy Reionization in the Cosmic Microwave Background, Phys. Rev. Lett. 119, 021301 (2017), arXiv:1607.01769 [astro-ph.CO].
  26. [29]S. Raghunathan et al. (SPT-3G, SPTpol), First Constraints on the Epoch of Reionization Using the Non-Gaussianity of the Kinematic Sunyaev-Zel’dovich Effect from the South Pole Telescope and Herschel-SPIRE Observations, Phys. Rev. Lett. 133, 121004 (2024), arXiv:2403.02337 [astro-ph.CO].DOI
  27. [30]E. Schaan et al. (ACTPol), Evidence for the kinematic Sunyaev-Zel’dovich effect with the Atacama Cosmology Tele- scope and velocity reconstruction from the Baryon Oscillation Spectroscopic Survey, Phys. Rev. D 93, 082002 (2016), arXiv:1510.06442 [astro-ph.CO].
  28. [31]E. Schaan et al. (Atacama Cosmology Telescope), Atacama Cosmology Telescope: Combined kinematic and thermal Sunyaev-Zel’dovich measurements from BOSS CMASS and LOWZ halos, Phys. Rev. D 103, 063513 (2021), arXiv:2009.05557 [astro-ph.CO].
  29. [32]B. Hadzhyiska, S. Ferraro, et al., Evidence for large baryonic feedback at low and intermediate redshifts from kinematic Sunyaev-Zel’dovich observations with ACT and DESI photometric galaxies, Phys. Rev. D 112, 083509 (2025), arXiv:2407.07152 [astro-ph.CO].
  30. [33]B. Ried Guachalla et al., Backlighting extended gas halos around luminous red galaxies: Kinematic Sunyaev-Zel’dovich effect from DESI Y1 and ACT data, Phys. Rev. D 112, 103512 (2025), arXiv:2503.19870 [astro-ph.GA].DOI
  31. [34]F. J. Qu et al., Precision Kinematic Sunyaev–Zel’dovich Measurements Across Halo Mass and Redshift with DESI DR2 and ACT DR6: Part I. Luminous Red Galaxies, arXiv e-prints (2026), arXiv:2604.19744 [astro-ph.CO].arxiv.org/abs/2604.19744
  32. [35]B. Hadzhyiska et al., Precision Kinematic Sunyaev–Zel’dovich Measurements Across Halo Mass and Redshift with DESI DR2 and ACT DR6: Part II. Bright Galaxy Survey and Emission-Line Galaxies, arXiv e-prints (2026), arXiv:2604.19745 [astro-ph.CO].
  33. [36]S. C. Hotinli, The galaxy–electron cross-spectrum with ACT, SPT-3G and DESILS LRGs, arXiv e-prints (2026), arXiv:2609.17676 [astro-ph.CO].arxiv.org/abs/2609.17676
  34. [37]A.-S. Deutsch, E. Dimastrogiovanni, M. C. Johnson, M. Münchmeyer, and A. Terrana, Reconstruction of the remote dipole and quadrupole fields from the kinetic Sunyaev Zel’dovich and polarized Sunyaev Zel’dovich effects, Phys. Rev. D 98, 123501 (2018), arXiv:1707.08129 [astro-ph.CO].DOI
  35. [38]J. Cayuso, R. Bloch, S. C. Hotinli, M. C. Johnson, and F. McCarthy, Velocity reconstruction with the cosmic microwave background and galaxy surveys, J. Cosmol. Astropart. Phys. 2023 (02), 051, arXiv:2111.11526 [astro-ph.CO].DOI
  36. [39]Y. Kvasiuk and M. Münchmeyer, Autodifferentiable likelihood pipeline for the cross-correlation of CMB and large-scale structure due to the kinetic Sunyaev-Zeldovich effect, Phys. Rev. D 109, 083515 (2024), arXiv:2305.08903 [astro-ph.CO].arxiv.org/abs/2305.08903
  37. [40]N. Dalal, O. Dore, D. Huterer, and A. Shirokov, The imprints of primordial non-gaussianities on large-scale structure: scale dependent bias and abundance of virialized objects, Phys. Rev. D 77, 123514 (2008), arXiv:0710.4560 [astro-ph].DOI
  38. [41]U. Seljak, Extracting Primordial Non-Gaussianity without Cosmic Variance, Phys. Rev. Lett. 102, 021302 (2009), arXiv:0807.1770 [astro-ph].DOI
  39. [42]Y. Akrami et al. (Planck), Planck 2018 results. IX. Constraints on primordial non-Gaussianity, Astron. Astrophys. 641, A9 (2020), arXiv:1905.05697 [astro-ph.CO].arxiv.org/abs/1905.05697
  40. [43]E. Chaussidon et al., Constraining primordial non-Gaussianity with DESI 2024 LRG and QSO samples, J. Cosmol. Astropart. Phys. 2025 (06), 029, arXiv:2411.17623 [astro-ph.CO].arxiv.org/abs/2411.17623
  41. [44]R. Bloch and M. C. Johnson, Kinetic Sunyaev Zel’dovich velocity reconstruction from Planck and unWISE, arXiv e-prints (2024), arXiv:2405.00809 [astro-ph.CO].arxiv.org/abs/2405.00809
  42. [45]Y. Omori et al. (DES, SPT), Dark Energy Survey Year 1 Results: Tomographic cross-correlations between Dark Energy Survey galaxies and CMB lensing from South Pole Telescope+Planck, Phys. Rev. D 100, 043501 (2019), arXiv:1810.02342 [astro-ph.CO].DOI
  43. [46]T. M. C. Abbott et al. (DES, SPT), Dark Energy Survey Year 1 Results: Joint Analysis of Galaxy Clustering, Galaxy Lensing, and CMB Lensing Two-point Functions, Phys. Rev. D 100, 023541 (2019), arXiv:1810.02322 [astro-ph.CO].DOI
  44. [47]Y. Omori et al. (DES, SPT), Joint analysis of Dark Energy Survey Year 3 data and CMB lensing from SPT and Planck. I. Construction of CMB lensing maps and modeling choices, Phys. Rev. D 107, 023529 (2023), arXiv:2203.12439 [astro-ph.CO].DOI
  45. [48]T. M. C. Abbott et al. (DES, SPT), Joint analysis of Dark Energy Survey Year 3 data and CMB lensing from SPT and Planck. III. Combined cosmological constraints, Phys. Rev. D 107, 023531 (2023), arXiv:2206.10824 [astro-ph.CO].DOI
  46. [49]S. Bocquet et al. (DES, SPT), SPT clusters with DES and HST weak lensing. I. Cluster lensing and Bayesian population modeling of multiwavelength cluster datasets, Phys. Rev. D 110, 083509 (2024), arXiv:2310.12213 [astro-ph.CO].arxiv.org/abs/2310.12213
  47. [50]S. Bocquet et al. (DES, SPT), Multiprobe cosmology from the abundance of SPT clusters and DES galaxy clustering and weak lensing, Phys. Rev. D 111, 063533 (2025), arXiv:2412.07765 [astro-ph.CO].arxiv.org/abs/2412.07765
  48. [51]S. Raghunathan et al. (SPT, DES), Detection of CMB-Cluster Lensing using Polarization Data from SPTpol, Phys. Rev. Lett. 123, 181301 (2019), arXiv:1907.08605 [astro-ph.CO].DOI
  49. [52]B. Ansarinejad et al. (SPT-3G, DES), Mass calibration of DES Year-3 clusters via SPT-3G CMB cluster lensing, J. Cosmol. Astropart. Phys. 2024 (07), 024, arXiv:2404.02153 [astro-ph.CO].
  50. [53]W. Quan et al. (SPT-3G), SPT-3G D1: Maps of the millimeter-wave sky from 2019 and 2020 observations of the SPT-3G Main field, arXiv e-prints (2026), arXiv:2603.20163 [astro-ph.CO].arxiv.org/abs/2603.20163
  51. [54]R. Zhou et al., DESI luminous red galaxy samples for cross-correlations, J. Cosmol. Astropart. Phys. 2023 (11), 097, arXiv:2309.06443 [astro-ph.CO].DOI
  52. [55]A. Slosar, C. Hirata, U. Seljak, S. Ho, and N. Padmanabhan, Constraints on local primordial non-Gaussianity from large scale structure, J. Cosmol. Astropart. Phys. 2008 (08), 031, arXiv:0805.3580 [astro-ph].DOI
  53. [56]A. Barreira, On the impact of galaxy bias uncertainties on primordial non-Gaussianity constraints, J. Cosmol. Astropart. Phys. 2020 (12), 031, arXiv:2009.06622 [astro-ph.CO].DOI
  54. [57]A. Barreira, G. Cabass, F. Schmidt, A. Pillepich, and D. Nelson, Galaxy bias and primordial non-Gaussianity: insights from galaxy formation simulations with IllustrisTNG, J. Cosmol. Astropart. Phys. 2020 (12), 013, arXiv:2006.09368 [astro-ph.CO].
  55. [58]N. Dalal and W. J. Percival, Estimating non-gaussian bias using counts of tracers, JCAP 04, 037, arXiv:2503.21024 [astro-ph.CO].arxiv.org/abs/2503.21024
  56. [59]J. M. Sullivan and U. Seljak, Local primordial non-Gaussian bias from time evolution, Phys. Rev. D 112, 083522 (2025), arXiv:2503.21736 [astro-ph.CO].arxiv.org/abs/2503.21736
  57. [60]R. Sunyaev and Y. B. Zeldovich, The Observations of Relic Radiation as a Test of the Nature of X-Ray Radiation from the Clusters of Galaxies, Comments Astrophys. Space Phys. 4, 173 (1972).
  58. [61]M. G. Hauser and E. Dwek, The Cosmic Infrared Background: Measurements and Implications, Annu. Rev. Astron. Astrophys. 39, 249 (2001), arXiv:astro-ph/0105539.DOI
  59. [62]U. Giri and K. M. Smith, Exploring KSZ velocity reconstruction with N-body simulations and the halo model, J. Cosmol. Astropart. Phys. 2022 (09), 028, arXiv:2100.07193 [astro-ph.CO].DOI
  60. [63]N. Hand, Y. Li, Z. Slepian, and U. Seljak, An optimal FFT-based anisotropic power spectrum estimator, J. Cosmol. Astropart. Phys. 2017 (07), 002, arXiv:1704.02357 [astro-ph.CO].DOI
  61. [64]R. Scoccimarro, Fast Estimators for Redshift-Space Clustering, Phys. Rev. D 92, 083532 (2015), arXiv:1506.02729 [astro-ph.CO].DOI
  62. [65]A. de Mattia and V. Ruhlmann-Kleider, Integral constraints in spectroscopic surveys, J. Cosmol. Astropart. Phys. 2019 (08), 036, arXiv:1904.08851 [astro-ph.CO].
  63. [66]R. K. Sachs and A. M. Wolfe, Perturbations of a Cosmological Model and Angular Variations of the Microwave Background, Astrophys. J. 147, 73 (1967).
  64. [67]M. J. Rees and D. W. Sciama, Large-scale Density Inhomogeneities in the Universe, Nature 217, 511 (1968).DOI
  65. [68]M. Birkinshaw and S. F. Gull, A test for transverse motions of clusters of galaxies, Nature 302, 315 (1983).DOI
  66. [69]S. C. Hotinli, J. Meyers, N. Dalal, A. H. Jaffe, M. C. Johnson, J. B. Mertens, M. Münchmeyer, K. M. Smith, and A. van Engelen, Transverse Velocities with the Moving Lens Effect, Phys. Rev. Lett. 123, 061301 (2019), arXiv:1812.03167 [astro-ph.CO].DOI
  67. [70]S. C. Hotinli, K. M. Smith, S. Ferraro, A. Beheshti, A. Kosowsky, E. Pierpaoli, and E. Schaan, First detection of the moving lens effect with ACT and DESI LSS, arXiv e-prints (2026), arXiv:2605.18938 [astro-ph.CO].arxiv.org/abs/2605.18938
  68. [71]Planck Collaboration, Planck intermediate results. XLVIII. Disentangling Galactic dust emission and cosmic infrared background anisotropies, Astron. Astrophys. 596, A109 (2016), arXiv:1605.09387 [astro-ph.CO].
  69. [72]A. Dey et al., Overview of the DESI Legacy Imaging Surveys, Astron. J. 157, 168 (2019), arXiv:1804.08657 [astro-ph.IM].DOI
  70. [73]The Dark Energy Survey Collaboration, The Dark Energy Survey, arXiv e-prints (2005), astro-ph/0510346.arxiv.org/abs/astro-ph/0510346
  71. [74]E. Camphuis et al. (SPT-3G), SPT-3G D1: CMB temperature and polarization power spectra and cosmology from 2019 and 2020 observations of the SPT-3G main field, Phys. Rev. D 113, 083504 (2026), arXiv:2506.20707 [astro-ph.CO].DOI
  72. [75]J. A. Sobrin et al. (SPT-3G), The Design and Integrated Performance of SPT-3G, Astrophys. J. Supp. 258, 42 (2022), arXiv:2106.11202 [astro-ph.IM].DOI
  73. [76]J. E. Carlstrom et al., The 10 Meter South Pole Telescope, Publ. Astron. Soc. Pac. 123, 568 (2011), arXiv:0907.4445 [astro-ph.IM].DOI
  74. [77]K. M. Górski, E. Hivon, A. J. Banday, B. D. Wandelt, F. K. Hansen, M. Reinecke, and M. Bartelmann, HEALPix: A Framework for High-Resolution Discretization and Fast Analysis of Data Distributed on the Sphere, Astrophys. J. 622, 759 (2005), arXiv:astro-ph/0409513 [astro-ph].DOI
  75. [78]M. Archipley et al. (SPT-3G), Millimeter-wave observations of Euclid Deep Field South using the South Pole Telescope - A data release of temperature maps and catalogs, Astron. Astrophys. 706, A17 (2026), arXiv:2506.00298 [astro-ph.GA].DOI
  76. [79]R. Zhou et al. (DESI), Target Selection and Validation of DESI Luminous Red Galaxies, Astron. J. 165, 58 (2023), arXiv:2208.08515 [astro-ph.CO].DOI
  77. [80]M. White et al., Cosmological constraints from the tomographic cross-correlation of DESI Luminous Red Galaxies and Planck CMB lensing, J. Cosmol. Astropart. Phys. 2022 (02), 007, arXiv:2111.09898 [astro-ph.CO].DOI
  78. [81]H. A. Feldman, N. Kaiser, and J. A. Peacock, Power spectrum analysis of three-dimensional redshift surveys, Astrophys. J. 426, 23 (1994), arXiv:astro-ph/9304022.DOI
  79. [82]D. Foreman-Mackey, D. W. Hogg, D. Lang, and J. Goodman, emcee: The MCMC Hammer, Publ. Astron. Soc. Pac. 125, 306 (2013), arXiv:1202.3665 [astro-ph.IM].
  80. [83]M. Rezaie et al., Local primordial non-Gaussianity from the large-scale clustering of photometric DESI luminous red galaxies, Mon. Not. Roy. Astron. Soc. 532, 1902 (2024), arXiv:2307.01753 [astro-ph.CO].DOI
  81. [85]E. Chaussidon et al., Angular clustering properties of the DESI QSO target selection using DR9 Legacy Imaging Surveys, Mon. Not. Roy. Astron. Soc. 509, 3904 (2022), arXiv:2108.03640.DOI
  82. [86]N. Battaglia, The tau of galaxy clusters, J. Cosmol. Astropart. Phys. 2016 (08), 058, arXiv:1607.02442.DOI
  83. [87]N. Anil Kumar, G. Sato-Polito, M. Kamionkowski, and S. C. Hotinli, Primordial trispectrum from kinetic Sunyaev-Zel’dovich tomography, Phys. Rev. D 106, 063533 (2022), arXiv:2205.03423 [astro-ph.CO].
  84. [88]P. Adshead and A. J. Tishue, Probing beyond local-type non-Gaussianity with kinematic Sunyaev-Zel’dovich tomography, Phys. Rev. D 110, 103549 (2024), arXiv:2407.21094 [astro-ph.CO].arxiv.org/abs/2407.21094
  85. [89]S. C. Hotlini, J. B. Mertens, M. C. Johnson, and M. Kamionkowski, Probing correlated compensated isocurvature perturbations using scale-dependent galaxy bias, Phys. Rev. D 100, 103528 (2019), arXiv:1908.08953 [astro-ph.CO].
  86. [90]N. A. Kumar, S. C. Hotlini, and M. Kamionkowski, Uncorrelated compensated isocurvature perturbations from kinetic Sunyaev-Zeldovich tomography, Phys. Rev. D 107, 043504 (2023), arXiv:2208.02829 [astro-ph.CO].DOI
  87. [91]P. Zhang and M. C. Johnson, Testing eternal inflation with the kinetic Sunyaev Zel’dovich effect, J. Cosmol. Astropart. Phys. 2015 (06), 046, arXiv:1501.00511 [astro-ph.CO].arxiv.org/abs/1501.00511
  88. [92]E. Vanzan, M. Kamionkowski, and S. C. Hotlini, Phenomenology of a vector-field-induced and possibly parity breaking compensated isocurvature perturbation, Phys. Rev. D 109, 083510 (2024), arXiv:2311.18121 [astro-ph.CO].DOI
  89. [93]J. I. Cayuso and M. C. Johnson, Towards testing CMB anomalies using the kinetic and polarized Sunyaev-Zel’dovich effects, Phys. Rev. D 101, 123508 (2020), arXiv:1904.10981 [astro-ph.CO].DOI
  90. [94]J. Adolf, S. Hotlini, and N. Dalal, Probing Dark Energy Microphysics with kSZ Tomography, Open J. Astrophys. 9 (2026), arXiv:2511.05653.
  91. [95]Z. Pan and M. C. Johnson, Forecasted constraints on modified gravity from Sunyaev-Zel’dovich tomography, Phys. Rev. D 100, 083522 (2019), arXiv:1906.04208 [astro-ph.CO].DOI
  92. [96]A. J. Tshue, S. C. Hotlini, P. Dadshead, E. D. Kovetz, and M. S. Madhavacheril, Neutrino mass constraints from kinetic Sunyaev Zel’dovich tomography, Phys. Rev. D 111, 123556 (2025), arXiv:2502.05260 [astro-ph.CO].
  93. [97]D. Contreras, M. C. Johnson, and J. B. Mertens, Towards detection of relativistic effects in galaxy number counts using kSZ tomography, J. Cosmol. Astropart. Phys. 2019 (10), 024, arXiv:1904.10133 [astro-ph.CO].
  94. [98]S. C. Hotlini and M. C. Johnson, Reconstructing large scales at cosmic dawn, Phys. Rev. D 105, 063522 (2022), arXiv:2012.09851 [astro-ph.CO].
  95. [99]P. A. Abell et al. (LSST Science, LSST Project), LSST Science Book, Version 2.0, arXiv e-prints (2009), arXiv:0912.0201 [astro-ph.IM].
  96. [100]L. Amendola et al., Cosmology and fundamental physics with the Euclid satellite, Living Rev. Rel. 21, 2 (2018), arXiv:1606.00180 [astro-ph.CO].DOI

Paper details

vlnc:W_rzpqhcbr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents