Mass calibration of DES Year-3 clusters via SPT-3G CMB cluster lensing

Abstract

We measure the stacked lensing signal in the direction of galaxy clusters in the Dark Energy Survey Year 3 (DES Y3) redMaPPer sample, using cosmic microwave background (CMB) temperature data from SPT-3G, the third-generation CMB camera on the South Pole Telescope (SPT). Here, we estimate the lensing signal using temperature maps constructed from the initial 2 years of data from the SPT-3G ‘Main’ survey, covering 1500 deg² of the Southern sky. We then use this lensing signal as a proxy for the mean cluster mass of the DES sample. The thermal Sunyaev–Zel’dovich (tSZ) signal, which can contaminate the lensing signal if not addressed, is isolated and removed from the data before obtaining the mass measurement. In this work, we employ three versions of the redMaPPer catalogue: a Flux-Limited sample containing 8865 clusters, a Volume-Limited sample with 5391 clusters, and a Volume&Redshift-Limited sample with 4450 clusters. For the three samples, we detect the CMB lensing signal at a significance of $10.3\sigma$σしぐま, $10.0\sigma$σしぐま, and $9.6\sigma$σしぐま and find the mean cluster masses to be ${M}_{200{\rm{m}}}=1.66\pm 0.13$ [stat.]$\pm 0.03$ [sys.], $1.97\pm 0.18$ [stat.]$\pm 0.05$ [sys.], and $2.11\pm 0.20$ [stat.]$\pm 0.05$ [sys.]$\times{10}^{14}\ {\rm{M}}_{\odot}$, respectively. This is a factor of $\sim 2$ improvement relative to the precision of measurements with previous generations of SPT surveys and the most constraining cluster mass measurements using CMB cluster lensing to date. Overall, we find no significant tensions between our results and masses given by redMaPPer mass–richness scaling relations of previous works, which were calibrated using CMB cluster lensing, optical weak lensing, and velocity dispersion measurements from various combinations of DES, SDSS and Planck data. We then divide our sample into 3 redshift and 3 richness bins, finding no significant discrepancies with optical weak-lensing calibrated masses in these bins. We forecast a $5.8\%$ constraint on the mean cluster mass of the DES Y3 sample with the complete SPT-3G surveys when using both temperature and polarization data and including an additional $\sim 1400$ deg² of observations from the ‘Extended’ SPT-3G survey.

FERMILAB-PUB-24-0123

DES-2023-0821

1 Introduction

Galaxy clusters are the most massive gravitationally collapsed objects and are the culmination of structure growth processes across cosmic time. As a result, cluster number counts as a function of cluster mass and redshift provide a sensitive probe of cosmological parameters that influence the growth of structure and the geometry of the Universe [see reviews by 1, 2]. These parameters include the matter density parameter, ${\mathrm{<ruby><rb>\Omega </rb><rt>おめが</rt></ruby>}}_{\text{m}}$Ωおめがm\Omega_{\textup{m}}roman_Ωおめが start_POSTSUBSCRIPT m end_POSTSUBSCRIPT; the normalisation of the matter power spectrum on the scale of $8\ h^{-1}$Mpc, $\sigma_{8}$σしぐま start_POSTSUBSCRIPT 8 end_POSTSUBSCRIPT; the dark energy equation of state parameter, $w$; as well as the sum of neutrino masses, $\sum m_{\nu}$νにゅー end_POSTSUBSCRIPT [see e.g. 3, 4, 5, 6]. These constraints are highly complementary to those derived from analyses of baryon acoustic oscillations (BAO; [7, 8]), cosmic microwave background (CMB; [9, 10, 11]), as well as auto- and cross-correlation analyses of optical weak gravitational lensing and galaxy clustering (3x2pt; [12, 13, 14]) as these measurements have different parameter degeneracies and independent sources of systematics.

However, cosmological analysis of galaxy cluster samples is currently limited by our ability to reconstruct the mass distribution of the cluster sample (a problem called mass calibration; see e.g. Section VI of [15] and Section 4 of [5], for a discussion of the impact of systematics on recent cluster cosmology analyses). In the near future, surveys such as eROSITA [16], LSST [17], Euclid [18] and CMB-S4 [19] will increase the cluster sample size compared to existing surveys by an order of magnitude, significantly reducing limitations due to statistical uncertainties. In preparation for these datasets, it is, therefore, crucial to improve our understanding of sources of systematic uncertainty that could impact commonly used cluster detection and mass calibration methods. In the optical regime, weak gravitational lensing is the most common cluster mass measurement approach [see 20, for a review]. Weak lensing offers the advantage of probing the total cluster mass with weak dependence on complex baryonic physics, which could affect the mass-observable scaling relations of the thermal Sunyaev–Zel’dovich (tSZ) decrement, X-ray luminosity, and cluster richness. Weak lensing is, however, impacted by various sources of systematics error including bias in photometric redshift estimates, galaxy shape modelling errors and contamination of the lensed galaxy sample with cluster member or foreground galaxies [21, 22, 23]. Calibration of these effects has led to a systematic floor smaller than $2\%$ on the halo mass [24, 25] if source galaxies can be reliably selected in the background of the cluster sample.

CMB cluster lensing is a promising alternative technique for measuring the masses of galaxy clusters. In this phenomenon, CMB photons passing through galaxy clusters’ gravitational potential wells are deflected and due to the small-scale CMB gradient, form arcminute scale dipoles with amplitudes of $\lesssim 10\mu{\rm K}$μみゅー roman_K [26, 27]. Measurements of these dipoles can, therefore, be used as a proxy for the cluster mass. Furthermore, CMB cluster lensing and optical weak lensing have mostly independent systematics (though projection effects will impact both observables; see Section 4.2.5 for details), enabling us to verify whether these systematics have been correctly characterised and accounted for. Additionally, since the source plane of CMB lensing is the surface of last scattering at $z\sim 1100$, mass calibration can also be carried out for higher cluster redshifts, where optical lensing starts to suffer from unreliable background source selection. This makes CMB cluster lensing an essential tool for cluster mass measurements in upcoming datasets such as CMB-S4, which greatly increase the size of high redshift cluster samples by detecting thousands of clusters at $z>1$ [28].

Over the past two decades, several different methods have been developed to measure the CMB cluster and galaxy lensing signal from CMB temperature and polarization maps [29, 30, 31, 27, 32, 33, 34, 35, 36]. In recent years, a number of studies have obtained the first significant detections of CMB cluster lensing using data from various CMB experiments. Using CMB temperature data from the SPT-SZ survey conducted with the South Pole Telescope (SPT) and a sample of 513 clusters [37] detected with the same data, [38] obtained a $3.1\sigma$σしぐま measurement of the CMB cluster lensing signal. Using data from the Atacama Cosmology Telescope (ACT) and various galaxy samples, [39] and [40] obtained a $3.2\sigma$σしぐま and $4.2\sigma$σしぐま detection of the signal, respectively. Similar studies using the Planck CMB data include [5] and [41], where the lensing signal was measured at $\sim 5\sigma$σしぐま for SZ-detected galaxy cluster samples, and [42], where a sample of 12.4 million galaxies selected from the WISE and SCOS surveys were used as tracers of dark matter halos, obtaining a $17\sigma$σしぐま measurement of the lensing signal.

Past studies have also applied CMB cluster lensing to calibrate the scaling relation between cluster mass and richness for different cluster samples detected with the redMaPPer algorithm [43]. [44] obtained a $10\%$ constraint on the richness–mass scaling relation using the Planck CMB data and optically detected clusters in the Sloan Digital Sky Survey (SDSS) data presented by [43]. Later, [45] used the SPT-SZ CMB temperature map to obtain a $17\%$ constraint on the amplitude of the mass–richness scaling relation of redMaPPer clusters detected in the Dark Energy Survey [DES 46] Year 1 data [47], while [48] obtained a $\sim 20\%$ measurement of the same scaling relation for the DES Year 3 (Y3) redMaPPer cluster sample, using CMB temperature maps from the SPTpol survey [49]. [35] also used the same datasets to obtain the first detection of the CMB cluster lensing signal using only the polarization data, obtaining a $\sim 28\%$ mass constraint for richness $\lambda>10$λらむだ > 10 clusters in the DES Y3 sample.

One of the main challenges in measuring the CMB cluster lensing signal is overcoming contamination due to various astrophysical foregrounds which could bias the cluster mass measurements if they are not accounted for. These sources of contamination include the tSZ and kinetic SZ (kSZ) effects, as well as the cosmic infrared background (CIB) due to the presence of dusty galaxies in the clusters (see [50] for a comprehensive discussion of the impact of these systematics on CMB cluster lensing measurements). Various techniques have been developed to overcome these sources of contamination, including cleaning the large-scale CMB gradient using a Quadratic Estimator (QE) to overcome the tSZ bias, as proposed by [51]. [52] later developed a modified QE that overcame tSZ, kSZ, and CIB contamination by inpainting the large-scale gradient in the CMB maps to remove the cluster emission. [35] presented a new estimator to measure the lensing dipole of stacked images, based on rotating cluster-centred CMB map cutouts along the direction of locally measured background CMB gradients. This approach is much simpler and less computationally expensive than the other alternative techniques and we adopt this method for measuring CMB cluster lensing in this study.

The layout of this paper is as follows. We present a summary of the galaxy cluster sample and the CMB maps used in our analysis in Section 2, followed by a description of our methods and pipeline verification with simulations in Section 3. We then present our results and compare them with various other DES cluster mass measurements from the literature in Section 4. We conclude by presenting a summary of our findings and their implications in Section 5. Throughout, we assume a $\mathrm{<ruby><rb>\Lambda </rb><rt>らむだ</rt></ruby>}$Λらむだ\Lambdaroman_ΛらむだCDM cosmology with $h=0.6774$, ${\mathrm{<ruby><rb>\Omega </rb><rt>おめが</rt></ruby>}}_m=0.307$Ωおめが𝑚0.307\Omega_{m}=0.307roman_Ωおめが start_POSTSUBSCRIPT italic_m end_POSTSUBSCRIPT = 0.307, and ${\mathrm{<ruby><rb>\Omega </rb><rt>おめが</rt></ruby>}}_{\mathrm{<ruby><rb>\Lambda </rb><rt>らむだ</rt></ruby>}}=0.693$ΩおめがΛらむだ0.693\Omega_{\Lambda}=0.693roman_Ωおめが start_POSTSUBSCRIPT roman_Λらむだ end_POSTSUBSCRIPT = 0.693. In this work, we express cluster masses as $M_{\rm 200m}$, defined as the mass enclosed within a sphere whose average density is 200 times that of the mean matter density of the Universe, $\bar{\rho}_{m}$ρろー end_ARG start_POSTSUBSCRIPT italic_m end_POSTSUBSCRIPT, at the cluster redshift.

Figure 1: The $\sim 1400$ deg² overlap between the $\sim 1500$ deg² SPT-3G Main field (blue) and the $\sim 5000$ deg² DES (orange) survey footprints. The grey band marks the galactic plane.

2 Datasets

In this section, we provide a brief description of the datasets and sample selection used in our CMB cluster lensing analysis.

2.1 SPT-3G CMB data

SPT is a 10-meter telescope located at the Amundsen-Scott South Pole station [53], optimised for low-noise observations of the temperature and polarization of the CMB. SPT-3G [54] is the third and latest receiver installed on the telescope, with the SPT-3G Main field covering a $\sim 1500$ deg² footprint defined by $310^{\circ}<{\rm RA}<50^{\circ}$ and $-70^{\circ}<{\rm DEC}<-42^{\circ}$. After masking point sources in the CMB data (see Section 2.3) and taking into account the masked area in the DES cluster sample, we are left with an overlap of $\sim 1350$ deg² (see Figure 1) between the SPT-3G survey and the DES cluster catalogue. In this work, we use data from the initial two years of the SPT-3G survey observation (2019-2020). While the analysis of this paper is only performed using temperature data, we note that the inclusion of the initial two years of SPT-3G polarization data is expected to reduce the mass uncertainty by $\sim 10\%$. We leave the polarization measurement to future works and provide forecasts for mass constraints using temperature and polarization data from the full survey depth, as well as data from the SPT-3G ‘Extended’ survey (which provides an additional $\sim 1400$ deg² of overlap with DES albeit at a lower sensitivity) in Section 4.6. In addition to the Main and Extended SPT-3G surveys, an additional $\sim 6000$ deg² of the Southern sky will be observed for one year in the SPT-3G ‘Wide’ survey. However, due to the relatively small additional overlap area with DES and the higher noise levels of these data, the Wide survey observations will not provide a significant improvement in the S/N of the CMB cluster lensing measurements, and we do not include these in our forecasts.

SPT-3G Main survey observations are conducted in the 95, 150, and 220 GHz bands with $1.6^{\prime},1.2^{\prime}$, and $1.0^{\prime}$ full width at half maximum beams and white noise levels of $\sim 5$, 4, and 15 $\mu{\rm K-arcmin}$μみゅー roman_K - roman_arcmin for the first two years of observations, in the three bands respectively. For each SPT-3G detector, the raw data are composed of digitised time-ordered data (TOD) that are converted to CMB temperature units (for details of the SPT-3G map making and data processing, see [55]). During map making, we apply a $300<\ell_{x}<13000$ bandpass filter to the TOD. In this analysis, we use maps based on a minimum-variance combination of the 95, 150, and 220 GHz data, with Sanson-Flamsteed flat-sky projection [56] and $0.5^{\prime}$ pixels.

2.2 tSZ-nulled SPT-3G map

As described in Section 3.1 our lensing estimator requires the estimation of the local CMB gradient at the location of each galaxy cluster. For this purpose, we use a tSZ-nulled CMB map constructed by performing an internal linear combination (ILC) of 95, 150, and 220 GHz data. As this map does not contain the tSZ signal, it allows for a more accurate estimation of the CMB gradient direction and amplitude, which in turn improves the S/N of our lensing dipole measurement.

2.3 DES Y3 redMaPPer galaxy clusters

DES has a $\sim 5000$ deg² footprint with imaging taken in the $g,r,i,z,$ and $Y$ bands via the Dark Energy Camera [57] installed on the 4m Blanco telescope at the Cerro Tololo Observatory. The survey has completed the sixth and final year of observations, and here we use cluster samples detected using data from the first 3 years of the survey [58]. We refer the reader to [59] for a description of the application of the redMaPPer algorithm to the DES survey. In this work, we perform our analysis based on three redMaPPer cluster samples: a Flux-Limited sample, a Volume-Limited sample, and the Volume-Limited sample limited to the redshift range $0.2<z<0.65$. Henceforth, we shall refer to the latter as the Volume&Redshift-Limited sample. While the Flux-Limited sample contains a significantly higher number of clusters at $z>0.65$, which would yield a lensing measurement with a higher S/N, the Volume-Limited sample is limited to varying redshifts which are determined based on the magnitude limit of the observations across the survey footprint. The Volume&Redshift-Limited sample is then created to ensure sample uniformity across the survey footprint and match the selection applied to the sample used for the DES Y3 cluster cosmology analysis.

In all cases, samples are limited to clusters with richness $\lambda\geq 20$λらむだ ≥ 20. Upon masking clusters within $1^{\circ}$ of the edges of the SPT-3G footprint and within 10^′ of bright point sources ($\geq 6$ mJy at 150GHz), we are left with 4450, 5391, and 8865 clusters in the Volume&Redshift-Limited, Volume-Limited, and Flux-Limited samples, respectively. The photometric redshift and richness distributions of the three samples are shown in Figure 2. The median photometric redshifts of the Volume- and Flux-Limited samples are $\sim 0.47$ and $\sim 0.61$, with median uncertainties of $\sigma_{z}/(1+z)=0.006$σしぐま start_POSTSUBSCRIPT italic_z end_POSTSUBSCRIPT / ( 1 + italic_z ) = 0.006 and $0.008$, respectively.

To investigate the potential redshift or richness dependence of the cluster mass–richness scaling relation, we divide the Volume&Redshift-Limited sample into 3 redshift and richness bins with an approximately equal number of clusters, as shown in Table 1.

Figure 2: Left panel: The photometric redshift distribution of the DES Y3 flux and Volume-Limited redMaPPer cluster samples. The vertical dashed lines indicate the redshift cuts applied to the Volume-Limited sample. Here, we can see that most additional clusters in the Flux-Limited sample relative to the Volume-Limited sample are at $z>0.6$. Right panel: The richness distribution of the Flux-, Volume- and Volume&Redshift-Limited samples. Although Volume-Limited samples contain fewer clusters, they follow a richness distribution similar to the Flux-Limited sample.

Table 1: The number of clusters in the 3 redshift and richness subsamples of our Volume&Redshift-Limited sample. The bins were chosen to have approximately equal numbers of clusters.

bin definition	clusters per bin
$0.20<z<0.40$	$1477$
$0.40<z<0.53$	$1480$
$0.53<z<0.65$	$1493$
$20<\lambda<24$λらむだ < 24	$1491$
$24<\lambda<32$λらむだ < 32	$1530$
$32<\lambda$λらむだ	$1429$

3 Pipeline description and validation

3.1 Lensing estimator

We adopt the lensing estimator introduced by [35], which is briefly described here. Using the tSZ-nulled map described in Section 2.2, the algorithm first extracts $60^{\prime}\times 60^{\prime}$ cutouts, centred on the location of $N_{\textup{clus}}$ clusters and $N_{\textup{rands}}$ random locations. The code then determines the median gradient direction $\theta_{\triangledown}=\tan^{-1}(\triangledown_{y}/\triangledown_{x})$θしーた start_POSTSUBSCRIPT ▽ end_POSTSUBSCRIPT = roman_tan start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ( ▽ start_POSTSUBSCRIPT italic_y end_POSTSUBSCRIPT / ▽ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT ) from the central $6^{\prime}\times 6^{\prime}$ region of each cutout. The noise penalty in the gradient estimation is reduced by applying a Wiener filter given by:

$$W_{\ell}=\begin{cases}C_{\ell}(C_{\ell}+N_{\ell})^{-1},&\ell\leq 2000\\ 0,&\text{otherwise}\\ \end{cases},$$ (3.1)

to the $60^{\prime}\times 60^{\prime}$ cutouts, where $C_{\ell}$ and $N_{\ell}$ are the data and noise power spectra, with the latter calculated using half-difference maps.

Central $10^{\prime}\times 10^{\prime}$ cutouts, ${\mathbf{d}}$, are then extracted from the SPT-3G map and rotated along the direction of the gradients, allowing for the stacking of the lensing dipoles which are oriented along the direction of the local CMB gradient. At this stage, a weight is assigned to each cluster given by $w=w_{n}w_{g}$, where the $w_{n}$ component is based on the inverse noise variance $\sigma^{2}$σしぐま start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT at the location of the cluster. The weight, $w_{g}$, is based on the median magnitude of the local gradient $\sqrt{\triangledown^{2}_{x}+\triangledown^{2}_{y}}$, which serves to maximize the $S/N$ of the measured dipole amplitude given its proportionality to the gradient amplitude.

The cutouts are then mean subtracted and stacked to obtain the weighted stacks $\mathbf{s_{c}}$ and $\mathbf{s_{r}}$ at the location of clusters and random points, respectively. $\mathbf{s_{c}}$ is dominated by the mean large-scale CMB gradient (henceforth we refer to this as the background), which is estimated by $\mathbf{s_{r}}$ and corrected to obtain the final stacked dipole. To ensure that our background estimation is not biased due to sample variance, we set the value of $N_{\textup{rands}}=10\times N_{\textup{clus}}$. Here, we use the DES Y3 random catalogues which ensure that random points do not fall in the masked DES regions. The final stacked dipole is given by:

$$\mathbf{s}\equiv\mathbf{s_{c}}-\mathbf{s_{r}}=\frac{\sum^{N_{\textup{clus}}}_{c}w_{c}[\mathbf{d_{c}}-\langle\mathbf{d_{c}}\rangle]}{\sum^{N_{\textup{clus}}}_{c}w_{c}}-\frac{\sum^{N_{\textup{rand}}}_{r}w_{r}[\mathbf{d_{r}}-\langle\mathbf{d_{r}}\rangle]}{\sum^{N_{\textup{rand}}}_{r}w_{r}},$$ (3.2)

where $\mathbf{s}$ contains the dipole signal along with noise contributions from astrophysical and atmospheric foregrounds, instrumental noise, residual large-scale CMB gradient, the kSZ effect and in the case of temperature maps, the tSZ effect.

To remove the tSZ contribution from the stack (which could cause a bias toward lower masses if not taken into account), we follow the approach of [60] and rotate the cluster cutouts in random orientations prior to stacking. To ensure an accurate estimate of the tSZ contamination, we repeat this procedure 25 times and take the mean of the 25 stacks as our estimate of the tSZ signal. Fig. 3 illustrates the main steps involved in removing tSZ, with panel (a) showing our original stack (with tSZ contamination visible in the central pixels of the image), while panel (b) shows the mean of our 25 ‘random rotation’ stacks with an estimate of tSZ contamination. Finally, panel (c) shows our tSZ-free cluster stack which is obtained by subtracting the tSZ signal shown in panel (b) from panel (a).

Figure 3: (a) The rotated and weighted cluster stack $\mathbf{s_{c}}$ from our minimum variance temperature map, including the tSZ contamination. (b) The mean of 25 randomly rotated cluster stacks, with the tSZ signal visible at the centre of the cutout. (c) Panel (a)-(b): the final data stack after removing tSZ contamination.

3.2 Lensing dipole models

To create the lensing dipole models $\mathbf{m}\equiv\mathbf{m}(M)$, we generate noiseless cluster-lensed simulations for a set of $N_{\textup{clus}}$ clusters with the redshift distribution of the DES sample and cluster masses varying in the range $M\in[0,4]\times 10^{14}$$\rm{M}_{\odot}$ with linear bins of $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}M=0.1\times {10}^{14}$Δでるた𝑀0.1superscript1014\Delta M=0.1\times 10^{14}roman_Δでるた italic_M = 0.1 × 10 start_POSTSUPERSCRIPT 14 end_POSTSUPERSCRIPT$\rm{M}_{\odot}$. For each mass bin, the $N_{\textup{clus}}$ cutouts are then stacked, following the steps in the previous section. Here, the mean background is simply given by $\mathbf{m_{r}}\equiv\mathbf{m_{c}}(M=0)$ and is subtracted from the stacks in all mass bins. Since the uncertainties on the gradient direction ($\delta\theta_{\triangledown}$δでるた italic_θしーた start_POSTSUBSCRIPT ▽ end_POSTSUBSCRIPT) measurement will be lower in the case of noiseless simulations relative to the real data, the suboptimal stacking of the lensing dipole in the data compared to the models will cause a bias towards lower masses in the likelihoods, if not corrected for. For this reason, we add white noise and Gaussian foregrounds (mimicking those present in the data) to the models only when measuring $\delta\theta_{\triangledown}$δでるた italic_θしーた start_POSTSUBSCRIPT ▽ end_POSTSUBSCRIPT. In Section 3.4, we provide a more comprehensive description of the generated simulations, including the prescription for adding the impact of cluster miscentering and correlated structure to our modelled lensing profiles.

3.3 Stacked cluster mass likelihood

Equipped with the stacked dipole signal, $\mathbf{s}$, and the models, $\mathbf{m}$, we calculate the likelihood using

$$-2\ln{{\cal L}(M|\mathbf{s})}=\sum_{\textup{pixels}}(\mathbf{s}-\mathbf{m})\hat{\mathbf{C}}^{-1}(\mathbf{s}-\mathbf{m})^{T},$$ (3.3)

where $\hat{\mathbf{C}}$ is the covariance matrix which is estimated from the data using the jackknife resampling technique by dividing the data into $N_{\textup{jk}}=0.9N_{\textup{clus}}$ subsamples:

$$\hat{\mathbf{C}}=\frac{N_{\textup{jk}}-1}{N_{\textup{jk}}}\sum^{N_{\textup{jk}}}_{i=1}[\mathbf{s}_{i}-\langle\mathbf{s}\rangle][\mathbf{s}_{i}-\langle\mathbf{s}\rangle]^{T}.$$ (3.4)

Here, $\mathbf{s}_{i}$ is the data stack in the $i$-th subsample and $\langle\mathbf{s}\rangle$ is the ensemble average of all subsamples. Estimating the covariance matrix from the data offers the advantage of capturing all sources of noise impacting the stacked lensing signal. When computing the likelihoods for our redshift and richness subsamples, which contain roughly one-third of the clusters from our Volume&Redshift-Limited sample, we adopt the covariance matrix derived from the latter and multiply it by a factor of 3 to account for the increased shot noise in the subsamples. This approach sidesteps potential inaccuracies in estimating the covariance matrix that could result from using the smaller cluster samples from our redshift and richness subsamples.

3.4 Simulations & pipeline validation

In order to test the pipeline and estimate the expected S/N of our measurements, we follow a similar approach to [35] and create simulations of the lensed SPT-3G CMB temperature maps with properties similar to our minimum variance combination of the 95, 150, and 220 GHz maps from the real data. We generate a set of $N_{\textup{clus}}$ simulations by creating Gaussian realisations of the CMB in $60^{\prime}\times 60^{\prime}$ flat-sky maps. For each cluster, we model the convergence profile as $\kappa_{\textup{tot}}(M,z)=\kappa_{\textup{1h}}(M,z)+\kappa_{\textup{2h}}(M,z)$κかっぱ start_POSTSUBSCRIPT tot end_POSTSUBSCRIPT ( italic_M , italic_z ) = italic_κかっぱ start_POSTSUBSCRIPT 1h end_POSTSUBSCRIPT ( italic_M , italic_z ) + italic_κかっぱ start_POSTSUBSCRIPT 2h end_POSTSUBSCRIPT ( italic_M , italic_z ). We model the one-halo term as a Navarro-Frenk-White [NFW; 61] profile, with the concentration parameter given by [62]. We account for the impact of uncertainties due to cluster miscentering following [63]:

$$\tilde{\kappa}_{\textup{1h}}(\ell)=\kappa_{\textup{1h}}(\ell)\biggl{[}(1-f_{\textup{mis}})+f_{\textup{mis}}\exp{\biggl{(}-\frac{1}{2}\sigma^{2}_{s}\ell^{2}\biggr{)}}\biggr{]}.$$κかっぱ end_ARG start_POSTSUBSCRIPT 1h end_POSTSUBSCRIPT ( roman_ℓ ) = italic_κかっぱ start_POSTSUBSCRIPT 1h end_POSTSUBSCRIPT ( roman_ℓ ) [ ( 1 - italic_f start_POSTSUBSCRIPT mis end_POSTSUBSCRIPT ) + italic_f start_POSTSUBSCRIPT mis end_POSTSUBSCRIPT roman_exp ( - divide start_ARG 1 end_ARG start_ARG 2 end_ARG italic_σしぐま start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT roman_ℓ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ) ] . (3.5)

Here, we use the DES miscentering fraction $f_{\textup{mis}}=0.22\pm 0.11$ given by [59] and $\sigma_{s}=\sigma_{R}/D_{A}(z)$σしぐま start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT = italic_σしぐま start_POSTSUBSCRIPT italic_R end_POSTSUBSCRIPT / italic_D start_POSTSUBSCRIPT italic_A end_POSTSUBSCRIPT ( italic_z ), where $D_{A}(z)$ is the angular diameter distance at the cluster redshift. The magnitude of miscentering is modelled as a Rayleigh distribution with $\sigma_{R}=c_{\textup{mis}}R_{\lambda}$σしぐま start_POSTSUBSCRIPT italic_R end_POSTSUBSCRIPT = italic_c start_POSTSUBSCRIPT mis end_POSTSUBSCRIPT italic_R start_POSTSUBSCRIPT italic_λらむだ end_POSTSUBSCRIPT, where $R_{\lambda}=(\lambda/100)^{0.2}h^{-1}$λらむだ end_POSTSUBSCRIPT = ( italic_λらむだ / 100 ) start_POSTSUPERSCRIPT 0.2 end_POSTSUPERSCRIPT italic_h start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPTMpc is the DES redMaPPer cluster radius, and $\ln{c_{\textup{mis}}}=-1.13\pm 0.22$ [see 23]. Following the approach of [48], we model the two-halo term, $\kappa_{\textup{2h}}$κかっぱ start_POSTSUBSCRIPT 2h end_POSTSUBSCRIPT, which takes into account the contribution of correlated structures to the total lensing convergence, following Equation (13) of [64]. We convolve the CMB maps with the SPT-3G 150 GHz beam function [54] and add the noise measured from our minimum variance combination of the 95, 150, and 220 GHz maps from the real data. We also include cluster tSZ and kSZ signals based on the Agora simulation set [65] and include the foreground power due to CIB and radio galaxies based on measurements by [66]. To mimic the impact of filtering applied to the data during the map-making process, we follow previous works (including [35]; [45]) and apply a 2D transfer function of the form $F_{\bar{\ell}}=e^{-(\ell_{1}/\ell_{x})^{6}}e^{-(\ell_{x}/\ell_{2})^{6}}$, with $\ell_{1}=300$ and $\ell_{2}=13000$ to the simulations.

The central $10^{\prime}\times 10^{\prime}$ region of each simulation is then extracted and processed through the pipeline as described in Sections 3.1 and 3.2 for the simulated data and models, respectively. For the purpose of pipeline verification, we generated 25 sets of mock CMB cluster lensed simulations each with $N_{\textup{clus}}$, one set of random CMB simulation with $N_{\textup{rand}}=10\times N_{\textup{clus}}$ and 25 sets of model simulations per mass bin, each with $N_{\textup{clus}}$ lensed clusters. In the case of the 25 mock simulation sets, we estimated a unique jackknife covariance matrix for each simulation set and set $N_{\textup{clus}}=5500$, roughly matching the number of clusters in the DES-Y3 Volume-Limited sample described in Section 2.3. For each simulated cluster, we assign the mass and redshift corresponding to a real cluster in the DES-Y3 sample. In the case of the models, where the cluster masses are fixed in each mass bin, we set the redshift distribution of the clusters to that of the DES-Y3 sample. To convert the cluster richness to cluster mass, we use the $M-\lambda$λらむだ scaling relation based on the weak lensing analysis of [47] given by $\langle M_{\rm 200m}|\lambda,z\rangle=3.081\times 10^{14}\ (\lambda/40)^{1.356}((1+z)/1.35)^{-0.3}$λらむだ , italic_z ⟩ = 3.081 × 10 start_POSTSUPERSCRIPT 14 end_POSTSUPERSCRIPT ( italic_λらむだ / 40 ) start_POSTSUPERSCRIPT 1.356 end_POSTSUPERSCRIPT ( ( 1 + italic_z ) / 1.35 ) start_POSTSUPERSCRIPT - 0.3 end_POSTSUPERSCRIPT, resulting in a mean mass of $M_{\rm 200m}=2.21\times 10^{14}{\rm M}_{\odot}$ for the DES Volume&Redshift-Limited cluster sample used in this analysis. We then evaluate the significance of our lensing measurement using $S/N=\sqrt{2[\ln{{\cal L}}(M_{\rm 200m}=M_{\textup{DES}})-\ln{{\cal L}}(M_{\rm 200m}=0)]}$, where $M_{\textup{DES}}$ is our measured stacked mass of the DES cluster sample.

4 Results and Discussion

4.1 Pipeline verification results

Figure 4 shows the result of our pipeline verification test based on 25 sets of simulations (shown as thin orange curves). The thick black curve shows the combined likelihood with the standard deviation multiplied by a factor of $\sqrt{25}$, to estimate the S/N expected from a cluster dataset with the mass and redshift distribution of the DES Volume-Limited sample. The dashed vertical line represents the mean mass of the cluster sample ($2.21\times 10^{14}$$\textup{M}_{\odot}$) and we find that the pipeline successfully recovers the expected mass of the input sample.

Figure 4: Likelihoods showing the results of the pipeline test for 25 sets of simulations (thin orange curves) described in Section 3.4 and their combined likelihood (thick black curves). For the combined likelihood, the width of the distribution is scaled up by a factor of 5 in order to demonstrate the expected S/N from one simulation run with the same number of clusters as our real data. The dashed vertical line is the mean cluster mass of the input sample used to produce the lensing simulations. We find a good agreement between the input mass and the recovered mass.

The successful pipeline validation indicates that the intrinsic richness scatter of the DES sample does not have a significant impact on measuring the mean cluster mass of the sample. Furthermore, the pipeline test indicates the validity of the assumption that, by measuring the mean lensing signal, we can measure the mean cluster mass.

4.2 Mean cluster mass

Our CMB cluster lensing measurement results in a mean (stacked) cluster mass of

$${M}_{200{\rm{m}}} =1.66\pm 0.13\text{ [stat.]}\pm 0.03\text{ [sys.]}\times 10^{14}\,\textup{M}_{\odot}\quad(8.6\%),$$

$${M}_{200{\rm{m}}} =1.97\pm 0.18\text{ [stat.]}\pm 0.05\text{ [sys.]}\times 10^{14}\,\textup{M}_{\odot}\quad(10.2\%),$$

$${M}_{200{\rm{m}}} =2.11\pm 0.20\text{ [stat.]}\pm 0.05\text{ [sys.]}\times 10^{14}\,{\rm{M}}_{\odot}\quad(10.4\%).$$

for the DES-Y3 Flux-, Volume-, and Volume&Redshift-Limited samples, respectively. These are the most constraining mass measurements obtained from CMB cluster lensing to date.

Figure 5 shows a comparison of the mean cluster mass for our Volume&Redshift-Limited sample and the mean cluster mass of the same sample based on the mass–richness scaling relations of [47], [23] (calibrated using optical weak lensing), [52], [45] and [44] (calibrated using CMB cluster lensing) and [67] (calibrated using spectroscopic velocity dispersion measurement of cluster galaxies). For ease of comparison, all measurements are normalised relative to this work with the shaded region marking the $1\sigma$σしぐま uncertainty on our measurement. Overall, we find a good agreement between our measurement of the mean cluster mass and the results of other studies.

Figure 5: Comparison of the measured mean cluster mass of this work (Volume&Redshift-Limited sample) and the mean cluster mass of the same sample, based on the redMaPPer cluster mass–richness scaling relations of various other studies.

4.3 Systematics

In this section, we explore contributions from beam uncertainties, transfer function modelling, residual foregrounds, and cluster miscentering to the systematic error budget. We focus on these sources of systematics, as they have been shown to dominate systematic uncertainty in previous works (see Section 4.2 of [48]). Here, we ignore systematic contributions from underlying cosmology and the choice of halo profile, as these have been found to be small in previous analyses (see [48], [45]) and negligible given the current magnitude of statistical uncertainty. However, quantification of these systematics will become important in future experiments such as CMB-S4, where the statistical uncertainty is expected to be reduced to $1\%$ given the much larger sample size of $\sim 100,000$ clusters.

4.3.1 Cluster tSZ Signal and Residual Foregrounds

The tSZ signal is an important source of systematics, which could result in a bias towards lower masses if not correctly accounted for. In order to verify the success of the random rotation method (described in Section 3.1) in removing the tSZ signal from our stacked lensing dipole, we repeat the measurement using the tSZ-nulled minimum-variance map, described in Section 2.2, for our Volume&Redshift-Limited sample.

Figure 6: Comparison of mass likelihoods obtained using the SPT-3G tSZ-nulled temperature map (dashed orange curve) and our original maps (solid blue curve) in which the tSZ signal is estimated using random rotations and removed from the lensing stack. For comparison, we also include mean sample mass based on the optical weak lensing calibrated, mass–richness scaling relation of [47] (solid vertical line with the $1\sigma$σしぐま error region indicated by the horizontal dotted lines).

Figure 6 shows a comparison of the mass likelihoods obtained from either the baseline minimum variance map or the more noisy tSZ-nulled minimum variance map, with the two appearing in good agreement. Nulling the tSZ signal significantly increases the map variance, primarily due to the CIB and instrumental noise terms. Thus using the tSZ-nulled map approximately doubles the mass uncertainty compared to the baseline case where the tSZ signal is removed by the random rotation stacking procedure. We choose to proceed with the baseline map.

The success of the random rotation technique in mitigating tSZ contamination has also been verified using a larger sample of simulated clusters, in previous studies [60, 35]. Similarly, we use the simulations described in Section 3.4, to estimate the contribution from residual tSZ and kSZ, as well as foregrounds due to CIB and radio galaxies, to our systematic error. This is done by running our pipeline based on a simulation set without foregrounds, tSZ, and kSZ signals, and then repeating the run on the same simulation set with the tSZ, kSZ, and foregrounds added in. Based on this test, we find a contribution of $1.3\%$ to our error budget, equivalent to a shift of $0.12\sigma$σしぐま in our results.

4.3.2 Cluster miscentering

To estimate the systematic uncertainty due to our cluster miscentering model, we repeat the cluster mass measurement for the Volume&Redshift-Limited sample with our fiducial setup but change the miscentering parameters by their $1\sigma$σしぐま uncertainty to $f_{\textup{mis}}=0.30$ (see [47]) and $\ln{c_{\textup{mis}}}=-1.35$. In this case, the measured mean cluster mass increases by $1.7\%$ or $0.15\sigma$σしぐま.

4.3.3 Filtering model

As described in Section 3.4, we estimate the impact of filtering applied to our CMB maps using a 2D transfer function with high- and low-pass filter components set to $\ell_{1}=300$ and $\ell_{2}=13000$ respectively. While $\ell_{2}$ is set to angular scales that do not matter to our lensing reconstruction, we need to calculate the systematics due to the uncertainty on the position of the high-pass filter $\ell_{1}=300\pm 20$. To this end, we recompute our models assuming $\ell_{1}=280$ and 320 and evaluate the changes in the mean lensing mass of our Volume&Redshift-Limited sample. Based on this analysis, we find the systematic contribution of our filtering model to be $0.9\%$ (or $0.07\sigma$σしぐま).

4.3.4 Beam uncertainties

To estimate the uncertainties in our modelling of the telescope beam profile, we generate 10 Gaussian realisations of the beam and regenerate our models of the lensing dipole (described in Section 3.2) for these beams. To generate the simulated beams, we take the beam covariance matrix ${\mathrm{<ruby><rb>\Sigma </rb><rt>しぐま</rt></ruby>}}_{\mathrm{\ell }{\mathrm{\ell }}^{\prime }}=⟨\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}{B}_{\mathrm{\ell }}\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}{B}_{{\mathrm{\ell }}^{\prime }}⟩$Σしぐまℓsuperscriptℓ′delimited-⟨⟩𝛿subscript𝐵ℓ𝛿subscript𝐵superscriptℓ′\Sigma_{\ell\ell^{\prime}}=\langle\delta B_{\ell}\delta B_{\ell^{\prime}}\rangleroman_Σしぐま start_POSTSUBSCRIPT roman_ℓ roman_ℓ start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT = ⟨ italic_δでるた italic_B start_POSTSUBSCRIPT roman_ℓ end_POSTSUBSCRIPT italic_δでるた italic_B start_POSTSUBSCRIPT roman_ℓ start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT ⟩, with $\delta B_{\ell}$δでるた italic_B start_POSTSUBSCRIPT roman_ℓ end_POSTSUBSCRIPT denoting the deviation of $B_{\ell}$ from its mean $\bar{B}_{\ell}$. Here, the elements of the covariance matrix are given by ${\mathrm{<ruby><rb>\Sigma </rb><rt>しぐま</rt></ruby>}}_{\mathrm{\ell }{\mathrm{\ell }}^{\prime }}={(Q\mathrm{<ruby><rb>\Lambda </rb><rt>らむだ</rt></ruby>}{Q}^{\top })}_{\mathrm{\ell }{\mathrm{\ell }}^{\prime }}$Σしぐまℓsuperscriptℓ′subscript𝑄Λらむだsuperscript𝑄topℓsuperscriptℓ′\Sigma_{\ell\ell^{\prime}}=(Q\Lambda Q^{\top})_{\ell\ell^{\prime}}roman_Σしぐま start_POSTSUBSCRIPT roman_ℓ roman_ℓ start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT = ( italic_Q roman_Λらむだ italic_Q start_POSTSUPERSCRIPT ⊤ end_POSTSUPERSCRIPT ) start_POSTSUBSCRIPT roman_ℓ roman_ℓ start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT, where, the diagonal matrix $\mathrm{<ruby><rb>\Lambda </rb><rt>らむだ</rt></ruby>}$Λらむだ\Lambdaroman_Λらむだ contains the eigenvalues $\lambda_{i}$λらむだ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT of the covariance matrix, and the orthogonal matrix $Q$, contains the eigenvectors $v_{i}$ of the covariance matrix, satisfying $\mathrm{<ruby><rb>\Sigma </rb><rt>しぐま</rt></ruby>}{v}_i={\mathrm{<ruby><rb>\lambda </rb><rt>らむだ</rt></ruby>}}_i{v}_i$Σしぐまsubscript𝑣𝑖subscript𝜆𝑖subscript𝑣𝑖\Sigma v_{i}=\lambda_{i}v_{i}roman_Σしぐま italic_v start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT = italic_λらむだ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT italic_v start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT.

Under the assumption that ${\mathrm{<ruby><rb>\Sigma </rb><rt>しぐま</rt></ruby>}}_{\mathrm{\ell }{\mathrm{\ell }}^{\prime }}$Σしぐまℓsuperscriptℓ′\Sigma_{\ell\ell^{\prime}}roman_Σしぐま start_POSTSUBSCRIPT roman_ℓ roman_ℓ start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT encapsulates the statistical characteristics of the beam (i.e., the relevant coefficients follow a Gaussian distribution), we can simulate beam profiles as follows:

$$\hat{B}_{\ell}=\bar{B}_{\ell}+a_{i\ell}e_{i\ell},$$

where $e_{i\ell}$ represents the $\ell$th component of the $i$th column in $Q\sqrt{\mathrm{<ruby><rb>\Lambda </rb><rt>らむだ</rt></ruby>}}$ΛらむだQ\sqrt{\Lambda}italic_Q square-root start_ARG roman_Λらむだ end_ARG, and $a_{i}$ are standard Gaussian variables with zero mean and unit variance $\langle a_{i}a_{j}\rangle=\delta_{ij}$δでるた start_POSTSUBSCRIPT italic_i italic_j end_POSTSUBSCRIPT.

We then repeat our measurement of the mean cluster mass for the Volume&Redshift-Limited sample and take the $1\sigma$σしぐま uncertainty on these 10 new measurements as our estimate of the systematic uncertainty due to our beam modelling. Here, we find a $1.7\%$ contribution to our systematic error budget, equivalent to $0.15\sigma$σしぐま of our statistical error. We provide a summary of the different contributors to our systematic error budget in Table 2.

Source of Error	Magnitude of Error	frac. of $\sigma_{\text{stat}}$σしぐま start_POSTSUBSCRIPT stat end_POSTSUBSCRIPT
Cluster miscentering	$1.7\%$	$0.15\sigma$σしぐま
Beam uncertainties	$1.7\%$	$0.15\sigma$σしぐま
Residual foregrounds	$1.3\%$	$0.12\sigma$σしぐま
Filtering model	$0.9\%$	$0.07\sigma$σしぐま
Total	$2.4\%$	$0.25\sigma$σしぐま

Table 2: Systematic Error Budget in the Stacked Mass for DES RM Year 3 Volume&redshift-limited Sample

4.3.5 Optical weak lensing systematics

In this work, we perform direct comparisons of our results with cluster masses given by the optical weak lensing calibrated mass–richness scaling relation of [47]. As such, we present a brief discussion of systematics impacting optical weak lensing analyses, as well as outstanding cosmological tensions between analyses of the DES Y1 cluster sample and various other cosmological probes.

The constraints on ${\mathrm{<ruby><rb>\Omega </rb><rt>おめが</rt></ruby>}}_{\text{m}}$Ωおめがm\Omega_{\textup{m}}roman_Ωおめが start_POSTSUBSCRIPT m end_POSTSUBSCRIPT and $\sigma_{8}$σしぐま start_POSTSUBSCRIPT 8 end_POSTSUBSCRIPT from a joint cluster abundances and weak lensing analysis of the DES Y1 cluster sample are in $2.4\sigma$σしぐま tension with the DES Y1 $3\times 2$pt results, and in $5.6\sigma$σしぐま with the Planck CMB analysis [68]. The weak lensing measurements used in this analysis were based on the results of [47], where the systematic uncertainty was estimated to be $4.3\%$. As such, weak lensing mass calibration systematics alone are not sufficient to explain the tensions found in the analysis of [68].

In a later work, [69] explored the contamination of the DES-Y1 cluster sample with SPT-SZ selected clusters. Here, it was shown that $10-20\%$ of the $\lambda<40$λらむだ < 40 DES redMaPPer clusters are galaxy groups with masses of $\sim 3-5\times{10}^{13}\,{\rm{M}}_{\odot}$ that are misclassified as more massive clusters due to projection effects. The presence of such low-mass systems in the DES Y1 sample is likely a dominant contributing factor to the cosmological tensions presented in [68]. Indeed, it was shown in [68], that tensions can be significantly alleviated by limiting the sample to clusters with $\lambda\geq 30$λらむだ ≥ 30, which further supports this hypothesis.

It is important to note that systematics due to the presence of such low-mass contaminants will also bias the CMB lensing measurements presented in this study, and in the future, more work is needed to better characterise and minimise the impact of such contaminants via spectroscopic follow-up, use of different cluster detection algorithms, and comparison to future X-ray and SZ samples which probe lower cluster masses.

4.4 Cluster mass–richness scaling relation

Figure 7: Mass–richness scaling relation fitted to our measured cluster masses of the Volume&Redshift-Limited sample in three richness bins (red squares/solid line), compared to the cluster masses given by the mass–richness scaling relation of [47] (blue circles/dashed line). The shaded regions indicate the $1\sigma$σしぐま uncertainty of the scaling relations.

In Figure 7, we compare our binned cluster mass measurements with those given by the mass–richness scaling relation of [47]. We find good agreement between the CMB cluster lensing and optical weak lensing measurements across all richness bins. Based on fitting a two-parameter model (with no redshift evolution), to the cluster mass measurements in our three richness bins, we obtain a mass–richness scaling relation given by:

$$M_{200m}=[3.0\pm 0.4]\times 10^{14}\ {\rm M}_{\odot}(\lambda/40)^{1.6\pm 0.5}.$$λらむだ / 40 ) start_POSTSUPERSCRIPT 1.6 ± 0.5 end_POSTSUPERSCRIPT . (4.1)

This best-fit model has a chi-square value of 2.59 ($p$-value of 0.11). We note that the constraining power of the current data limits us to 3 richness bins, which matches the number of free parameters in the [47] model. As such, we are unable to perform statistical tests to quantitatively assess the level of agreement between our data and the [47] scaling relation which contains 3 free parameters, and leave this to future analyses.

4.5 Mean cluster mass as a function of redshift

Figure 8: Comparison of the mean cluster mass measurements of this work for three redshift bins, to masses based on the optical weak lensing calibrated, mass–richness scaling relation of [47].

Figure 8 shows the mean cluster mass for three redshift sub-samples of our Volume&Redshift-Limited sample described in Table 1. The red data points show the measurements from this work compared to the mean masses of the same subsamples (blue data points), obtained using the optical weak lensing calibrated mass–richness scaling relation of [47]. We find a reasonable agreement between the measurements, noting only a $\sim 1.7\sigma$σしぐま deviation between the two measurements in the lowest redshift bin. As this modest divergence is not statistically significant, we do not investigate it further at this stage. However, it would be interesting to see if such a discrepancy persists in future SPT-3G measurements at a higher level of statistical significance.

4.6 Forecasts

In this section, we provide forecasts for the expected mass constraints one could achieve by including polarization data and upon the completion of the SPT-3G survey, including data from the SPT-3G ‘Extended’ survey (providing a total of $\sim 2800$ deg² overlap with DES, albeit at varying sensitivities). We summarise these forecasts in Table 3. Here, we can see that upon completion of the survey in 2026 and by including polarization and the Extended survey observations, we can improve the current mass constraints by a factor of $\sim 1.8$, obtaining a $5.8\%$ stacked cluster mass constraint. This is much more competitive with optical weak lensing mass constraints and will allow for more precise cosmological parameter estimation.

SPT-3G survey field	Data	Years observed	Map depth	DES overlap	Mass constraint
			[$\mu$μみゅーK.arcmin]	[deg2]	[$\%$]
Main (This work)	T	2019+20	3.1	$1350$	$10.4$
Main	T	2019-23, 2025-26	1.6	$1350$	$7.4$
Main	T+Pol	2019-23, 2025-26	1.6	$1350$	$6.6$
Extended	T+Pol	2019-23	6.1	$1420$	$12.6$
Main+Extended	T+Pol	–	–	$2770$	$5.8$

Table 3: Mass constraint forecasts with the addition of various SPT-3G survey data. Here, the map depth is given by the inverse quadrature sum of the noise in the 95, 150 and 220 GHz frequency bands.

5 Conclusions

In this study, we presented a measurement of the mean cluster masses of three DES-Y3 galaxy cluster samples using the CMB cluster lensing measurements from the initial two years of observations of the SPT-3G survey. Here, we restrict our measurements to the temperature data and the ‘Main’ SPT-3G survey which has a $1350$ deg² overlap with DES (after masking) and leave the addition of SPT-3G polarization maps, as well as data from the SPT-3G ‘Extended’ survey to future works when additional SPT-3G observations are available.

The DES-Y3 cluster samples used in this analysis consist of a Flux-Limited sample with 8865 clusters, a Volume-Limited sample with 5391 clusters and a Volume&Redshift-Limited sample with 4450 clusters. The latter sample is designed to match the selection function of the cluster sample used for DES cluster cosmology analyses, and thus is the primary focus of this work. For the three samples, we detect the CMB lensing dipole with a significance of $10.3\sigma$σしぐま, $10.0\sigma$σしぐま and $9.6\sigma$σしぐま and find the mean cluster masses to be:

$${M}_{200{\rm{m}}} =1.66\pm 0.13\text{ [stat.]}\pm 0.03\text{ [sys.]}\times 10^{14}\,\textup{M}_{\odot}\quad(8.6\%),$$

$${M}_{200{\rm{m}}} =1.97\pm 0.18\text{ [stat.]}\pm 0.05\text{ [sys.]}\times 10^{14}\,\textup{M}_{\odot}\quad(10.2\%),$$

$${M}_{200{\rm{m}}} =2.11\pm 0.20\text{ [stat.]}\pm 0.05\text{ [sys.]}\times 10^{14}\,{\rm{M}}_{\odot}\quad(10.4\%).$$

This measurement represents a factor of $\sim 2$ improvement in precision relative to CMB cluster lensing measurements based on previous generations of SPT surveys (see, e.g. [45], [48] and [35]) and is much more competitive with optical weak lensing mass constraints. Overall, we find good agreement between our measurements and those given by the redMaPPer mass–richness scaling relations of previous works (e.g. [67], [44], [23], [45], [48] and [47]) calibrated using techniques including CMB cluster lensing, optical weak-lensing, and velocity dispersion measurements from various combinations of DES, SDSS, and Planck data.

We verify that our measurements are not significantly biased due to contamination from the residual tSZ signal by comparing the mass of our Volume&Redshift-Limited sample to the mass of the same sample obtained using CMB cluster lensing of a tSZ-nulled ILC map. We find a good agreement between the two measurements, with the higher noise levels in the tSZ-nulled map resulting in a factor of $\sim 2$ larger uncertainty on the mean cluster mass.

Although it was not possible to measure cluster masses for different redshift and richness bins in previous SPT works due to the low S/N of the lensing dipole, the improved sensitivity of the SPT-3G data and the greater overlap with DES enables us to divide our Volume&Redshift-Limited sample into 3 redshift and richness bins, each containing $\sim 1/3$ of the clusters in the full sample. For these subsamples, we obtain mass constraints ranging from $10-20\%$ in precision (with a mean precision of $14\%$). Our results across these sub-samples do not reveal any significant discrepancies when compared to the optical weak lensing calibrated masses given by the scaling relation of [47], although we observe that our mass measurement in the lowest redshift bin is $\sim 1.7\sigma$σしぐま lower.

Finally, we perform forecasts for expected mass constraints using various combinations of the upcoming SPT-3G data. We find that upon completion of the survey in 2026, the combination of temperature and polarization data would yield mass constraints of $6.6\%$ and $12.6\%$ in the Main and Extended SPT-3G surveys, respectively, which translates to a $5.8\%$ mass constraint upon the combination of the full SPT-3G datasets. This level of precision will make CMB cluster lensing much more competitive with optical weak lensing in future years. Moreover, it will serve as an important test of the robustness of cluster mass measurements for precision cosmology due to the complementary systematics of the two measurements.