betaFeedback

WorksPhysics and Astronomy

Euclid Quick Data Release (Q1). Mass-to-light ratio of elliptical galaxies: constraints from strong lensing

Abstract

Mass is among the most fundamental properties describing a galaxy. The relation between the luminosity and the total mass of a galaxy contains information on the efficiency of the star formation process and on the distribution of dark matter. We used strong gravitational lensing data from the Euclid Quick Data Release 1 (Q1) to constrain the total mass enclosed within an aperture of 5 h−15\,h^{-1}kpc, M5M_5, of elliptical galaxies. Our analysis is based on a sample of 95 lenses, which we built by ensuring that their selection function is reproducible. This enabled us to correct our inference for selection effects. The average log⁡10(M5 h M⊙−1)\log_{10}{(M_5\,h\,M_\odot^{-1})} for a galaxy with an rr-band luminosity of 1010.7L⊙ h−210^{10.7}L_{\odot}\,h^{-2} at z=0.5z=0.5 is 11.30±0.0211.30\pm0.02. The mass M5M_5 scales with luminosity to the power βL=0.71±0.07β_\mathrm{L}=0.71\pm0.07, and decreases with increasing redshift at fixed luminosity. Assuming a scenario in which galaxies grow by mergers and stellar populations evolve passively, these correlations imply that M5M_5 increases with time for individual galaxies. These measurements provide a useful benchmark for testing models of galaxy structure and evolution that make predictions on the distribution of total mass on the scales of 5 h−15\,h^{-1}kpc.

Introduction

Elliptical galaxies are some of the best-studied objects in the Universe. The properties of their luminous component are known in great detail. However, key aspects of their inner mass structure, such as the distribution of dark matter, are still uncertain. Mass is the quantity that is most robustly predicted in simulations of galaxies. Therefore, observations that directly trace mass offer a better chance for making accurate comparisons between theoretical models and observations. Models with different recipes for processes dominated by baryons, such as star formation or the ejection of gas from active galactic nuclei, make different predictions on the inner mass distribution (e.g. Duffy et al. [18]; Peirani et al. [51]), and so do simulations with different models of dark matter (Despali et al. [17]). Therefore, precise measurements of mass can help us to build more realistic models of galaxies.

Strong gravitational lensing offers the most direct means of obtaining measurements of mass in galaxies. A strong lens can provide a direct estimate of the total projected mass enclosed within the multiple images of the lensed background source, with a precision on the order of 1%. Hundreds of such measurements have been obtained so far (Auger et al. [3]; Brownstein et al. [10]; Sonnenfeld et al. [61]; Shu et al. [56]; Oldham et al. [48]). However, there are two aspects of strong lensing observations that complicate their interpretation. The first one is the fact that the Einstein radius, which is the main observable in strong lensing, is not purely an intrinsic property of the foreground lens galaxy, but also depends on the redshift of the background source. Consequently, comparing theoretical models to strong lensing data is not as simple as doing this with more traditional observations, such as the stellar mass or the half-radius. To overcome this obstacle, it is helpful to have a model for the density profile of the lens galaxies. The most commonly adopted model of this type is a power law, in which the density is assumed to scale as ρ(r)∝r−γ\rho(r) \propto r^{-\gamma}. The density slope γ\gamma of early-type galaxies has been the focus of several strong lensing-based observational studies (Koopmans et al. [40]; Auger et al. [4]; Bolton et al. [8]; Sonnenfeld et al. [63]; Tan et al. [66]; Sahu et al. [53]), and also of kinematics-only analyses (Tortora et al. [68]; Cappellari et al. [14]). Curiously, very little attention has been paid to the second parameter of the power-law model, the normalisation of the density1. This will be the main focus of our work.

The second obstacle in the interpretation of strong lensing data is the fact that strong lenses are a biased subset of the general population of galaxies. More massive and more compact galaxies have a larger cross-section for creating strong lensing events, and are therefore overrepresented in strong lens samples (Sonnenfeld et al. [62]). Therefore, in order to generalise measurements obtained on strong lenses to the full population of galaxies, selection effects must be taken into account.

The Euclid satellite (Euclid Collaboration: Mellier et al. [29]) is revolutionising the field of strong lensing. Hundreds of new galaxy-scale strong lenses were discovered in the first Euclid Quick Data Release (Q1; Euclid Quick Release Q1 [34]; Euclid Collaboration: Walmsley et al. [32]; Euclid Collaboration: Rojas et al. [30]; Euclid Collaboration: Euclid Collaboration: Lines et al. [28]; Euclid Collaboration: Li et al. [27]; Euclid Collaboration: Holloway et al. [25]). The Q1 data cover an area of 63 deg2^2 (Euclid Collaboration: Aussel et al. [21]), meaning that more than 10510^5 lenses are expected to be discovered in the entire Euclid Wide Survey (which will ultimately cover an area of 14 000 deg2^{2}). In this work, we use the Q1 Euclid strong lensing data to carry out a statistical study of the mass distribution in elliptical galaxies, including a detailed treatment of selection effects. The Q1 strong lens sample is sufficiently large to produce competitive statistical measurements, while sufficiently small to allow for human-led lens modelling, which at this stage is a necessary step in our analysis. The results obtained in this work will help us guide in the analysis of the much larger sample of lenses from the Data Release 1.

The main quantity that we seek to constrain is the total projected mass enclosed within a cylinder of fixed physical radius. We set this radius to 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}, as this is close to the average Einstein radius of the lenses in the Euclid sample. We want to measure how the total fixed-aperture mass varies with luminosity, half-light radius, and redshift, across the population of elliptical galaxies. We choose to work with luminosity, rather than stellar mass, because it can be determined much more robustly. Our main result is therefore a measurement of the mass-to-light ratio integrated within an aperture, 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}, where baryons and dark matter contribute with comparable amounts to the total mass of a typical massive elliptical galaxy ([4]). As this quantity can be readily predicted with hydrodynamical simulations, our measurement offers one of the most direct ways of comparing models of galaxy evolution with observations.

The structure of this paper is as follows. In Sect. 2 we explain the strategy that we adopted for our measurement. In Sect. 3 we present the sample and the data used for our study. In Sect. 4 we describe the procedure that we used to obtain redshift information for the lens galaxies. In Sect. 5 we introduce the model used for our analysis. Section 6 shows the results. We discuss our modelling choices in Sect. 7 and conclude in Sect. 8. Throughout our paper, we assume a flat Λ\LambdaCDM cosmology model, with Ωm=0.3\Omega_{\mathrm{m}} = 0.3. We keep the dependence on the value of the Hubble constant explicit through the parameter h=H0/(100 km s−1 Mpc−1)h = H_{0}/(100\,\mathrm{km\,s}^{-1}\,\mathrm{Mpc}^{-1}).

Strategy

Our goal is to infer the properties of the general population of elliptical galaxies, using observations from a sample of strong lenses. The problem can be summarised as follows. Each strong lens consists of a foreground galaxy, described with a set of parameters ψg\psi_{\mathrm{g}}, and a background source, described with a different set of parameters ψs\psi_{\mathrm{s}}. Foreground galaxies and background sources are drawn from populations described by probability distributions Pg(ψg)P_{\mathrm{g}}(\psi_{\mathrm{g}}) and Ps(ψs)P_{\mathrm{s}}(\psi_{\mathrm{s}}), respectively. With these definitions, the distribution of the population of strong lenses is given by Sonnenfeld [58].

PSL(ψg,ψs)∝Pg(ψg) Ps(ψs) Psel(ψg,ψs).(1)P_{\mathrm{SL}}(\psi_{\mathrm{g}},\psi_{\mathrm{s}}) \propto P_{\mathrm{g}}(\psi_{\mathrm{g}})\,P_{\mathrm{s}}(\psi_{\mathrm{s}})\,P_{\mathrm{sel}}(\psi_{\mathrm{g}},\psi_{\mathrm{s}}). \tag*{(1)}

In the above equation, Psel(ψg,ψs)P_{\mathrm{sel}}(\psi_{\mathrm{g}},\psi_{\mathrm{s}}) is the lens selection probability, that is, the probability of a foreground-background galaxy pair with parameters (ψg,ψs)(\psi_{\mathrm{g}},\psi_{\mathrm{s}}) to produce a strong lens that is included in our lens sample. Our primary target is Pg(ψg)P_{\mathrm{g}}(\psi_{\mathrm{g}}). In order to obtain it, we must invert Eq. (1) above, as the strong lensing data only constrains the left-hand side term, PSLP_{\mathrm{SL}}, directly. Doing so requires good knowledge of the background source distribution before lensing, PsP_{\mathrm{s}}, and the selection function, PselP_{\mathrm{sel}}. To obtain the former, we used a prior from the literature (described in Sect. 5.2). The latter depends critically on the procedure used to define the strong lens sample. To make our analysis as robust as possible, we made the choice of restricting our study to a subset of Euclid lenses with a well-understood selection probability, so as to simplify the evaluation of the PselP_{\mathrm{sel}} term.

The foreground galaxy parameters ψg\psi_{\mathrm{g}}, which are described in detail in Sect. 5.1, include quantities related to the stellar component, such as luminosity, size, and redshift. Measuring how elliptical galaxies are distributed in the parameter space defined by these properties does not require strong lensing data. Therefore, we complemented our strong lens sample with a larger sample of non-lens elliptical galaxies, in order to get a firmer handle on the Pg(ψg)P_{\mathrm{g}}(\psi_{\mathrm{g}}) distribution. We also did this because the difference in the properties of lenses and non-lenses encodes information on the mass distribution of the galaxy population ([62]).

The key ingredient in our analysis is the mass model of the galaxy population. For the sake of simplicity, we chose a singular isothermal sphere (SIS) model, with a three-dimensional density profile ρ(r)∝r−2\rho(r) \propto r^{-2}. Although this is a simpler model compared to many strong lensing studies from the literature, we verified that our results do not change under reasonable deviations from isothermality (Sect. 7).

The main observable that we used to constrain the model is the Einstein radius, θE\theta_{\mathrm{E}}, which is approximately the radius of the circle that encloses the multiple images of a lensed source. For each lens, we measured θE\theta_{\mathrm{E}} directly from the Euclid images. The Einstein radius can be related to the Einstein mass MEM_{\mathrm{E}}, which is the total projected mass enclosed within an aperture of circularised radius θE\theta_{\mathrm{E}} centred on a lens,

ME=π(θEDg)2Σcr(zg,zs).(2)M_{\mathrm{E}} = \pi(\theta_{\mathrm{E}}D_{\mathrm{g}})^{2}\Sigma_{\mathrm{cr}}(z_{\mathrm{g}},z_{\mathrm{s}}). \tag*{(2)}

In the above equation, DgD_{\mathrm{g}} is the angular diameter distance of the lens2, and Σcr\Sigma_{\mathrm{cr}} is the critical surface mass density for lensing. This is defined as

Σcr=c24πGNDsDgDgs,(3)\Sigma_{\mathrm{cr}} = \frac{c^{2}}{4\pi G_{\mathrm{N}}}\frac{D_{\mathrm{s}}}{D_{\mathrm{g}}D_{\mathrm{gs}}}, \tag*{(3)}

where DsD_{\mathrm{s}} and DgsD_{\mathrm{gs}} are the angular diameter distances of the source with respect to the observer and to the lens, respectively. Therefore, in order to convert measurements of θE\theta_{\mathrm{E}} into measurements of mass, information on the redshift of the lens and the source is needed. We only have spectroscopic redshifts for approximately 20% of the lens galaxies, and only three of the lensed sources in our sample. Although spectroscopic follow-up observations of Euclid lenses have been planned, we do not realistically expect to obtain a complete spectroscopic coverage of the full Euclid strong lens sample in the near future. Therefore, we opted for a solution that allowed us to take advantage of all available strong lensing data, regardless of spectroscopy.

To obtain redshift information for the lens galaxies, we chose to leverage the non-lens elliptical galaxy population, for which a much larger number of spectra is available. In particular, we used a spectroscopic sample of elliptical galaxies to infer the mapping between Euclid photometry and redshift, which we then used as a prior for the lens sample.

To obtain source redshifts, we adopted an approach called photogeometric redshift, first introduced by Ruff et al. [52] and later refined by Sonnenfeld [59]. The idea is to simultaneously infer the redshift of the source and the mass of the lens, while taking into account how the lensing cross-section varies as a function of these two properties. In this approach, the flux of the source carries important information, as brighter galaxies tend to be preferentially located at a lower redshift. We used a prior on the distribution of galaxies in redshift and flux space to exploit this information.

Data

Our analysis rests mainly on Euclid photometric data for a sample of elliptical galaxy strong lenses. We describe this sample and their data in Sect. 3.1. At the same time, we also studied a sample of non-lens elliptical galaxies to better understand those properties of the galaxy population that are directly observable without lensing, including redshift. We describe this sample of non-lenses in Sect. 3.2. All of the Euclid data and data products used in this work are taken from Q1. We used images in the band IEI_{\mathrm{E}} of the VIS instrument (Euclid Collaboration: Cropper et al. [23]), and in bands YEY_{\mathrm{E}}, JEJ_{\mathrm{E}}, and HEH_{\mathrm{E}} of NISP (Euclid Collaboration: Jahnke et al. [26]).

Lenses

We built the sample of lenses in such a way as to prioritise the reproducibility of its selection function. We did so by selecting strong lenses in a region of data space where we are reasonably confident that: 1) the Euclid lens sample is close to complete; 2) the lensing nature of the objects in this region can be determined with no ambiguity from Euclid photometric data alone; and 3) the Einstein radius of these lenses can be determined with certainty. In short, we want to ensure completeness and purity of our sample, and robustness of our measurements with respect to modelling choices. We refer to the subset of systems that satisfy these requirements as the ‘golden lenses’. The starting point of our selection of the golden lenses is the ensemble of Euclid strong lenses located in the region covered by Q1. This ensemble consists of approximately 700 lens candidates with visual inspection grade A or B, discovered in the searches described in Euclid Collaboration: Walmsley et al. [32], W26 from here on, Euclid Collaboration: Rojas et al. [30], Euclid Collaboration: Lines et al. [28], Euclid Collaboration: Ecker et al. [24], Euclid Collaboration: Bergamini et al. [22], and Euclid Collaboration: Xu et al. [33]. We then applied cuts to this initial sample:

  1. Lens galaxy IEI_{\mathrm{E}} flux, Fg,IEF_{\mathrm{g},I_{\mathrm{E}}}, larger than 5 μ\muJy (22.15 in AB magnitude), as measured by fitting the lens light with a de Vaucouleurs profile.

  2. Lens galaxy IEI_{\mathrm{E}} axis ratio larger than 0.5.

  3. Isolated: no neighbours within 3″ of the lens.

  4. Lensed source images detected with signal-to-noise ratio S/N >10> 10, and subtending an angle of at least 150° with respect to the lens centre.

  5. Total lensed source S/N larger than 10, when measured over the footprint where the source surface brightness exceeds that of the lens.

  6. An Einstein radius larger than 0″.7.

The application of these cuts requires a model for the surface brightness distribution of the lens and the source. We describe the lens modelling process in the latter part of this section. First, we explain the rationale for the choices listed above.

We applied the first cut to ensure that the lens passed the pre-selection of the search described in W26, which is the most comprehensive lens search carried out so far in Q1 data. From the MERge processing function (MER; Euclid Collaboration: Romelli et al. [31]) catalogue, W26 selected objects for which the quantity FLUX_DETECTION_TOTAL is larger than 6.33078 μ\muJy (22.5 AB magnitude). Typically, this flux indicator includes light from the lensed arcs, and is therefore larger than the de Vaucouleurs flux that we obtained after lens modelling. However, we found lenses with a larger Fg,IEF_{\mathrm{g},I_{\mathrm{E}}}. With our more conservative limit at Fg,IE>5 μF_{\mathrm{g},I_{\mathrm{E}}}>5\,\muJy, only lenses for which the ratio between FLUX_DETECTION_TOTAL and Fg,IEF_{\mathrm{g},I_{\mathrm{E}}} is smaller than 0.7262 would have been missed by W26. Less than 2% of the lenses that we modelled have a flux ratio smaller than this threshold.

The rationale for the second cut is to eliminate disk galaxies from the sample, since our focus is on elliptical galaxies. The third cut is to avoid working with systems of multiple galaxies acting as a lens, which complicates the interpretation of lensing measurements. We applied the fourth cut to select only lenses with unambiguously detected multiple images that bracket the lens galaxy. Using the best-fit lens model image, we first defined a footprint of the lensed source by selecting all regions of contiguous pixels with surface brightness larger than three times the median root mean square fluctuation of the sky background. We measured the S/N of each separate region, and discarded those with S/N <10< 10. Finally, we measured the angle Δϕ\Delta\phi subtended by all pixels associated with regions with S/N >10> 10, with respect to the lens centre, and applied the cut Δϕ>150∘\Delta\phi> 150^\circ. The choice of 150° for the minimum subtended angle is arbitrary, but its exact value has little impact on the sample selection, because the distribution in Δϕ\Delta\phi of our lens candidates is bimodal, with Δϕ\Delta\phi nearing 360° for most of the golden lenses.

The fifth cut selects against lenses for which the contrast between source and lens light is low. These are lenses that are difficult to identify visually, so this cut is needed to maximise completeness. Finally, the sixth cut was needed to deal with the fact that it is very difficult to distinguish real lenses from false positives, such as ring galaxies, when the Einstein radius is small. Although removing low-θE\theta_{\mathrm{E}} lenses from the sample leads to a larger spread between the properties of lenses and non-lens galaxies (Sonnenfeld et al. [62]), this is not a problem for our analysis, because we forward model all of our selection cuts, including that on the Einstein radius. The final sample consists of 95 golden lenses. These are high-confidence strong lenses, based on our ability to reproduce the morphology of the lensed arcs with a simple lens model. Their images are shown in Figure 1.

Colour-composite images of the 95 golden lenses used in our study

Figure 1. Colour-composite images of the 95 golden lenses used in our study. Data from the IEI_{\mathrm{E}}, YEY_{\mathrm{E}}, and JEJ_{\mathrm{E}} bands are used for the blue, green, and red channels of the images, respectively. Cyan regions, if present, are due to masked pixels in the NISP images. This can be a severe issue for some of the lensed arcs. Nevertheless, NISP data are only used to characterise the colours of the lens galaxy, which are well measured despite the masking of the source light. Each image is 10′′^{\prime\prime} on the side.

This number is about a factor of seven smaller than the initial sample of grade A or B lenses in Q1. The reasons for such a sharp compression are various. A few tens of lens candidates were eliminated at the catalogue level, based on their lens galaxy being too faint, bringing the sample down to 693 candidates. Of these, 119 were found not to show evidence of multiple imaging, in the sense of our golden lens definition. These objects are likely gravitational lenses, but do not have multiple images on opposite sides of the lens galaxy, and therefore do not pass our cuts. A further 97 were eliminated due to the presence of neighbours, and 64 lenses were discarded due to the lens galaxy having a clear disk morphology (most of which would likely not have passed the second cut in our golden lens definition, had they been modelled). The remaining 408 lenses were fitted with a lens model (described later in this section). Of them, 47 were discarded for the lens being too faint, after being fit with a de Vaucouleurs profile. Then, a further 190 were found not to show evidence of multiple images with sufficiently high S/N. That is, they failed to meet the requirements of point 4 in the above list of cuts. Of the remaining ones, 45 were discarded for their Einstein radius being too small (14 of them cannot be distinguished from ring galaxies), one for the lens being too disky,

and 13 for the source-to-lens contrast being too low (criterion 5 in the above list). Finally, we removed 17 candidates from the remaining 112, because we deemed them likely not to be lenses, based on our difficulty in qualitatively matching the supposed arc features with a lens model. We discuss these and other ambiguous lens candidates in Appendix A. Although we cannot exclude with certainty that some of these systems are actual lenses, we verified that the results of our analysis are robust with respect to small changes to the lens sample.

Of the 95 golden lenses, 18 have spectroscopic redshifts. Twelve of these redshifts come from the Dark Energy Spectroscopic Instrument Data Release 1 (DESI DR1; DESI Collaboration et al. [16]), five from the Australian Dark Energy Survey (OzDES; Lidman et al. [41]), and one from the PRIsm MUlti-object Survey (PRIMUS; Coil et al. [15]). We have spectroscopic redshifts of only three lensed sources: EUCL J175107.51+653150.6 at zs=0.406z_{\mathrm{s}} = 0.406 (NGC 6505; O’Riordan et al. [49]), EUCL J180354.65+643421.6 at zs=1.897z_{\mathrm{s}} = 1.897, and EUCL J175049.89+665454.5 at zs=1.956z_{\mathrm{s}} = 1.956 (Euclid Collaboration: Rojas et al. [30]).

We fitted 408 lens candidates with a lens model. The main goals of this modelling procedure are to measure the Einstein radius of the lenses, to provide a description of the surface brightness distribution of the lens and the background source, and to measure the colours of the lens galaxy. The data used for modelling were coadded images, as well as noise maps and models of the point-spread function (PSF), as produced by MER. This PSF is based on stacked PSFEx models (Bertin [6]), sampled at 0\farcs1/pixel, and not accounting for wavelength dependence. We obtained these data from ESA Datalabs (Navarro et al. [46]).

We carried out the modelling process by first fitting a PSF-convolved Sérsic model to the IEI_{\mathrm{E}} lens light distribution, while masking out lensed arcs and neighbouring objects. We used the object detection software SExtractor (Bertin & Arnouts 1996) for this purpose. We visually inspected the lens-subtracted images, along with the SExtractor segmentation maps, and classified the detected regions into lensed source images and foreground contaminants. We used the former to obtain a starting guess for the lens model, and masked out the latter. Then, we added an exponential profile to describe the background source, and a singular isothermal ellipsoid (SIE) to model the mass of the lens. We imposed the centre of mass to be the same as the centre of light. We used the lens modelling software Glafic (Oguri [47]) to predict the surface brightness distribution of the lensed source, and convolved it with the PSF to compare it with the IEI_{\mathrm{E}} data. Finally, we optimised for the lens and source model parameters. For most systems in the sample, the model obtained after this optimisation was sufficiently good for our purposes, in the sense that it reproduced the positions and shapes of all of the observed lensed images, so that the Einstein radius was well constrained. For 40 lenses, however, we manually added elements of complexity to the model, such as a second Sérsic component for the lens light, external shear, or a spatial offset between the lens light and mass. A few of the lenses in our sample have two lensed sources at two different redshifts, as shown by Euclid Collaboration: Li et al. [27]. In those cases, we only modelled the arcs produced by the lowest-redshift source (i.e. the innermost arcs), as that is sufficient to measure the Einstein radius. Once we obtained a satisfactory model, we fixed its structural parameters and optimised solely for the amplitude of each light component in the three NISP bands. In parallel to this last step, we also fitted a model with a de Vaucouleurs profile (i.e. a Sérsic profile with index n=4n = 4) for the lens light, optimising its structural parameters on the IEI_{\mathrm{E}} image while keeping the mass and source parameters fixed. The motivation for this additional fit is that our science analysis is based on a de Vaucouleurs model for the lens, which we prefer over the more general Sérsic in virtue of its reduced number of degrees of freedom (two, instead of three).

Figure 2 shows examples of models of a few lenses. Although in several cases our model is unable to reproduce small-scale features in the images of the lensed source, it can qualitatively match the positions, sizes, shapes, and orientations of the multiple images.

Example lens models

Figure 2. Example lens models. From left to right, each column shows the original image, the full model, the lens-subtracted image, the source-only model, the original (before lensing) source-plane image along with the caustic lines of the lens, the residual image, and the VIS-band residual normalised by the uncertainty. All colour images have the same brightness and colour scale. Although colour-composite images are shown here, only the IE data were used to optimise the lens model parameters. In the last panel, masked pixels are set to zero. Each image is 10″ on the side.

Figure 3 shows the distribution in Einstein radius, half-light radius, lens flux, and ratio between YEY_{\mathrm{E}}- and IEI_{\mathrm{E}}-band flux of the lenses. We see a clear correlation between flux and YEY_{\mathrm{E}}- and IEI_{\mathrm{E}}-flux ratio. This reflects the underlying correlation between colour and redshift, which we exploit to gain redshift information, as explained in § 4. Figure 3 also shows the distribution in flux and angular half-light radius of the background sources, before lensing. We see a clear correlation between the two. Although we do expect brighter sources to be, on average, more extended, part of this correlation could also be due to a degeneracy between these two quantities. This is because neither the total flux nor the half-light radius is directly observable, but is derived under the assumption of an exponential surface brightness profile, often extrapolated to regions that are not constrained by the data. Table 1 lists a selection of properties of the golden lenses. A more complete list with all of the parameters of the model is available online3. We also provide individual lens models in the code-independent organised lens standard (COOLEST; Galan et al. [36]).

Lens namezgz_{\mathrm{g}}zg,photz_{\mathrm{g,phot}}log⁡10(Fg,IE/μJy)\log_{10}(F_{\mathrm{g,IE}}/\mu\mathrm{Jy})θe\theta_{\mathrm{e}} (″)θE\theta_{\mathrm{E}} (″)zsz_{\mathrm{s}}log⁡10(Fs,IE/μJy)\log_{10}(F_{\mathrm{s,IE}}/\mu\mathrm{Jy})
EUCL J032408.71−271429.5...0.75±0.050.75 \pm0.051.381.350.75...−0.47
EUCL J032618.89−281123.00.7750.72±0.050.72 \pm0.051.100.860.75...−1.01
EUCL J032723.88−263653.0...0.97±0.050.97 \pm0.051.191.591.43...−0.84
EUCL J032751.20−272533.4...0.63±0.040.63 \pm0.041.310.750.80...−0.51
EUCL J032923.51−290907.60.7300.71±0.050.71 \pm0.051.291.470.83...−1.06
EUCL J032929.39−262115.1...0.26±0.030.26 \pm0.032.280.971.31...−0.77
EUCL J033101.78−285247.7...0.93±0.050.93 \pm0.050.870.510.77...−0.03
EUCL J033410.71−281210.10.2890.33±0.040.33 \pm0.042.150.801.35...0.29
EUCL J033414.91−280046.4...0.40±0.050.40 \pm0.051.920.960.89...−0.07
EUCL J033459.29−263431.60.3460.32±0.030.32 \pm0.032.451.702.49...−0.63

Table 1. Selected observables properties of the golden lenses.

Figure 3. Distribution of observed properties of the lenses. These are the Einstein radius, flux, half-light radius, and YEY_{\mathrm{E}}- to IEI_{\mathrm{E}}-band flux ratio of the lenses, and the unlensed flux and half-light radius of the background sources. The distribution of the non-lens elliptical galaxy sample (i.e. the parent population) is also shown. Scatter points of lenses with spectroscopic redshift are circled in green. The histograms of the parent and lens population are normalised to one. The latter include the lenses with spec-zz. The spec-zz sample histogram is normalised to the fraction of lenses with spectroscopic redshift. AB magnitude scales are shown in correspondence to the flux and colour columns.

We estimated the completeness of the golden lens sample to be 64%. Appendix B explains how we obtained this value. For the rest of this paper, we assume that our sample is effectively complete, meaning that, if any lenses are missing, they are indistinguishable from the rest of the sample.

Non-lens sample

The purpose of the non-lens sample of elliptical galaxies is to provide prior information on the properties of the elliptical galaxies population, to be used in the strong lensing analysis. For this reason, it is important that this sample is built with the same selection steps that we applied to the golden lens samples (where applicable), and that their photometric measurements are carried out consistently with our lens models.

To select a representative sample of non-lens elliptical galaxies, we drew random positions over the Q1 area, and searched for galaxies lying within 5′′5^{\prime\prime} of each drawn position. From the MER catalogue, we first selected all objects with FLUX_DETECTION_TOTAL >> 3.63078 μ\muJy (i.e. the same cut used for the pre-selection of lenses in W26), SERSIC_SERSIC_VIS_AXIS_RATIO >> 0.4 (we applied a more stringent cut on axis ratio in a second step, described below), and POINT_LIKE_PROB << 0.2. We also applied a lower limit to the YEY_{\mathrm{E}} to IEI_{\mathrm{E}}-band flux ratio, log⁡10(Fg,YE/Fg,IE)>0.3\log_{10}(F_{g,Y_{\mathrm{E}}}/F_{g,I_{\mathrm{E}}}) > 0.3, using aperture fluxes within one FWHM in PSF-matched images. The reason for this colour cut is that the golden lenses are all observed to be redder than this limit (Figure 3). With this procedure, we searched over an effective area of 0.19 deg2^2, and found 1549 galaxies (which extrapolate to ∼5×105\sim5 \times10^{5} over the entire Q1 area). We visually inspected them to select galaxies with elliptical morphology, and no neighbours within a 3′′3^{\prime\prime} aperture (this is the same isolation criterion that we adopted for the lenses), Notes. Spectroscopic redshift zgz_{\mathrm{g}} (where available), photometric redshift zg,photz_{\mathrm{g,phot}} (from the procedure described in Sect. 4), IE flux, half-light radius θe\theta_{\mathrm{e}}, and Einstein radius θE\theta_{\mathrm{E}} of the lens galaxies; IE flux and spectroscopic redshift (where available) of the lensed source. All fluxes are corrected by Galactic extinction, using the Fitzpatrick [35] dust extinction curve. Lens fluxes and sizes were obtained by fitting a de Vaucouleurs profile to the IE image. Only the first ten lenses are shown in print. The full table is available online.

reducing the sample size to 657. Finally, we fitted the IE image of each galaxy with a de Vaucouleurs model, and selected all galaxies with Fg,IE>5 μJyF_{\mathrm{g,IE}} > 5\,\mu\mathrm{Jy} and axis ratio larger than 0.5.

The final sample consists of 534 non-lens elliptical galaxies. Of them, 34 have spectroscopic redshifts. We measured their

NISP fluxes via forced photometry4 to the best-fit IE de Vaucouleurs model. Fluxes, half-light radii, and YE- to IE-band flux ratios of these galaxies are shown in Figure 3 alongside the lens sample. We can see that the two populations overlap, with the non-lenses extending to a region of low flux and relatively blue colour that is not covered by the lenses. Those are generally low-redshift and low-mass galaxies, with a very small cross-section for strong lensing.

Photometric redshifts and physical properties

Our basis for obtaining photo-zzs is an empirical relation between redshift and the (IEI_{\mathrm{E}}, YEY_{\mathrm{E}}, JEJ_{\mathrm{E}}, HEH_{\mathrm{E}}) fluxes of elliptical galaxies, which we derived as follows. We selected a subset of 1000 galaxies from the sample of objects with DESI DR1 spectra that lie in the Q1 area. When drawing this subset, we gave a larger weight to the brightest galaxies at each redshift, so as to obtain a sample that is broadly representative of strong lens galaxies. We visually inspected this sample, and selected 252 isolated elliptical galaxies. Then, we fitted a de Vaucouleurs profile to their IEI_{\mathrm{E}} image, and obtained NISP band fluxes via forced photometry.

Figure 4 shows the ratio between the flux in each NISP band and the IEI_{\mathrm{E}} flux, as a function of redshift, of these elliptical galaxies with DESI spectra. We can see a clear correlation for all three bands. In light of these data, we fitted the following model to the distribution in the four-band flux of elliptical galaxies as a function of redshift,

NISP to $I_{\mathrm{E}}$ flux ratio of elliptical galaxies as a function of redshift

Figure 4. NISP to IEI_{\mathrm{E}} flux ratio of elliptical galaxies as a function of redshift. Empty circles: 252 elliptical galaxies with DESI DR1 spectroscopy. Filled circles: 18 golden lenses (not used in the inference of the photo-z). Dashed lines indicate the model μλ(zg,Fg,IE)\mu_{\lambda}(z_{\mathrm{g}}, F_{\mathrm{g},I_{\mathrm{E}}}), evaluated at Fg,IE=F∗(zg)F_{\mathrm{g},I_{\mathrm{E}}}=F_{*}(z_{\mathrm{g}}), that we inferred. The right axis indicates colours in units of AB magnitude.

P({Fg,λ}∣z)=∏λ∈NISPNlog⁡10(Fl,λ/Fg,IE)[μλ(zg,Fg,IE),σλ2],(4)P(\{F_{\mathrm{g},\lambda}\}|z)=\prod_{\lambda\in\mathrm{NISP}}\mathcal{N}_{\log_{10}(F_{\mathrm{l},\lambda}/F_{\mathrm{g},I_{\mathrm{E}}})}\left[\mu_{\lambda}(z_{\mathrm{g}},F_{\mathrm{g},I_{\mathrm{E}}}),\sigma_{\lambda}^{2}\right], \tag*{(4)}

where the product is over the three NISP bands, and the notation Nx[μ,σ2]\mathcal{N}_{x}[\mu,\sigma^{2}] indicates a Gaussian distribution in the variable xx with mean μ\mu and variance σ2\sigma^{2}. The mean of each Gaussian takes the following form,

μλ(zg,Fg,IE)=μλ,0+ζλ,1(zg−zt){1−s[b(zg−zt)]}\mu_{\lambda}(z_{\mathrm{g}},F_{\mathrm{g},I_{\mathrm{E}}})=\mu_{\lambda,0}+\zeta_{\lambda,1}(z_{\mathrm{g}}-z_{\mathrm{t}})\{1-s[b(z_{\mathrm{g}}-z_{\mathrm{t}})]\}
μλ(zg,Fg,IE)=+ζλ,2(zg−zt)s[b(zg−zt)]+ωlog⁡10(Fg,IEF⋆(zg)),(5)\phantom{\mu_{\lambda}(z_{\mathrm{g}},F_{\mathrm{g},I_{\mathrm{E}}})=}+\zeta_{\lambda,2}(z_{\mathrm{g}}-z_{\mathrm{t}})s[b(z_{\mathrm{g}}-z_{\mathrm{t}})]+\omega\log_{10}\left(\frac{F_{\mathrm{g},I_{\mathrm{E}}}}{F_{\star}(z_{\mathrm{g}})}\right), \tag*{(5)}

where ss is a sigmoid function,

s(x)=11+e−x,(6)s(x)=\frac{1}{1+e^{-x}}, \tag*{(6)}

F⋆F_{\star} is a redshift-dependent characteristic flux,

F⋆(zg)=[DL(0.5)DL(zg)]2100 μJy,(7)F_{\star}(z_{\mathrm{g}})=\left[\frac{D_{\mathrm{L}}(0.5)}{D_{\mathrm{L}}(z_{\mathrm{g}})}\right]^{2}100\,\mu\mathrm{Jy}, \tag*{(7)}

and DLD_{\mathrm{L}} is the luminosity distance. In words, we modelled the ratio between the near-infrared and optical flux, in each NISP band, as a Gaussian, with a mean that scales with redshift as a two-piece linear function, with an additional dependence on the IEI_{\mathrm{E}} flux. The parameter ztz_{\mathrm{t}} indicates the redshift at which the slope of the colour-redshift relation changes, while bzb_{z} governs the smoothness of the transition. In all, the model parameters are: μλ,0\mu_{\lambda,0}, ζλ,1\zeta_{\lambda,1}, ζλ,2\zeta_{\lambda,2}, ωλ\omega_{\lambda}, and σλ\sigma_{\lambda}, for each NISP filter; ztz_{\mathrm{t}}, and bzb_{z}. Their inferred values are listed in Table 2. Throughout the rest of our analysis, we fix the value of each of these parameters to the median of its marginal posterior probability. In Figure 4 we show, for each NISP band, the mean colour μλ(zg,Fg,IE)\mu_{\lambda}(z_{\mathrm{g}}, F_{\mathrm{g},I_{\mathrm{E}}}), evaluated at the reference flux F∗(zg)F_{*}(z_{\mathrm{g}}).

Parameter(s)Posterior (68%)Eq. Ref.Description
ztz_t0.33±0.100.33 \pm0.10(5)Transition redshift of the redshift-colour relation.
aza_z6.4±8.26.4 \pm8.2(5)Rapidity of the transition from the low- to the high-redshift regime.
μY,0,μJ,0,μH,0\mu_{Y,0}, \mu_{J,0}, \mu_{H,0}0.41±0.030.41 \pm0.03, 0.55±0.040.55 \pm0.04, 0.67±0.050.67 \pm0.05(5)Average VIS–NISP colour at z=ztz = z_t, for each NISP band.
ζY,1,ζJ,1,ζH,1\zeta_{Y,1}, \zeta_{J,1}, \zeta_{H,1}0.07±0.230.07 \pm0.23, 0.21±0.210.21 \pm0.21, 0.42±0.160.42 \pm0.16(5)Scaling of colour with redshift, at the low-zz end.
ζY,2,ζJ,2,ζH,2\zeta_{Y,2}, \zeta_{J,2}, \zeta_{H,2}0.46±0.030.46 \pm0.03, 0.51±0.030.51 \pm0.03, 0.59±0.030.59 \pm0.03(5)Scaling of colour with redshift, at the high-zz end.
ωY,ωJ,ωH\omega_Y, \omega_J, \omega_H0.035±0.0110.035 \pm0.011, 0.040±0.0150.040 \pm0.015, 0.047±0.0180.047 \pm0.018(5)Scaling of colour with IEI_{\rm E} flux.
σY,σJ,σH\sigma_Y, \sigma_J, \sigma_H0.042±0.0020.042 \pm0.002, 0.051±0.0030.051 \pm0.003, 0.059±0.0030.059 \pm0.003(4)Scatter in colour at fixed flux and redshift.

Table 2. Parameters of the photometric redshift model. Notes. The second column reports the median and 68% credible region of the marginal posterior probability. The third column indicates the equation where each parameter is first defined.

We checked the accuracy of our model with the 18 lenses for which we have a spectroscopic redshift. For each lens, we computed the posterior probability of the redshift given all available data, which include the lens fluxes, the half-light radius, the Einstein radius, and the intrinsic source flux. All of these observables can inform the redshift of the lens, when interpreted in the context of a lens model. The correct prior probability to be used for this purpose is given by the distribution of Eq. (1), marginalised over the properties of the source, over the mass of the lens, and over the uncertainty of the model. Figure 5 shows the resulting posterior probability, as a function of the spectroscopic redshift. There is good agreement between the posterior predicted redshifts and the spectroscopic ones, with no catastrophic outliers. The average difference between the two is Δz=0.014\Delta z = 0.014, which translates into an error on Σcr\Sigma_{\mathrm{cr}} of less than 0.1%, much smaller than the precision on our measurement of the average mass-to-light ratio of elliptical galaxies. Therefore, we are not concerned with this bias impacting the results of our work.

Figure 5. Photo-z accuracy test. Posterior predicted redshift as a function of spectroscopic redshift for the lenses with spectroscopic data. Error bars bracket an enclosed probability of 68%. Purple data points were obtained by simply using the model of Eq. (4) and assuming a uniform prior on the lens redshift. Orange data points were obtained by using all available information, and our full model of the lens population as a prior.

In order to gauge the relative importance of the photo-z model of Eq. (4), in Figure 5 we also show posterior predicted redshifts obtained by assuming a uniform prior on the lens redshift, and ignoring any piece of information other than the colours. A comparison with the full posterior shows that colour information alone can constrain the redshift with a typical precision of 0.05.

This uncertainty is set by the observed scatter in the colour of elliptical galaxies at fixed redshift.

One caveat in this analysis is that the spectroscopic redshifts that we used to constrain our model are all drawn from DESI, which specifically targeted luminous red galaxies by means of cuts in flux–colour space ([71]). These galaxies are not necessarily a representative subsample of the entire population of elliptical galaxies, and therefore the mapping between photometry and redshift that we derived might be biased. As most of the spectroscopic redshifts for our lens sample are also from DESI, this bias might be underestimated in our comparison of Figure 5. However, luminous red galaxies are fairly representative of the typical Euclid lens, since, of the lenses that lie in the region of the sky covered by DESI DR1, roughly a third are DESI targets. When considering the six lenses with a spectroscopic redshift from a survey other than DESI, we found that their average photo-z is larger than their average redshift by only 0.01, with a 0.02 uncertainty. Taking this as an estimate of the bias due to the spectroscopic training sample, we conclude that this bias is not significant in the context of our mass measurement.

Using the photo-z, we derived physical properties of the galaxies in our samples. We obtained luminosities by fitting spectral templates to the observed fluxes, which we corrected for Galactic extinction using the model of Fitzpatrick [35]. For the fit, we used elliptical-type templates from the Kinney–Calzetti spectral atlas ([13, 39]), as provided by the software LePhare ([2, 37]). We complemented these with ultra-red templates, which we built by applying dust attenuation to the ELLE_{\mathrm{LL}} template. This was made necessary by the fact that, for the reddest galaxies in our sample, none of the standard Kinney-Calzetti templates was able to provide a good fit. We normalised the templates to the same rest-frame SDSS rr-band luminosity, LrL_{r}, and sorted them by their increasing Lr/LgL_{r}/L_{g}, where LgL_{g} is the rest-frame SDSS gg-band luminosity. Throughout the rest of the paper, we refer to the ratio Lr/LgL_{r}/L_{g} as the rest-frame colour. Then, for a given redshift, we computed their Euclid fluxes and interpolated them, obtain flux as a continuous variable TT, which spans templates from the bluest to the reddest. Given zgz_g and {Fg,λ}\{F_{g,\lambda}\}, we optimised for TT and for the normalisation of the template, to obtain luminosity and rest-frame colour.

Figure 6 shows physical Einstein radii, half-light radii, luminosities, rest-frame colours, and redshifts, for the lenses in our sample. The median Einstein radius is close to 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}. This motivated our choice of the mass enclosed within that aperture as the main target for our analysis.

Figure 6. Physical properties of the lenses in our sample. These include redshift, luminosity, rest-frame colour, half-light radius, and Einstein radius. For the lenses with spectroscopic redshift, values are shown with scatter points. For the remaining lenses, the curve corresponding to the 68% credible region of the photo-zz is shown. The gold histograms include both spectroscopic and photometric lenses. Contours show the 68% and 95% enclosed probability of the distribution of non-lens elliptical galaxies. The dashed black curve shows the luminosity-size relation of non-lens elliptical galaxies, evaluated at zg=0.5z_{\rm g} = 0.5.

Model

We want to fit a model distribution of the form of Eq. (1) to the ensemble of golden lenses. The main quantities that we want to constrain are a set of parameters describing the population of elliptical galaxies. We indicate them collectively as η\boldsymbol{\eta}, and we introduce them in Sect. 5.1. The posterior probability of η\boldsymbol{\eta} is

P(η∣d)∝P(η) P(d∣η).(8)P(\boldsymbol{\eta} \mid d) \propto P(\boldsymbol{\eta})\,P(d \mid\boldsymbol{\eta}) . \tag*{(8)}

where dd is the ensemble of data, P(η)P(\boldsymbol{\eta}) is the prior probability on the population parameters, and P(d∣η)P(d \mid\boldsymbol{\eta}) is the likelihood of observing the data given the model. The data consist of the Einstein radius, the unlensed source IEI_{\rm E} flux, the IEI_{\rm E} and NISP lens fluxes, and the lens half-light radius,

d=(θE,Fs,IE,{Fg,λ},θe).(9)d = (\theta_{\rm E}, F_{s,I_{\rm E}}, \{F_{g,\lambda}\}, \theta_e) . \tag*{(9)}

In so doing, we compress all of the information from the lens model into the pair of parameters (θE,Fs,IE)(\theta_{\rm E}, F_{s,I_{\rm E}}). The data for each lens are independent of each other, therefore we can write

P(d∣η)=∏iP(di∣η)=∏i∫dψg,idψs,i P(di∣ψg,i,ψs,i)PSL(ψg,i,ψs,i∣η).(10)\begin{aligned} P(d \mid\boldsymbol{\eta}) = \prod_i P(d_i \mid\boldsymbol{\eta}) \\ &= \prod_i \int\mathrm{d}\boldsymbol{\psi}_{g,i}\mathrm{d}\boldsymbol{\psi}_{s,i}\,P(d_i \mid\boldsymbol{\psi}_{g,i}, \boldsymbol{\psi}_{s,i})P_{\rm SL}(\boldsymbol{\psi}_{g,i}, \boldsymbol{\psi}_{s,i} \mid\boldsymbol{\eta}) . \tag*{(10)} \end{aligned}

The second equality means that the likelihood of the data for the iith lens can be obtained while marginalising over its individual parameters. The probability distribution PSL(ψg,ψs∣η)P_{\rm SL}(\boldsymbol{\psi}_g,\boldsymbol{\psi}_s \mid\boldsymbol{\eta}) is, in general, not analytical. This means that, in order to obtain an accurate result, we have to actively ensure that it is always properly normalised. That is,

∫Pg(ψg∣η)Ps(ψs)Psel(ψg,ψs) dψgdψs=1,(11)\int P_g(\boldsymbol{\psi}_g \mid\boldsymbol{\eta})P_s(\boldsymbol{\psi}_s)P_{\rm sel}(\boldsymbol{\psi}_g,\boldsymbol{\psi}_s)\,\mathrm{d}\boldsymbol{\psi}_g\mathrm{d}\boldsymbol{\psi}_s = 1 , \tag*{(11)}

for any value of the parameters η\boldsymbol{\eta}. Assuming that the background sources are randomly distributed in the sky, the integral over the angular position of the source can be computed separately from that over the other variables. We can therefore introduce the strong lensing cross-section, defined as the following integral over the source position β\boldsymbol{\beta},

σSL(ψg,ψs(−β))=∫Psel(ψg,ψs(−β)) d2β,(12)\sigma_{\rm SL}(\boldsymbol{\psi}_g,\boldsymbol{\psi}_s^{(-\beta)}) = \int P_{\rm sel}(\boldsymbol{\psi}_g,\boldsymbol{\psi}_s^{(-\beta)})\,\mathrm{d}^2\beta, \tag*{(12)}

where ψs(−β)\boldsymbol{\psi}_s^{(-\beta)} is the set of source parameters excluding β\beta, and the integral extends over the whole sky. With this definition, Eq. (11) becomes

∫Pg(ψg∣η)Ps(ψs)σSL(ψg,ψs(−β)) dψgdψs(−β)=1.(13)\int P_g(\boldsymbol{\psi}_g \mid\boldsymbol{\eta})P_s(\boldsymbol{\psi}_s)\sigma_{\rm SL}(\boldsymbol{\psi}_g,\boldsymbol{\psi}_s^{(-\beta)})\,\mathrm{d}\boldsymbol{\psi}_g\mathrm{d}\boldsymbol{\psi}_s^{(-\beta)} = 1 . \tag*{(13)}

In the rest of this section, we describe the details of the model.

Foreground galaxy population

We describe the population of elliptical galaxies with a probability distribution of the following form,

Pg(ψg)=P⋆(zg,Lr,Lr/Lg,Re) PM(M5∣zg,Lr,Re),(14)P_g(\boldsymbol{\psi}_g) = P_\star(z_g, L_r, L_r/L_g, R_e)\,P_M(M_5 \mid z_g, L_r, R_e) , \tag*{(14)}

where P⋆P_\star describes the properties of the stellar component, while PMP_M describes the mass distribution. In Eq. (14), M5M_5 is the total projected mass enclosed within an aperture of 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}. When comparing our spherical mass distribution with the SIE-based lens models of the golden lenses, we focus exclusively on the Einstein radius, which is a circularly-averaged quantity. Therefore, M5M_5 uniquely determines the projected mass distribution, which is given by

Σ(R)=M52π(5 h−1 kpc)2(5 h−1 kpcR).(15)\Sigma(R) = \frac{M_5}{2\pi(5\,h^{-1}\,\mathrm{kpc})^2}\left(\frac{5\,h^{-1}\,\mathrm{kpc}}{R}\right) . \tag*{(15)}

We parameterise the first factor of Eq. (14) as

P⋆(zg,Lr,Lr/Lg,Re)=P(zg) P(Lr∣zg) P(Lr/Lg∣zg) P(Re∣zg,Lr).(16)P_\star(z_g, L_r, L_r/L_g, R_e) = P(z_g)\,P(L_r \mid z_g)\,P(L_r/L_g \mid z_g)\,P(R_e \mid z_g,L_r) . \tag*{(16)}

For the marginal distribution in redshift, P(zg)P(z_g), we adopt a generalised gamma distribution,

P(zg)=∣cz∣Γ(az)z0(zgz0)azcz−1exp⁡[−(zgz0)cz].(17)P(z_g) = \frac{|c_z|}{\Gamma(a_z) z_0}\left(\frac{z_g}{z_0}\right)^{a_zc_z-1}\exp\left[-\left(\frac{z_g}{z_0}\right)^{c_z}\right] . \tag*{(17)}

For the luminosity distribution we adopt a Schechter function,

P(Lr∣zg)∝[LrL⋆(zg)]αe−Lr/L⋆(zg).(18)P(L_r \mid z_g) \propto\left[\frac{L_r}{L_\star(z_g)}\right]^\alpha e^{-L_r/L_\star(z_g)} . \tag*{(18)}

with a redshift-dependent characteristic luminosity,

log⁡10[L∗(zg)h2Lr,⊙]=log⁡10(L,0∗h2Lr,⊙)+ζL(zg−0.5).(19)\log_{10}\left[\frac{L^{*}(z_{\rm g})h^{2}}{L_{r,\odot}}\right] = \log_{10}\left(\frac{L^{*}_{,0}h^{2}}{L_{r,\odot}}\right) + \zeta_{L}(z_{\rm g} - 0.5). \tag*{(19)}

The luminosity distribution is truncated at a minimum value Lmin⁡L_{\min}, set by our selection cut on the IEI_{\rm E}-band flux. The exact value of Lmin⁡L_{\min} depends on the redshift and colour of a galaxy. We modelled the distribution in rest-frame colour as a Gaussian in log⁡10(Lr/Lg)\log_{10}(L_{r}/L_{g}),

P(Lr/Lg∣zg)=Nlog⁡10(Lr/Lg)[μC(zg),σC2],(20)P(L_{r}/L_{g}|z_{\rm g}) = \mathcal{N}_{\log_{10}(L_{r}/L_{g})}[\mu_{C}(z_{\rm g}),\sigma_{C}^{2}], \tag*{(20)}

with a redshift-dependent mean defined as

μC(zg)=μC,0+ζC(zg−0.5).(21)\mu_{C}(z_{\rm g}) = \mu_{C,0} + \zeta_{C}(z_{\rm g} - 0.5). \tag*{(21)}

Finally, we modelled the size distribution as a Gaussian in the logarithm of the half-light radius,

P(Re∣zg,Lr)=Nlog⁡10(Reh/kpc)[μR(zg,Lr),σR2].(22)P(R_{\rm e}|z_{\rm g},L_{r}) = \mathcal{N}_{\log_{10}(R_{\rm e}h/{\rm kpc})}[\mu_{R}(z_{\rm g},L_{r}),\sigma_{R}^{2}]. \tag*{(22)}

The mean of this Gaussian scales with luminosity and redshift,

μR(zg,Lr)=μR,0+ζR(zg−0.5)+ρL,1log⁡10(Lr/Lt){1−s[bLlog⁡10(Lr/Lt)]}+ρL,2log⁡10(Lr/Lt)s[bLlog⁡10(Lr/Lt)],(23)\begin{aligned} \mu_{R}(z_{\rm g},L_{r}) ={}& \mu_{R,0} + \zeta_{R}(z_{\rm g} - 0.5) \\ &+ \rho_{L,1}\log_{10}(L_{r}/L_{t})\{1-s[b_{L}\log_{10}(L_{r}/L_{t})]\} \\ &+ \rho_{L,2}\log_{10}(L_{r}/L_{t})s[b_{L}\log_{10}(L_{r}/L_{t})], \tag*{(23)} \end{aligned}

where ss is the sigmoid function defined in Eq. (6).

We fitted the parameters of the stellar distribution, P∗P_{*}, to the sample of non-lens galaxies described in Sect. 3.2. The best-fit values of the parameters are reported in Table 3. We express rr-band luminosities in terms of the solar value, which we take to be Lr,⊙=6.00×1011 W Hz−1L_{r,\odot} = 6.00 \times10^{11}\ {\rm W\,Hz}^{-1} (Willmer 2018). The distribution P∗P_{*} is also shown in Figure 6 (blue contours).

Parameter(s)Posterior (68%)Eq. def.Description
z0z_00.32±0.130.32 \pm0.13(17)Scale parameter of the marginal distribution in redshift.
aza_z2.4±0.82.4 \pm0.8(17)Exponent of the marginal distribution in redshift.
czc_z1.55±0.271.55 \pm0.27(17)Shape parameter of the marginal distribution in redshift.
log⁡10[L⋆,0h2/Lr,⊙]\log_{10}[L_{\star,0}h^2/L_{r,\odot}]10.55±0.0410.55 \pm0.04(19)Scale of the luminosity function at z=0.5z=0.5.
ζL\zeta_L0.33±0.160.33 \pm0.16(19)Redshift dependence of the luminosity scale parameter.
α\alpha−0.98±0.03-0.98 \pm0.03(18)Faint-end slope of the luminosity function.
μC,0\mu_{C,0}0.316±0.0010.316 \pm0.001(21)Average colour at z=0.5z=0.5.
ζC\zeta_C−0.078±0.003-0.078 \pm0.003(21)Redshift dependence of the average colour.
σC\sigma_C0.0101±0.00040.0101 \pm0.0004(20)Intrinsic scatter in colour, at fixed redshift.
μR,0\mu_{R,0}0.40±0.180.40 \pm0.18(23)Average size at z=0.5z=0.5 and Lr=LtL_r=L_t.
ζR\zeta_R−0.36±0.06-0.36 \pm0.06(23)Scaling of the average size with redshift.
ρL,1\rho_{L,1}0.34±0.090.34 \pm0.09(23)Faint-end slope of the luminosity-size relation.
ρL,2\rho_{L,2}0.93±0.210.93 \pm0.21(23)Bright-end slope of the luminosity-size relation.
log⁡10[Lth2/Lr,⊙]\log_{10}[L_t h^2/L_{r,\odot}]10.48±0.2810.48 \pm0.28(23)Transition luminosity of the luminosity-size relation
bLb_L0.20±0.260.20 \pm0.26(23)Rapidity of the transition of the luminosity-size relation.
σR\sigma_R0.187±0.0060.187 \pm0.006(22)Intrinsic scatter in size, at fixed redshift and luminosity.

Table 3. Parameters of the galaxy distribution in redshift, luminosity, colour, and size. Notes. The second column reports the median and 68% credible interval of the marginal posterior probability. The third column indicates the equation where each parameter is first defined.

We describe the distribution in M5M_{5} of the population of elliptical galaxies as a Gaussian in log⁡10(M5h/M⊙)\log_{10}(M_{5}h/M_{\odot}), where the factor hh follows from the fact that, for a given angular Einstein radius, the Einstein mass is inversely proportional to the Hubble constant. The model is

PM(M5∣zg,Lr,Re)=Nlog⁡10(M5h/M⊙)[μM(zg,Lr,Re),σM2],(24)P_{M}(M_{5}|z_{\rm g},L_{r},R_{\rm e}) = \mathcal{N}_{\log_{10}(M_{5}h/M_{\odot})}[\mu_{M}(z_{\rm g},L_{r},R_{\rm e}),\sigma_{M}^{2}], \tag*{(24)}

with mean

μM(zg,Lr,Re)=μM,0+βz(zg−0.5)+βL[log⁡10(Lrh2/Lr,⊙)−10.7]+βR[log⁡10(Reh/kpc)−0.5].(25)\mu_M(z_g,L_r,R_e)=\mu_{M,0}+\beta_z(z_g-0.5) +\beta_L[\log_{10}(L_r h^2/L_{r,\odot})-10.7] +\beta_R[\log_{10}(R_e h/\mathrm{kpc})-0.5]. \tag*{(25)}

In summary, the set of model parameters that we wish to infer from the Euclid strong lensing data is

η=(μM,0,βz,βL,βR,σM).(26)\boldsymbol{\eta}=(\mu_{M,0},\beta_z,\beta_L,\beta_R,\sigma_M). \tag*{(26)}

This is the simplest model that allows us to reach the main goal of our work, which is to determine the average M5M_5 of the population of elliptical galaxies, and its scaling with redshift, luminosity, and size. We consider the model satisfactory if it can reproduce the observed distribution in Einstein radius of our lens sample. We verify whether this requirement is met with posterior predictive tests, as explained in Sect. 6.

Background source population

We modelled the distribution in flux and redshift space of the background source population. We started from the observed galaxy distribution in rest-frame UV (1500Å) luminosity and redshift space, as measured by Parsa et al. [50]. We discuss the impact of the choice of template in Sect. 7. Then, for a given value of the UV luminosity and the redshift, we computed the IEI_{\mathrm E} flux using the spectral energy distribution template SB1, from the Kinney–Calzetti atlas, as provided by LePhare. Figure 7 (filled contours) shows the resulting prior distribution in redshift and IEI_{\mathrm E}-flux space, Ps(zs,Fs,IE)P_s(z_s,F_{s,I_{\mathrm E}}).

Figure 7. Prior probability distribution of the background source population, in redshift and IEI_{\mathrm E} flux space. Filled contours: distribution before lensing (from Parsa et al. [50]). Solid lines: lensed distribution. Dashed histogram (bottom-right panel): observed distribution in source flux. The three stars indicate the three lensed sources for which we have a spectroscopic redshift. The lensed distribution contours are obtained using the best-fit model parameters, and therefore do not take into account the uncertainty on the lens population properties.

Given Ps(zs,Fs,IE)P_s(z_s,F_{s,I_{\mathrm E}}), the posterior probability of a lensed source given a model population η\boldsymbol{\eta} and the other observables is

P(zs∣θE,Fs,IE∣{Fg,λ},θE,η)∝Ps(zs∣Fs,IE)×∫P(θE∣ψg,zs)P({Fg,λ},θE∣ψg)Pg(ψg∣η)σSL(ψg,ψs) dψg,(27)\begin{aligned} P(z_s|\theta_{\mathrm E},F_{s,I_{\mathrm E}}|\{F_g,\lambda\},\theta_{\mathrm E},\boldsymbol{\eta}) \propto P_s(z_s|F_{s,I_{\mathrm E}}) \\ &\times\int P(\theta_{\mathrm E}|\psi_g,z_s)P(\{F_g,\lambda\},\theta_{\mathrm E}|\psi_g)P_g(\psi_g|\boldsymbol{\eta})\sigma_{\mathrm{SL}}(\psi_g,\psi_s)\,\mathrm{d}\psi_g, \tag*{(27)} \end{aligned}

which we refer to as the photogeometric redshift of the source. Figure 8 shows the photogeometric redshift of the three lensed sources with spectroscopic redshifts, given the redshift of the lens, and marginalised over the uncertainty on the model parameters η\boldsymbol{\eta}. In all cases, the photogeometric redshift is consistent with the spectroscopic redshift. Figure 8 also shows the prior on zsz_s conditional on the flux, Ps(zs∣Fs,IE)P_s(z_s|F_{s,I_{\mathrm E}}). In the case of EUCL J171507.51+653150.6, this term dominates the posterior. This is because that source is particularly bright, and flux information alone rules out solutions with a high redshift. For the other two sources, the addition of the Einstein radius constraint shifts the distribution to higher redshifts, because lower redshift solutions require a particularly large, hence unlikely, lens mass.

Figure 8. Photogeometric redshifts of the three lensed sources with a spectroscopic redshift. The blue colour histogram shows the prior probability Ps(zs∣Fs,IE)P_s(z_s|F_{s,\mathrm{IE}}). The gold colour histogram shows P(zs∣zg,θE,Fs,IE,{Fg,λ},θe)P(z_s|z_g,\theta_E,F_{s,\mathrm{IE}},\{F_{g,\lambda}\},\theta_e), obtained from Eq. (27) by marginalising over the parameters η\boldsymbol{\eta}, given the posterior probability of the model.

Lensing cross-section

In order to compute the normalisation of the probability distribution, Eq. (13), we need an expression for the lensing cross-section. The lensing cross-section depends primarily on the mass distribution of the lens, on its redshift, on the redshift of the background source, and on the flux of the source. However, it also depends on the surface brightness distribution of the foreground galaxy, because our definition of a golden lens requires a minimum contrast between the source and lens light.

We can compute σSL\sigma_{\mathrm{SL}} by means of simulation. For a given set of properties of the lens galaxy and of the background source, we can draw a large number of source positions, sampled uniformly in a sufficiently large region of the sky, and, for each configuration, create a mock Euclid image. We can then check, for each image, if the resulting lens satisfies our golden lens definition. The final value of σSL\sigma_{\mathrm{SL}} is the ratio between the number of golden lenses to the total number of generated images, multiplied by the source plane area from which the source positions were sampled. However, doing so for every possible combination of lens and source parameters, and for any possible value of the model parameters η\boldsymbol{\eta}, is unpractical. Therefore, we made a few simplifying assumptions. We fixed the axis ratio of the lens light and lens mass to the same value of q=0.8q=0.8 (the lensing cross-section has a very mild dependence on axis ratio, as shown by Sonnenfeld et al. [62]). Then, for the purposes of generating images of background sources, we assumed a fixed scaling relation between the angular size of the source and its flux,

log⁡10(θs,e/1′′)=−0.850+0.418log⁡10(Fs,IE/μJy).(28)\log_{10}(\theta_{s,e}/1^{\prime\prime})=-0.850+0.418\log_{10}(F_{s,\mathrm{IE}}/\mu\mathrm{Jy}). \tag*{(28)}

We obtained the above relation by fitting the flux–size distribution of the background sources in our sample with a linear model. Finally, we made the assumption that the lensing cross-section scales with Einstein radius as

σSL(θE,Fg,IE,θe,Fs,IE)σSL(θE∗,Fg,IE,θe,Fs,IE)=(θEθE∗)2,(29)\frac{\sigma_{\mathrm{SL}}(\theta_E,F_{g,\mathrm{IE}},\theta_e,F_{s,\mathrm{IE}})}{\sigma_{\mathrm{SL}}(\theta_E^\ast,F_{g,\mathrm{IE}},\theta_e,F_{s,\mathrm{IE}})}=\left(\frac{\theta_E}{\theta_E^\ast}\right)^2, \tag*{(29)}

where θE∗\theta_E^\ast is a reference value of the Einstein radius. Equation (29) says that, for two lenses with identical lens light distribution and source flux, the ratio of their lensing cross-section is equal to the ratio of their Einstein radius squared. This is the theoretical scaling of the lensing cross-section for an SIE lensing a point source, and we expect it to hold in our regime as well. The investigation of the lensing cross-section of lenses with extended lensed sources by Sonnenfeld et al. [62] supports this claim. One important exception to Eq. (29) is if θE\theta_E falls below our minimum Einstein radius of 0\farcs7, in which case σSL=0\sigma_{\mathrm{SL}}=0.

Practical implementation

We sampled the posterior probability distribution P(η∣d)P(\boldsymbol{\eta}|\boldsymbol{d}) with a Markov chain Monte Carlo (MCMC). At each draw of η\boldsymbol{\eta} and for each lens, this requires computing integrals of the kind of Eq. (10). To simplify calculations, we assumed that all of the observables, including the Einstein radius, are measured exactly. This is a good approximation, as the typical uncertainty on θE\theta_E is on the order of a few % (Etherington et al. [20]), which is much smaller than the uncertainty on the mass model parameters, and also much smaller than the intrinsic scatter in M5M_5. The contribution to the likelihood of a lens without a spectroscopic redshift of neither the lens nor the source is

P(θE,{Fg,λ},θe,Fs,IE∣η)=∫Pg[zg,{Fg,λ},θe,M5(θE,zg,zs)∣η]Ps(zs∣Fs,IE)∣dM5dθE∣dzgdzs,(30)P(\theta_E,\{F_{g,\lambda}\},\theta_e,F_{s,\mathrm{IE}}|\boldsymbol{\eta})=\int P_g[z_g,\{F_{g,\lambda}\},\theta_e,M_5(\theta_E,z_g,z_s)|\boldsymbol{\eta}]P_s(z_s|F_{s,\mathrm{IE}})\left|\frac{\mathrm{d}M_5}{\mathrm{d}\theta_E}\right|\mathrm{d}z_g\mathrm{d}z_s, \tag*{(30)}

where we used Psel=1P_{\mathrm{sel}}=1 (i.e. we assumed the lens sample to be effectively complete), and the last factor in the integrand is the result of the integration over M5M_5, which we performed analytically in virtue of the likelihood on the Einstein radius being a Dirac delta function. We computed the integrals above (one per each lens) by Monte Carlo integration, drawing samples of zgz_g from the lens photo-zz posterior probability, and samples of zsz_s from the distribution in source redshift, conditional on the source flux. For each zg<zsz_g<z_s pair, we computed the value of M5M_5 necessary to match the observed Einstein radius. For zg>zsz_g>z_s, the integrand is zero.

To compute the normalisation of the probability distribution, Eq. (13), we proceeded as follows. We drew a sample of 10 000 lens and source pairs from PSL(φg,φs∣η∗)P_{\mathrm{SL}}(\boldsymbol{\varphi}_g,\boldsymbol{\varphi}_s|\boldsymbol{\eta}^\ast), where η∗\boldsymbol{\eta}^\ast is a reference value of the model parameters. For each lens-source pair, we computed the lensing cross-section following the procedure described in Sect. 5.3. Then, for any given value of η\boldsymbol{\eta}, we drew samples of (φg,φs)(\boldsymbol{\varphi}_g,\boldsymbol{\varphi}_s) from the same random seed. This created a new sample with identical lens light and source properties, and with a shifted distribution in M5M_5. We made use of the σSL\sigma_{\mathrm{SL}} scaling approximation of Eq. (29) to compute the lensing cross-section for this new sample. Finally, we computed the integral of Eq. (13) by Monte Carlo integration. We iterated this procedure, each time setting η∗\eta^* to a value close to the peak of the posterior probability. The final results converged after two iterations.

We assumed uniform priors on all of the mass model parameters, over the ranges listed in Table 4. We fixed the parameters of the photo-zz model and of the distribution in redshift, luminosity and size to the median of their marginal posterior probability.

ParameterPriorPosterior (68%)Eq. def.Description
μM,0\mu_{M,0}U(10.5,11.8)U(10.5, 11.8)11.300±0.02111.300 \pm0.021(25)Average log⁡10(M5hM⊙−1)\log_{10}(M_5 h M_\odot^{-1}) for a galaxy at z=0.5z = 0.5, with Lr=1010.7Lr,⊙h−2L_r = 10^{10.7}L_{r,\odot}h^{-2} and log⁡10(Reh/kpc)=0.5\log_{10}(R_e h/\mathrm{kpc}) = 0.5.
σM\sigma_MU(0.02,0.2)U(0.02, 0.2)0.099±0.0100.099 \pm0.010(24)Intrinsic scatter in log⁡10(M5hM⊙−1)\log_{10}(M_5 h M_\odot^{-1}) at fixed redshift, luminosity, and size.
βL\beta_LU(−3,3)U(-3, 3)0.71±0.070.71 \pm0.07(25)Power-law scaling of M5M_5 with luminosity.
βR\beta_RU(−3,3)U(-3, 3)−0.15±0.07-0.15 \pm0.07(25)Power-law scaling of M5M_5 with size.
βz\beta_zU(−3,3)U(-3, 3)−0.33±0.05-0.33 \pm0.05(25)Power-law scaling of M5M_5 with redshift.

Table 4. Inference on the mass model parameters.

Results

Table 4 reports the median and 68% credible interval of the marginal posterior probability distribution of the model parameters. The full probability distribution is shown in Figure 9 by means of pairwise projections. The parameter μM,0\mu_{M,0}, which describes the average log⁡10(M5h/M⊙)\log_{10}(M_5 h/M_\odot) at a reference luminosity, size, and redshift, is recovered with a 68% uncertainty of 0.02. This implies a 5% uncertainty on the mass-to-light ratio. We see a negative covariance between μM,0\mu_{M,0} and the intrinsic scatter parameter σM\sigma_M, which can be explained as follows. While our model describes the general population of galaxies, lensing data samples only the high mass tail of the galaxy distribution. A model with an overall lower mass and higher intrinsic scatter can produce a lens population with similar Einstein radius distribution to that of a model with a higher mass and smaller scatter. Therefore, if the only lensing-related constraints are those provided by θE\theta_{\rm E}, these two scenarios are degenerate.

Figure 9. Posterior probability distribution of the parameters of the model. The first two columns show the posterior probability of the average redshift and flux of the lensed sources (predicted from the model, and not fitted for). Blue filled contours show the fiducial inference, obtained under the SIE model assumption. Green dashed contours are obtained with a power-law density profile with γ=1.9\gamma= 1.9. Black dashed contours are obtained with a power-law density profile with a slope evolving as γ(z)=2−0.3(z−0.5)\gamma(z) = 2 - 0.3(z - 0.5). Red dashed contours are obtained by repeating the fiducial analysis, but with a different choice for the spectral template used to describe the population of background sources. These three alternative models are introduced in Sect. 7. Contour levels correspond to 68% and 95% enclosed probability. Dotted lines indicate the observed average source flux, under the assumption of an SIE (blue colour) or γ=1.9\gamma= 1.9 lens model (green colour).

We detected a positive correlation of M5M_5 with luminosity, βL=0.71±0.07\beta_L = 0.71 \pm0.07. This value is consistent with the scaling between stellar mass and the total projected mass enclosed within 3.5 h−1h^{-1} kpc measured by Sonnenfeld [60]. We found a preference for a negative value of the mass–size correlation parameter, βR=−0.15±0.07\beta_R = -0.15 \pm0.07, although it is strongly degenerate with the luminosity dependence parameter, βL\beta_L.

We combined these results to compute the mass-to-light ratio of elliptical galaxies as a function of luminosity. In particular, we computed the mean ratio between mass and luminosity, each within an aperture of 5 h−1h^{-1} kpc, of galaxies at zg=0.5z_{\rm g} = 0.5, and with average size for their luminosity. The result is shown in Figure 10. At a luminosity log⁡10(Lrh2/L⊙)=10.7\log_{10}(L_r h^2/L_\odot) = 10.7 (close to the average of the golden lens sample), the mean mass-to-light ratio within 5 h−1h^{-1} kpc is (6.7±0.3) M⊙hLr,⊙−1(6.7 \pm0.3)\,M_\odot h L_{r,\odot}^{-1}. For small luminosities, the mass-to-light ratio is a decreasing function of LrL_r. This is a consequence of the luminosity-dependence parameter βL\beta_L being smaller than one. In galaxies of decreasing luminosity, an aperture of 5 h−1h^{-1} kpc corresponds to a relatively larger radius compared to the half-light radius, and therefore to a region more and more dominated by dark matter. The upturn in M/LrM/L_r at high LrL_r is instead driven by the steepening of the luminosity-size relation. At a given luminosity, a larger half-light radius means that the fraction of light enclosed within a fixed physical aperture decreases, while M/LrM/L_r increases.

Mean mass-to-light ratio as a function of galaxy luminosity

Figure 10. Mean mass-to-light ratio, integrated within an aperture of 5h−1 kpc5h^{-1}\,\mathrm{kpc}, as a function of galaxy luminosity (blue bands). The bands enclose 68% and 95% of the probability. The model assumes a fixed redshift zg=0.5z_{\mathrm{g}} = 0.5, and average size for the luminosity. The three curves show theoretical predictions from models consisting of stars and a dark matter halo. We described the stars with an SPS model, obtained from Galaxev ([11]). We assumed solar metallicity, an exponentially declining star formation history with timescale τ=1 Gyr\tau= 1\,\mathrm{Gyr}, 7 Gyr since the initial burst, and minimum dust (τV=0.04\tau_{\mathrm{V}} = 0.04). We adopted two choices for the stellar IMF, as indicated in the legend. For the dark matter halo, we assumed an NFW profile or an adiabatically contracted (AC) profile.

In order to interpret these results, we show in Figure 10 theoretically predicted mass-to-light ratios, obtained from models consisting of the sum of a stellar component and a dark matter halo. We modelled the stellar component with stellar population synthesis (SPS) models obtained from the software Galaxev ([11]). We assumed solar metallicity, an exponentially declining star formation history, and minimal dust. We adopted both a Chabrier and a Salpeter stellar initial mass function (IMF). For each value of the luminosity, we obtained the corresponding stellar mass given the SPS model. Then, using the stellar-to-halo mass relation measured by Shuntov et al. [57], we set the dark matter halo mass MhM_h to the average log⁡10(Mh/M⊙)\log_{10}(M_h/M_\odot) of galaxies of the given stellar mass. We considered dark matter halos with a Navarro, Frenk & White (NFW; Navarro et al. [45]) profile with a mass-concentration relation from Dutton & Macciò [19], as well as adiabatically contracted profiles, obtained with the prescription of Blumenthal et al. [7]. The models straddle the range of values of mass-to-light ratio allowed by the data. This indicates that our measurements can potentially rule out regions in the model parameter space defined by the stellar IMF and the dark matter profile. However, given that we set the stellar population parameters arbitrarily when building these models, we refrain from drawing quantitative conclusions from this comparison.

Figure 10 also shows dynamical measurements obtained from the work of Tortora et al. [67], for comparison. Tortora et al. [67] estimated the mass of a large sample of massive [log⁡10(M∗h2/M⊙)>10.95][\log_{10}(M_* h^2/M_\odot) > 10.95] early-type galaxies, using their stellar velocity dispersion as a constraint. They, too, assumed an isothermal density profile, and used the physical Jeans equation to constrain their mass. The Tortora et al. [67] values are, on average, consistent with our measurements. However, they show a significantly steeper decreasing trend with luminosity. The reason for the sharper decrease reflects a stronger sensitivity of the dynamical estimates to the half-light radius. For a singular isothermal sphere, M5∝σSIS2M_5 \propto\sigma_{\rm SIS}^2, where σSIS\sigma_{\rm SIS} describes the velocity dispersion of the massive particles that form the distribution. In the context of Jeans modelling, σSIS\sigma_{\rm SIS} is, to first approximation, similar to the velocity dispersion of the tracer particles (i.e. the stars). At fixed stellar mass, the stellar velocity dispersion is a decreasing function of ReR_{\rm e} (e.g. Bernardi et al. [5]). As a result, the inferred M5M_5 appears to decrease with increasing ReR_{\rm e}. By contrast, our measurement shows a very mild dependence of M5M_5 on ReR_{\rm e}, quantified by the parameter βR=−0.15±0.07\beta_R = -0.15 \pm0.07. Since higher-luminosity galaxies have on average larger sizes, this different sensitivity to ReR_{\rm e} is responsible for the difference in slope observed in Figure 10. A problem with the dynamical measurements is that, in practice, the stellar velocity dispersion depends not only on the overall mass, but also on the slope of the density profile. At fixed M5M_5, it is larger for galaxies with a steeper density slope. More compact galaxies have a steeper density slope ([63]), and therefore a larger stellar velocity dispersion. But, under the isothermal assumption, the dependence on the density slope is absorbed into a dependence on ReR_{\rm e}, introducing a bias. In contrast, our lensing-based measurements are less sensitive to the choice of the density profile, as we show in Sect. 7. Nevertheless, the fact that the dynamical data points are, on average, in agreement with our estimate, is an important consistency check.

We also found a negative correlation between redshift and mass, at fixed luminosity. This is expected, because the luminosity of elliptical galaxies decreases with time, as their stellar populations become older. The result is that, at the same luminosity, lower redshift galaxies are more massive. Rather than comparing the properties at fixed luminosity, a more interesting description is one that follows the evolution of M5M_5 along the evolutionary tracks of galaxies. However, matching progenitors and descendants in a sample of galaxies is not trivial. Elliptical galaxies grow mainly as a result of mergers, but the amount of growth is uncertain ([43]; [12]; [1]). We considered two extreme scenarios. One in which there is no growth between z=1z = 1 and z=0z = 0, and one in which the stellar mass grows as Δlog⁡10(M∗/M⊙)=0.3\Delta\log_{10}(M_*/M_\odot) = 0.3 (roughly a factor of two increase over the same period). To further simplify the problem, we assumed that the stellar population of the elliptical galaxies and of the accreted stars were all generated in a single burst at z=2.5z = 2.5. We used Galaxev to compute the luminosity evolution in the two cases. The result is shown in

Notes. The second column reports the assumed prior probability distribution, where the notation U(x,y)U(x, y) indicates a uniform distribution between xx and yy. The third column reports the median and 68% credible interval of the marginal posterior probability. The fourth column indicates the equation where each parameter is first defined.

Figure 11. The top panel shows Lpass(z)L_{\mathrm{pass}}(z), which we normalised so that log⁡10[Lpass(0.5)h2/Lr,⊙]=10.7\log_{10}[L_{\mathrm{pass}}(0.5)h^2/L_{r,\odot}] = 10.7. In the no-growth scenario, the luminosity drops by 0.36 dex from z=1z = 1 to z=0z = 0, while in the rapidly growing case it decreases by 0.10 dex. The bottom panel shows the corresponding evolution in the average M5M_5, assuming that, at each redshift, the size stays equal to the average for galaxies of the same luminosity (middle panel). In the case of no growth, the data are consistent with a non-evolving mass.

Evolution of individual galaxies under two stellar mass growth rates

Figure 11. Evolution of individual galaxies, under the assumption of a purely merger-induced growth. Two stellar mass growth rates are considered. Top: luminosity evolution, assuming a simple stellar population formed at z=2.5z = 2.5, with solar metallicity. We adopted H0=70 km s−1 Mpc−1H_{0} = 70\,\mathrm{km\,s^{-1}\,Mpc^{-1}} to convert stellar ages to redshift. Middle: average size of galaxies with luminosity equal to LpassL_{\mathrm{pass}}. Bottom: average log⁡10(M5 h/M⊙)\log_{10}(M_{5}\,h/M_{\odot}) of galaxies with luminosity equal to LpassL_{\mathrm{pass}}, and average size for the luminosity, as a function of redshift. Regions correspond to 95% and 68% enclosed probability.

The inferred variation in the average M5M_{5} between z=1z = 1 and z=0z = 0 is (−0.01±0.05)(-0.01 \pm0.05) dex. In the rapidly growing case, the variation is (0.18±0.05)(0.18 \pm0.05) dex. Although we cannot distinguish between these two scenarios, we can conclude that, if there is any merger-induced growth between z=1z = 1 and z=0z = 0, M5M_{5} increases along the progenitors. Using the rapidly growing case as a reference, we estimate the response in M5M_{5} to an increase in stellar mass to be approximately dlog⁡10(M5/M⊙)/dlog⁡10(M∗/M⊙)=0.54±0.15\mathrm{d}\log_{10}(M_{5}/M_{\odot})/\mathrm{d}\log_{10}(M_{*}/M_{\odot}) = 0.54 \pm0.15.5 This measurement can be used to test models of the merger evolution of elliptical galaxies. For example, Liu et al. [42] showed that major and minor mergers have a different impact on the inner stellar mass distribution of elliptical galaxies, for the same amount of overall stellar mass growth.

Finally, we investigated how the lens population differs from the general population of galaxies in terms of mass. The former is given by the left-hand side of Eq. (1), while the latter is described by Pg(ψg)P_{\mathrm{g}}(\psi_{\mathrm{g}}). Figure 12 shows the predicted distribution of M5M_{5} as a function of luminosity, at a reference redshift zg=0.5z_{\mathrm{g}} = 0.5. At fixed luminosity, lenses are on average more massive than regular galaxies (dashed lines in Figure 12). The difference is 0.08 dex at log⁡10(Lrh2/Lr,⊙)=10.6\log_{10}(L_{r}h^{2}/L_{r,\odot}) = 10.6, and 0.05 dex at the high-luminosity end. Figure 12 only shows the distribution obtained from the best-fit model parameter values. In reality, the amplitude of the difference between the two distributions depends on the value of the intrinsic scatter parameter (larger for larger σM\sigma_{M}). Nevertheless, this result indicates that particular care must be taken in interpreting strong lensing measurements directly, especially if a precision of 20% or better is desired.

Distribution in luminosity and stellar mass of elliptical galaxies and golden lenses

Figure 12. Distribution in luminosity and M5M_{5} of elliptical galaxies and golden lenses. For each sample, contours show the distribution predicted for the best-fit values of the model parameters (i.e. the median values listed in Table 4), for galaxies at zg=0.5z_{\mathrm{g}} = 0.5. Dashed lines indicate the mean of each distribution.

All results in this paper are based on galaxies with no neighbours within 3′′3'' in projection. Therefore, they should be interpreted as applying to the subset of elliptical galaxies with no close projected companions.

We compared our results with the lensing-based measurements of Sonnenfeld [60], who obtained the total projected mass of early-type galaxies within 3.5 h−1 kpc3.5\,h^{-1}\,\mathrm{kpc}, as a function of stellar mass, using strong lenses from the Sloan ACS Lens Survey (SLACS; Bolton et al. [9]). To do so, we first extrapolated the Sonnenfeld [60] measurement to an aperture of 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}, using the power-law density profile model of Sonnenfeld [60]. Second, we obtained from Auger et al. [3] the VV-band luminosity corresponding to the stellar mass of each SLACS lens. Third, we converted from VV- to rr-band luminosity using a spectral template with colour log⁡10(Lr/Lg)=0.35\log_{10}(L_r/L_g)=0.35, which is the typical value for elliptical galaxies at z=0.2z=0.2 (Figure 6). The result is that the average log⁡10(M5h/M⊙)\log_{10}(M_{5h}/M_\odot) at log⁡10(Lrh2/Lr,⊙)=10.67\log_{10}(L_r h^2/L_{r,\odot})=10.67 is 11.33±0.0111.33\pm0.01. Our model predicts a value of 11.38±0.0311.38\pm0.03 for galaxies of the same luminosity and redshift, therefore is within two standard deviations of the Sonnenfeld [60] value.

We checked the goodness of fit of our model by investigating how well it can reproduce the observed distribution in Einstein radius, by means of a posterior predictive test. In particular, we used the 10%-, the 50%-, and the 90%-ile of the distribution in θE\theta_{\mathrm{E}} as test quantities. We drew samples from the posterior probability of the model parameters. For each draw, we generated a population of 95 golden lenses and measured the three test quantities. Figure 13 shows the resulting posterior predicted distributions. The percentages on the left and right of each panel indicate the probability that a smaller or larger value than the observed one would have occurred by chance, under the assumption of the model being correct. Although the model tends to predict larger values of the percentiles of the Einstein radius distribution, the chance of observing even smaller values by chance is reasonably large (5.7% for the 90%-ile, which is the test quantity with the worst agreement). Therefore the test gives us no reason to question the goodness of fit of the model to the Einstein radius distribution.

Goodness-of-fit posterior predictive test

Figure 13. Goodness-of-fit posterior predictive test. Each panel shows a posterior predicted percentile of the Einstein radius distribution, as indicated on the horizontal axis. The percentages indicate the fraction of samples for which the posterior predicted value is smaller or larger than the observed value.

Discussion

Our study is based on a sample of lenses with minimal spectroscopic coverage, which limited our ability to obtain precise measurements of mass on an individual lens basis. We overcame this problem with statistics, and with two key modelling choices. The first one is the assumption of a singular isothermal density profile for the mass distribution. To investigate the impact of this assumption, we repeated the analysis with a power-law density profile, ρ(r)∝r−γ\rho(r)\propto r^{-\gamma}, with γ\gamma set to plausible values that are consistent with measurements from the literature. One caveat is that our inference of the fluxes of the lensed sources also depends on the choice of the density profile, because it relies on a model for the lensing magnification. Our measurement of Fs,IEF_{\mathrm{s,IE}} was obtained using an SIE model. In order to be self-consistent with the new power-law assumption, we multiplied them by γ−1\gamma-1, which is the ratio between the radial magnification of an SIE and that of a power-law lens, evaluated at the Einstein radius. In other words, we accounted for the fact that, if we had modeled our lenses with a power-law model, we would have inferred smaller intrinsic source fluxes, because of its larger radial magnification compared to an SIE.

We first fixed γ=1.9\gamma=1.9 for all galaxies. This is close to the value inferred by Bolton et al. [8] for galaxies at z=0.6z=0.6. The resulting posterior probability distribution is shown in Figure 9 as dashed green curves. The inference on the average mass parameter, μM,0\mu_{M,0}, is virtually unchanged, while that on the other parameters is consistent with the fiducial model. We also considered a power-law model with a density slope that evolves as γ(z)=2.0−0.3(z−0.5)\gamma(z)=2.0-0.3(z-0.5), which is very close to the trend measured by Sonnenfeld et al. [63] over the range 0<z<0.80<z<0.8. As Figure 9 shows, the inferred parameter values are still unchanged. Neither of these results are fortuitous coincidences, but rather the consequence of having chosen 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc} as the reference scale on which to measure the total mass. Because the median Einstein radius of the golden lenses is very close to 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}, the data constrain M5M_5 directly, with a very small sensitivity to the choice of the density profile. More extreme deviations from an isothermal profile are in tension with existing constraints from lensing and stellar dynamics (Sonnenfeld et al. [63]). We conclude that the isothermal assumption is currently not limiting the accuracy of our measurement.

We could also modify our density profile with a mass-sheet transformation,

Σ(R)→λmstΣ(R)+(1−λmst)Σcr,(31)\Sigma(R)\to\lambda_{\mathrm{mst}}\Sigma(R)+(1-\lambda_{\mathrm{mst}})\Sigma_{\mathrm{cr}}, \tag*{(31)}

which leaves our lensing observables unchanged (Schneider & Sluse [54]). For small projected distances from the Einstein radius, the projected mass profile of a mass-sheet-transformed isothermal model can be approximated by a power law. Therefore, we expect the inference to be robust also with respect to modest mass-sheet transformations (λmst\lambda_{\mathrm{mst}} close to 1).

One additional caveat with the isothermal assumption, and with the tests described so far, is that we did not allow for scatter in the density profile. Given that the Einstein radii of our lenses have a non-zero spread around 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc}, a scatter in the density slope γ\gamma would be absorbed by the parameter that describes the intrinsic scatter in M5M_5, σM\sigma_M. Therefore, we urge caution in taking our measured value of σM\sigma_M at face value. For a quantitative estimate of the degeneracy between the scatter in density slope and mass, we refer the reader to the analysis of the SLACS sample of Sonnenfeld [60].

The second critical assumption of our model is the choice of the prior in flux-redshift space of the population of background sources, which we based on a measurement of the UV luminosity function of galaxies from Parsa et al. [50]. Our implicit assumption is that the method that we used to measure source fluxes is consistent with that used by Parsa et al. [50]. This assumption, however, is difficult to verify and could introduce biases in the inferred redshift distribution of the background sources. Moreover, we assumed a fixed spectral template, SB1, to convert from rest-frame UV to observed IEI_{\mathrm{E}} flux. To assess the sensitivity of our results to this particular choice, we repeated the analysis while using the template SB6, which has a significantly redder spectrum. The resulting posterior probability is shown in Figure 9 (red dashed contours). The inferred average lensed source redshift is significantly lower than that of the fiducial analysis. However, the values of the model parameters show very small differences, indicating that the choice of source spectral template is not a limiting factor in our analysis. This is not surprising, given that the critical surface mass density is a slowly varying function of source redshift. The SB1 template produces good agreement between the posterior predicted average source flux and the observed one (Figure 7). By contrast, when using the SB6 template, the posterior predicted average source flux is larger than the observed one by 0.14 dex (Figure 9). For this reason, we chose the SB1 template for our fiducial analysis. Nevertheless, our understanding of the background source population is still subject to systematic uncertainties. We have an ongoing spectroscopic follow-up programme. With the aid of more spectroscopic observations, we plan to measure the distribution of the background source population directly on Euclid data.

Another assumption that we made is that of the effective completeness of our sample of golden lenses. That is, we assumed that our sample is missing golden lenses at random, with no correlation between the properties of a lens and the probability of us including it in our sample. One way to test this assumption is by examining additional aspects of the properties of the lenses, such as the inferred source position β\beta and the total magnification ∣μtot∣|\mu_{\rm tot}|. We defined the latter as the ratio between the total flux of the lensed source and the inferred intrinsic flux. Figure 14 shows ∣μtot∣|\mu_{\rm tot}| as a function of ∣β∣/θE|\beta|/\theta_{\rm E} for the observed sample, along with the distribution predicted with our fiducial model (blue contours). Lenses with a larger value of ∣β∣/θE|\beta|/\theta_{\rm E} have a more asymmetric image configuration, with a main arc well outside of the critical curve and a fainter counter-image near the lens centre, and smaller total magnification. At the oposite limit, lenses with ∣β∣/θE≈0|\beta|/\theta_{\rm E} \approx0 produce Einstein rings, with a large magnification. Qualitatively, the model reproduces this trend. However, it predicts a tail of low-magnification and large source-lens separation that is not observed in our sample (most clearly seen in the histograms of Figure 14). Such lenses must exist in reality, but are evidently identified with a lower efficiency compared to the highly-magnified ones. This means that our assumption of the golden lens sample being effectively complete is not entirely accurate. In order to understand the impact of this inaccuracy on our measurements, we repeated our analysis with an alternative model for the strong lensing cross-section. In addition to the criteria listed in Sect. 3.1, we imposed a minimum magnification of ∣μtot∣=5|\mu_{\rm tot}| = 5, for a system to be classified as a lens. Nine of the 95 golden lenses have a measured magnification smaller than this threshold (nevertheless, we still included them in this new analysis). This new model provides a better match to the observed distribution in magnification and source position (Figure 14), and the inferred model parameters are virtually identical to the fiducial case (Figure 9). This is not surprising, because, for the purpose of inferring the mass distribution of the foreground galaxy population, the key property of the selection function model is the relative probability of galaxies of different mass to be strong lenses. Adding or removing a low-magnification tail to the sample changes the total number of lenses, but does not affect this property. Nevertheless, the test suggests that the magnification may play a role in the detectability of a lens, giving us an important clue on how to improve our knowledge of the selection function. However, given the low sensitivity of our measurements to the details of the lensing cross-section model, we do not attempt to refine it further, at this stage.

Figure 14. Distribution in total magnification and projected distance between the source and the lens, normalised by the Einstein radius. Orange stars and histograms: observations. Blue: prediction from the fiducial model, with each population parameter fixed to the median value of its marginal posterior probability. Magenta: prediction from a model with a modified lensing cross-section, with σSL=0\sigma_{\rm SL}=0 for lenses with total magnification ∣μtot∣<5|\mu_{\rm tot}| < 5.

We also plan to improve our model for the distribution of elliptical galaxies in redshift–flux–colour space. The current model is sufficiently accurate for the purposes of this work, but is somewhat simplistic. As we argued in Sect. 4, the limiting factor is the availability of spectroscopic redshifts for the lens galaxies, which we need to test the accuracy of our photo-zs. With a larger pool of spectroscopic observations, both of lens and non-lens elliptical galaxies, we can increase the complexity of the model, possibly adopting a non-analytical approach, such as a self-organising map (e.g. Wright et al. [70]). We can also include ground-based photometric observations to improve the precision of individual photo-zs.

Conclusions

We measured the total projected mass enclosed within an aperture of 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc} of isolated elliptical galaxies, over the redshift range 0<z<10 < z < 1. Although our measurement is based on strong lenses, which are a biased subset of the full galaxy population, we corrected for selection effects and constrained the properties of galaxies before lensing. This is one of very few such measurements obtained so far, and the one that covers the broadest redshift range.

We used our measurement to compute the total mass-to-light ratio of elliptical galaxies as a function of their properties. We found that galaxies at z=0.5z = 0.5, with luminosity log⁡10(Lrh2/L⊙)=10.7\log_{10}(L_r h^2/L_{\odot}) = 10.7, and average size for their luminosity, have a ratio between their total mass within 5 h−1 kpc5\,h^{-1}\,\mathrm{kpc} and the total luminosity within the same aperture of (6.7±0.3) M⊙h/L⊙(6.7 \pm0.3)\,M_{\odot}h/L_{\odot}. We also measured how M5M_5 varies with time. Although the exact evolutionary tracks of massive elliptical galaxies are uncertain, we can conclude that, in a scenario with a purely merger-driven growth and a passively evolving stellar population, a growth in stellar mass brings along a corresponding increase in M5M_5.

According to our model, strong lens galaxies have, on average, a 10–20% larger M5M_5 compared to non-lens galaxies, with the difference being more pronounced at lower luminosity. Since the precision of mass measurements from strong lensing is typically higher than this value, it is important to account for selection effects when generalising measurements made on strong lenses to the full population of galaxies. This statement applies to studies based on statistical samples of lenses or individual lenses alike.

In summary, our measurement provides an observational benchmark with which to test models of galaxy structure, a unique one for the broad range of its redshift coverage, 0<z<10 < z < 1. These measurements were obtained using data from the Q1 area of Euclid, which constitutes less than 0.5% of the total area that will be covered by the Euclid Wide Survey. As the amount of data increases, we will use Euclid strong lensing observations to investigate an increasing number of aspects of galaxy structure.

Acknowledgements. The Euclid Consortium acknowledges the European Space Agency and a number of agencies and institutes that have supported the development of Euclid, in particular the Agenzia Spaziale Italiana, the Austrian Forschungsförderungsgesellschaft funded through BMIMI, the Belgian Science Policy, the Canadian Euclid Consortium, the Deutsches Zentrum für Luft- und Raumfahrt, the DTU Space and the Niels Bohr Institute in Denmark, the French Centre National d’Etudes Spatiales, the Fundação para a Ciência e a Tecnologia, the Hungarian Academy of Sciences, the Ministerio de Ciencia, Innovación y Universidades, the National Aeronautics and Space Administration, the National Astronomical Observatory of Japan, the Netherlandse Onderzoekschool Voor Astronomie, the Norwegian Space Agency, the Research Council of Finland, the Romanian Space Agency, the State Secretariat for Education, Research, and Innovation (SERI) at the Swiss Space Office (SSO), and the United Kingdom Space Agency. A complete and detailed list is available on the Euclid web site (www.euclid-ec.org/consortium/community/). This work has made use of the Euclid Quick Release Q1 data from the Euclid mission of the European Space Agency (ESA), 2025, https://doi.org/10.57780/esa-2853f3b. This research makes use of ESA Datalabs (datalabs.esa.int), an initiative by ESA’s Data Science and Archives Division in the Science and Operations Department, Directorate of Science. AS acknowledges support from the National Key R&D Program of China (No. 2023YFA1607800, 2023YFA1607802), and from the Yangyang Development Fund.

Authors and affiliations

Euclid Collaboration: A. Sonnenfeld1, 2, 3, A. Nersesian4, 5, K. Rojas6, P. Holloway7, J. W. Nightingale8, C. Tortora9, S. Schuldt10, 11, 12, T. Li7, C. J. Storfer13, R. Gavazzi14, 15, D. Sluse4, G. Despali16, 17, 18, T. E. Collett7, S. Serjeant19, F. Courbin20, 21, 22, R. B. Metcalf16, 17, S. Vegetti23, B. Altieri24, S. Andreon25, N. Auricchio17, C. Baccigalupi26, 27, 28, 29, M. Baldi30, 17, 18, M. Balestra31, S. Bardelli17, P. Battaglia17, A. Biviano27, 26, M. Bolzonella17, E. Branchini32, 33, 25, M. Brescia34, 9, J. Brinchmann35, 36, 37, S. Camera38, 39, 40, G. Cañas-Herrera41, V. Capobianco40, C. Carbone12, M. Castellano42, G. Castignani17, S. Cavuoti9, 43, K. C. Chambers13, A. Cimatti44, C. Colodro-Conde45, G. Congedo46, C. J. Conselice47, L. Conversi48, 24, Y. Copin49, A. Costille14, H. M. Courtois50, M. Cropper51, H. Degaudenzi52, G. De Lucia27, C. Dolding51, H. Dole53, F. Dubath52, C. A. J. Duncan46, X. Dupac24, S. Dusini54, S. Escoffier55, M. Farina56, R. Farinelli17, F. Faustini42, S. Ferriol49, F. Finelli17, 57, P. Fosalba58, 59, M. Frailis27, E. Franceschi17, M. Fumana12, L. Gabarra60, S. Galeotta27, K. George61, B. Gillis46, C. Giocoli17, 18, J. Gracia-Carpio62, A. Grazian31, F. Grupp62, 63, W. G. Hartley52, S. V. H. Haugan64, G. Helou65, 66, S. Hemmati66, H. Hoekstra41, W. Holmes67, I. M. Hook68, F. Hormuth69, A. Hornstrup70, 71, K. Jahnke72, M. Jhabvala73, S. Kermiche55, A. Kiessling67, B. Kubik49, K. Kuijken41, M. Kümmel63, M. Kunz74, H. Kurki-Suonio11, 75, A. M. C. Le Brun76, D. Le Mignant14, S. Ligori40, P. B. Lilje64, V. Lindholm11, 75, I. Lloro77, M. Magliocchetti56, G. Mainetti78, O. Mansutti27, S. Marcin6, O. Marggraf79, M. Martinelli42, 80, N. Martinet14, F. Marulli16, 17, 18, R. J. Massey81, N. Mauri44, 18, E. Medinaceli17, S. Mei82, 83, M. Meneghetti17, 18, E. Merlin31, G. Meylan84, P. Monaco85, 27, 28, 26, A. Mora86, M. Moresco16, 17, L. Moscardini16, 17, 18, R. Nakajima79, C. Neissner87, 88, R. C. Nichol89, S.-M. Niemi90, C. Padilla87, S. Paltani52, F. Pasian27, W. J. Percival91, 92, 93, V. Pettorino90, S. Pires94, G. Polenta95, M. Poncet96, L. A. Popa97, L. Pozzetti17, F. Raison62, A. Renzi98, 54, 17, J. Rhodes67, G. Riccio9, I. Risso32, 33, 25, F. Rizzo27, E. Romelli27, B. Rusholme66, R. Saglia63, 62, Z. Sakr99, 100, 101, D. Sapone102, M. Schirmer72, P. Schneider79, T. Schrabback103, A. Secroun55, E. Sihvola104, P. Simon79, C. Sirignano98, 54, G. Sirri18, L. Stanco54, P. Tallada-Crespí105, 88, A. N. Taylor46, I. Tereno106, 107, N. Tessore51, S. Toft108, 109, R. Toledo-Moreo110, 111, F. Torradeflot108, 105, A. Tsyganov112, I. Tutusaus59, 58, 100, J. Valiviita11, 75, T. Vassallo27, 61, J. Wang66, C. Weller63, 62, S. Borgani27, 26, G. Zamorani17, F. M. Zerbi25, E. Zucca17, I. T. Andika63, M. Ballardini113, 114, 17, C. Benoit-Lévy115, S. Borgani85, 26, 27, 28, 116, E. Bozzo52, C. Burigana117, 57, R. Cabanac100, A. Cappi115, 17, F. Caro42, T. Castro118, 27, 26, J. A. Escartin Vigo62, J. García-Bellido99, T. Gasparetto42, C. M. Gutierrez45, 119, A. Hall46, Y. Kang103, J. Macias-Perez120, R. Maoli121, 42, M. Pöntinen11, E. Sarpa27, V. Scottez122, 123, M. Sereno17, 18, K. Tanidis124, M. Tenti18, M. Tucci52, M. Wiesmann64, J. A. Acevedo Barroso67, Y. Akrami99, 125, G. Angora9, 113, N. Archidiacono126, 127, G. Aricò128, F. Atrio-Barandela128, M. Baes5, L. Bazzanini113, J. R. Bermejo-Climent129, 130, D. Bertacca95, M. Bethermin131, F. Beutler46, A. Blanchard100, H. Böhringer62, 61, 132, F. Bouché133, 43, M. L. Brown47, S. Bruton65, A. Calabrese42, C. S. Carbone134, F. Cogato17, P. Corcho-Caballero135, M. Correnti42, 95, B. Csizi103, O. Cucciati17, T. de Boer13, F. De Paolis136, 137, 138, G. Desprez135, A. Díaz-Sánchez139, S. Di Domizio32, 33, J. M. Diego140, P.-A. Duc131, V. Duret55, M. Y. Elkhashab27, 28, 85, 26, Y. Fang63, A. Farina25, 33, A. Finoguenov11, F. Fontanot27, 26, A. Franco136, 137, 138, A. Fumagalli27, Y. Fu41, 135, K. Ganga82, E. Gaztanaga59, 58, 7, Z. Ghaffari27, 26, F. Giacomini18, F. Gianotti17, G. Gozaliasilam141, 11, 100, A. Gruppuso17, 18, M. Guidi30, 17, H. Hildebrandt142, G. Hjorth146, M. Jauzac144, 81, 145, 100, J. J. E. Kajava147, 10, 148, K. Kansal149, 150, D. Karagiannis151, 113, J. Kim60, C. C. Kirkpatrick104, A. Kovács130, 129, I. Kovačić5, K. Koyama7, S. Kruk24, M. C. Lam46, M. Lattanzi114, S. Lee153, 67, M. D. Lepinzan27, G. Leroy144, 81, G. F. Lesci16, 117, J. Lesgourgues154, L. Linke103, G. Maggio27, A. Manjón-García139, F. Mannucci155, F. R. Marleau103, C. J. A. P. Martins156, 35, M. Miluzio24, 157, G. Morgante17, K. Naidoo7, 72, A. Navarro-Alsina79, S. Nessseris99, L. Nicastro17, F. Oppizzi33, 54, 98, F. Pace38, 39, 40, 107, D. Paoletti17, 57, K. Paterson72, L. Patrizii18, C. Pattison7, A. Pisani55, L. Pizzuti158, D. Potter159, G. W. Pratt94, S. Quail60, 17, M. Radovich31, F. Ricci160, 42, W. Roster62, M. Sahlén161, D. B. Sanders1, C. Scarlata162, J. Schaye41, A. Schneider159, D. Sciotti164, E. Sellentin168, L. C. Smith164, J. G. Sorce165, 53, I. Spanduto13, M. Talia16, 17, F. Tarsitano166, 167, G. Testera33, R. Teyssier168, S. Tosi32, 25, 33, A. Troja27, C. Uhlemann169, 8, C. Valieri18, A. Venhola170, D. Vergani17, G. Vernardos171, 172, E. Vilenius51, S. Vinciguerra14, M. von Wietersheim-Kramsta81, 144, N. A. Walton164, A. H. Wright142, H. W. Yeung46, and D. Scott173 ⁷ Institute of Cosmology and Gravitation, University of Portsmouth, Portsmouth PO1 3FX, UK

⁸ School of Mathematics, Statistics and Physics, Newcastle University, Herschel Building, Newcastle-upon-Tyne, NE1 7RU, UK

⁹ INAF-Osservatorio Astronomico di Capodimonte, Via Moiariello 16, 80131 Napoli, Italy

¹⁰ Finnish Centre for Astronomy with ESO (FINCA), Quantum, Vesilinnantie 5, University of Turku, 20014 Turku, Finland

¹¹ Department of Physics, P.O. Box 64, University of Helsinki, 00014 Helsinki, Finland

¹² INAF-IASF Milano, Via Alfonso Corti 12, 20133 Milano, Italy

¹³ Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, HI 96822, USA

¹⁴ Aix-Marseille Université, CNRS, CNES, LAM, Marseille, France

¹⁵ Institut d’Astrophysique de Paris, UMR 7095, CNRS, and Sorbonne Université, 98 bis boulevard Arago, 75014 Paris, France

¹⁶ Dipartimento di Fisica e Astronomia "Augusto Righi" - Alma Mater Studiorum Università di Bologna, via Piero Gobetti 93/2, 40129 Bologna, Italy

¹⁷ INAF-Osservatorio di Astrofisica e Scienza dello Spazio di Bologna, Via Piero Gobetti 93/3, 40129 Bologna, Italy

¹⁸ INFN-Sezione di Bologna, Viale Berti Pichat 6/2, 40127 Bologna, Italy

¹⁹ School of Physical Sciences, The Open University, Milton Keynes, MK7 6AA, UK

²⁰ Institut de Ciències del Cosmos (ICCUB), Universitat de Barcelona (IEEC-UB), Martí i Franquès 1, 08028 Barcelona, Spain

²¹ Institució Catalana de Recerca i Estudis Avançats (ICREA), Passeig de Lluís Companys 23, 08010 Barcelona, Spain

²² Institut de Ciencies de l’Espai (IEEC-CSIC), Campus UAB, Carrer de Can Magrans, s/n Cerdanyola del Vallés, 08193 Barcelona, Spain

²³ Max-Planck-Institut für Astrophysik, Karl-Schwarzschild-Str. 1, 85748 Garching, Germany

²⁴ ESAC/ESA, Camino Bajo del Castillo, s/n., Urb. Villafranca del Castillo, 28692 Villanueva de la Cañada, Madrid, Spain

²⁵ INAF-Osservatorio Astronomico di Brera, Via Brera 28, 20122 Milano, Italy

²⁶ IFPU, Institute for Fundamental Physics of the Universe, via Beirut 2, 34151 Trieste, Italy

²⁷ INAF-Osservatorio Astronomico di Trieste, Via G. B. Tiepolo 11, 34143 Trieste, Italy

²⁸ INFN, Sezione di Trieste, Via Valerio 2, 34127 Trieste TS, Italy

²⁹ SISSA, International School for Advanced Studies, Via Bonomea 265, 34136 Trieste TS, Italy

³⁰ Dipartimento di Fisica e Astronomia, Università di Bologna, Via Gobetti 93/2, 40129 Bologna, Italy

³¹ INAF-Osservatorio Astronomico di Padova, Via dell’Osservatorio 5, 35122 Padova, Italy

³² Dipartimento di Fisica, Università di Genova, Via Dodecaneso 33, 16146, Genova, Italy

³³ INFN-Sezione di Genova, Via Dodecaneso 33, 16146, Genova, Italy

³⁴ Department of Physics "E. Pancini", University Federico II, Via Cinthia 6, 80126, Napoli, Italy

³⁵ Instituto de Astrofísica e Ciências do Espaço, Universidade do Porto, CAUP, Rua das Estrelas, PT4150-762 Porto, Portugal

³⁶ Faculdade de Ciências da Universidade do Porto, Rua do Campo de Alegre, 4150-007 Porto, Portugal

³⁷ European Southern Observatory, Karl-Schwarzschild-Str. 2, 85748 Garching, Germany

³⁸ Dipartimento di Fisica, Università degli Studi di Torino, Via P. Giuria 1, 10125 Torino, Italy

³⁹ INFN-Sezione di Torino, Via P. Giuria 1, 10125 Torino, Italy

⁴⁰ INAF-Osservatorio Astrofisico di Torino, Via Osservatorio 20, 10025 Pino Torinese (TO), Italy

⁴¹ Leiden Observatory, Leiden University, Einsteinweg 55, 2333 CC Leiden, The Netherlands

⁴² INAF-Osservatorio Astronomico di Roma, Via Frascati 33, 00078 Monteporzio Catone, Italy

⁴³ INFN – Sezione di Napoli, Via Cinthia 6, 80126, Napoli, Italy

⁴⁴ Dipartimento di Fisica e Astronomia "Augusto Righi" - Alma Mater Studiorum Università di Bologna, Viale Berti Pichat 6/2, 40127 Bologna, Italy

⁴⁵ Instituto de Astrofísica de Canarias, E-38205 La Laguna, Tenerife, Spain

⁴⁶ Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK

⁴⁷ Jodrell Bank Centre for Astrophysics, Department of Physics and Astronomy, University of Manchester, Oxford Road, Manchester M13 9PL, UK

⁴⁸ European Space Agency/ESRIN, Largo Galileo Galilei 1, 00044 Frascati, Roma, Italy

⁴⁹ Université Claude Bernard Lyon 1, CNRS/IN2P3, IP2I Lyon, UMR 5822, Villeurbanne, F-69100, France

⁵⁰ UCB Lyon 1, CNRS/IN2P3, IUF, IP2I Lyon, 4 rue Enrico Fermi, 69622 Villeurbanne, France

⁵¹ Mullard Space Science Laboratory, University College London, Holmbury St Mary, Dorking, Surrey RH5 6NT, UK

⁵² Department of Astronomy, University of Geneva, ch. d’Ecogia 16, 1290 Versoix, Switzerland

⁵³ Université Paris-Saclay, CNRS, Institut d’astrophysique spatiale, 91405, Orsay, France

⁵⁴ INFN-Padova, Via Marzolo 8, 35131 Padova, Italy

⁵⁵ Aix-Marseille Université, CNRS/IN2P3, CPPM, Marseille, France

⁵⁶ INAF-Istituto di Astrofisica e Planetologia Spaziali, via del Fosso del Cavaliere, 100, 00100 Roma, Italy

⁵⁷ INFN-Bologna, Via Irnerio 46, 40126 Bologna, Italy

⁵⁸ Institut d’Estudis Espacials de Catalunya (IEEC), Edifici RDIT, Campus UPC, 08860 Castelldefels, Barcelona, Spain

⁵⁹ Institute of Space Sciences (ICE, CSIC), Campus UAB, Carrer de Can Magrans, s/n, 08193 Barcelona, Spain

⁶⁰ Department of Physics, University of Oxford, Keble Road, Oxford OX1 3RH, UK

⁶¹ University Observatory, LMU Faculty of Physics, Scheinerstr. 1, 81679 Munich, Germany

⁶² Max Planck Institute for Extraterrestrial Physics, Giessenbachstr. 1, 85748 Garching, Germany

⁶³ Universitäts-Sternwarte München, Fakultät für Physik, Ludwig-Maximilians-Universität München, Scheinerstr. 1, 81679 München, Germany

⁶⁴ Institute of Theoretical Astrophysics, University of Oslo, P.O. Box 1029 Blindern, 0315 Oslo, Norway

⁶⁵ California Institute of Technology, 1200 E California Blvd, Pasadena, CA 91125, USA

⁶⁶ Caltech/IPAC, 1200 E. California Blvd., Pasadena, CA 91125, USA

⁶⁷ Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA, 91109, USA

⁶⁸ Department of Physics, Lancaster University, Lancaster, LA1 4YB, UK

⁶⁹ Felix Hormuth Engineering, Goethestr. 17, 69181 Leimen, Germany

⁷⁰ Technical University of Denmark, Elektrovej 327, 2800 Kgs. Lyngby, Denmark

⁷¹ Cosmic Dawn Center (DAWN), Denmark

⁷² Max-Planck-Institut für Astronomie, Königstuhl 17, 69117 Heidelberg, Germany

⁷³ NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA 74 Université de Genève, Département de Physique Théorique and Centre for Astroparticle Physics, 24 quai Ernest-Ansermet, CH-1211 Genève 4, Switzerland

75 Helsinki Institute of Physics, Gustaf Hällströminkatu 2, University of Helsinki, 00014 Helsinki, Finland

76 Laboratoire d’étude de l’Univers et des phenomenes eXtremes, Observatoire de Paris, Université PSL, Sorbonne Université, CNRS, 92190 Meudon, France

77 SKAO, Jodrell Bank, Lower Withington, Macclesfield SK11 9FT, UK

78 Centre de Calcul de l’IN2P3/CNRS, 21 avenue Pierre de Coubertin 69627 Villeurbanne Cedex, France

79 Universität Bonn, Argelander-Institut für Astronomie, Auf dem Hügel 71, 53121 Bonn, Germany

80 INFN-Sezione di Roma, Piazzale Aldo Moro, 2 - c/o Dipartimento di Fisica, Edificio G. Marconi, 00185 Roma, Italy

81 Department of Physics, Institute for Computational Cosmology, Durham University, South Road, Durham, DH1 3LE, UK

82 Université Paris Cité, CNRS, Astroparticule et Cosmologie, 75013 Paris, France

83 CNRS-UCB International Research Laboratory, Centre Pierre Binétruy, IRL2007, CPB-IN2P3, Berkeley, USA

84 Institute of Physics, Laboratory of Astrophysics, Ecole Polytechnique Fédérale de Lausanne (EPFL), Observatoire de Sauverny, 1290 Versoix, Switzerland

85 Dipartimento di Fisica - Sezione di Astronomia, Università di Trieste, Via Tiepolo 11, 34131 Trieste, Italy

86 Telespazio UK S.L. for European Space Agency (ESA), Camino bajo del Castillo, s/n, Urbanizacion Villafranca del Castillo, Villanueva de la Cañada, 28692 Madrid, Spain

87 Institut de Física d’Altes Energies (IFAE), The Barcelona Institute of Science and Technology, Campus UAB, 08193 Bellaterra (Barcelona), Spain

88 Port d’Informació Científica, Campus UAB, C. Albareda s/n, 08193 Bellaterra (Barcelona), Spain

89 School of Mathematics and Physics, University of Surrey, Guildford, Surrey, GU2 7XH, UK

90 European Space Agency/ESTEC, Keplerlaan 1, 2201 AZ Noordwijk, The Netherlands

91 Waterloo Centre for Astrophysics, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada

92 Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada

93 Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2L 2Y5, Canada

94 Université Paris-Saclay, Université Paris Cité, CEA, CNRS, AIM, 91191, Gif-sur-Yvette, France

95 Space Science Data Center, Italian Space Agency, via del Politecnico snc, 00133 Roma, Italy

96 Centre National d’Etudes Spatiales – Centre spatial de Toulouse, 18 avenue Edouard Belin, 31401 Toulouse Cedex 9, France

97 Institute of Space Science, Str. Atomistilor, nr. 409 Măgurele, Ilfov, 077125, Romania

98 Dipartimento di Fisica e Astronomia "G. Galilei", Università di Padova, Via Marzolo 8, 35131 Padova, Italy

99 Instituto de Física Teórica UAM-CSIC, Campus de Cantoblanco, 28049 Madrid, Spain

100 Institut de Recherche en Astrophysique et Planétologie (IRAP), Université de Toulouse, CNRS, UPS, CNES, 14 Av. Edouard Belin, 31400 Toulouse, France

101 Université St Joseph; Faculty of Sciences, Beirut, Lebanon

102 Departamento de Física, FCFM, Universidad de Chile, Blanco Encalada 2008, Santiago, Chile

103 Universität Innsbruck, Institut für Astro- und Teilchenphysik, Technikerstr. 25/8, 6020 Innsbruck, Austria

104 Department of Physics and Helsinki Institute of Physics, Gustaf Hällströminkatu 2, University of Helsinki, 00014 Helsinki, Finland

105 Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas (CIEMAT), Avenida Complutense 40, 28040 Madrid, Spain

106 Departamento de Física, Faculdade de Ciências, Universidade de Lisboa, Edifício C8, Campo Grande, PT1749-016 Lisboa, Portugal

107 Instituto de Astrofísica e Ciências do Espaço, Faculdade de Ciências, Universidade de Lisboa, Campo Grande, 1749-016 Lisboa, Portugal

108 Cosmic Dawn Center (DAWN)

109 Niels Bohr Institute, University of Copenhagen, Jagtvej 128, 2200 Copenhagen, Denmark

110 Universidad Politécnica de Cartagena, Departamento de Electrónica y Tecnología de Computadoras, Plaza del Hospital 1, 30202 Cartagena, Spain

111 European University of Technology EUt+, European Union

112 Centre for Information Technology, University of Groningen, P.O. Box 11044, 9700 CA Groningen, The Netherlands

113 Dipartimento di Fisica e Scienze della Terra, Università degli Studi di Ferrara, Via Giuseppe Saragat 1, 44122 Ferrara, Italy

114 Instituto Nazionale di Fisica Nucleare, Sezione di Ferrara, Via Giuseppe Saragat 1, 44122 Ferrara, Italy

115 Université Côte d’Azur, Observatoire de la Côte d’Azur, CNRS, Laboratoire Lagrange, Bd de l’Observatoire, CS 34229, 06304 Nice cedex 4, France

116 ICIS - Centro Nazionale di Ricerca in High Performance Computing, Big Data e Quantum Computing, Via Magnanelli 2, Bologna, Italy

117 INAF, Istituto di Radioastronomia, Via Piero Gobetti 101, 40129 Bologna, Italy

118 Department of Mathematical Physics, Institute of Physics, University of São Paulo, R. do Matão 1371, 05508-090, São Paulo, SP, Brazil

119 Universidad de La Laguna, Dpto. Astrofísica, E-38206 La Laguna, Tenerife, Spain

120 Univ. Grenoble Alpes, CNRS, Grenoble INP, LPSC-IN2P3, 53, Avenue des Martyrs, 38000 Grenoble, France

121 Dipartimento di Fisica, Sapienza Università di Roma, Piazzale Aldo Moro 2, 00185 Roma, Italy

122 Institut d’Astrophysique de Paris, 98bis Boulevard Arago, 75014, Paris, France

123 ICL, Junia, Université Catholique de Lille, LITL, 59000 Lille, France

124 Center for Astrophysics and Cosmology, University of Nova Gorica, Nova Gorica, Slovenia

125 CERCA/ISO, Department of Physics, Case Western Reserve University, 10900 Euclid Avenue, Cleveland, OH 44106, USA

126 Dipartimento di Fisica "Aldo Pontremoli", Università degli Studi di Milano, Via Celoria 16, 20133 Milano, Italy

127 INFN-Sezione di Milano, Via Celoria 16, 20133 Milano, Italy

128 Departamento de Física Fundamental. Universidad de Salamanca. Plaza de la Merced s/n. 37008 Salamanca, Spain

129 Konkoly Observatory, HUN-REN CSFK, MTA Centre of Excellence, Budapest, Konkoly Thege Miklós út 15-17. H-1121, Hungary

130 MTA-CSFK Lendület Large-Scale Structure Research Group, Konkoly-Thege Miklós út 15-17, H-1121 Budapest, Hungary

131 Université de Strasbourg, CNRS, Observatoire astronomique de Strasbourg, UMR 7550, 67000 Strasbourg, France

132 Max-Planck-Institut für Physik, Boltzmannstr. 8, 85748 Garching, Germany

133 Scuola Superiore Meridionale, Via Mezzocannone 4, 80138, Napoli, Italy

134 Instituto de Astrofísica e Ciências do Espaço, Faculdade de Ciências, Universidade de Lisboa, Tapada da Ajuda, 1349-018 Lisboa, Portugal

135 Kapteyn Astronomical Institute, University of Groningen, PO Box 800, 9700 AV Groningen, The Netherlands 136 Department of Mathematics and Physics E. De Giorgi, University of Salento, Via per Arnesano, CP-193, 73100, Lecce, Italy

137 INFN, Sezione di Lecce, Via per Arnesano, CP-193, 73100, Lecce, Italy

138 INAF-Sezione di Lecce, c/o Dipartimento Matematica e Fisica, Via per Arnesano, 73100, Lecce, Italy

139 Departamento Física Aplicada, Universidad Politécnica de Cartagena, Campus Muralla del Mar, 30202 Cartagena, Murcia, Spain

140 Instituto de Física de Cantabria, Edificio Juan Jordá, Avenida de los Castros, 39005 Santander, Spain

141 Department of Computer Science, Aalto University, PO Box 15400, Espoo, FI-00 076, Finland

142 Ruhr University Bochum, Faculty of Physics and Astronomy, Astronomical Institute (AIRUB), German Centre for Cosmological Lensing (GCCL), 44780 Bochum, Germany

143 DARK, Niels Bohr Institute, University of Copenhagen, Jagtvej 155, 2200 Copenhagen, Denmark

144 Department of Physics, Centre for Extragalactic Astronomy, Durham University, South Road, Durham, DH1 3LE, UK

145 Astrophysics Research Centre, University of KwaZulu-Natal, Westville Campus, Durban 4041, South Africa

146 School of Mathematics, Statistics & Computer Science, University of KwaZulu-Natal, Westville Campus, Durban 4041, South Africa

147 Department of Physics and Astronomy, Vesilinnantie 5, University of Turku, 20014 Turku, Finland

148 Serco for European Space Agency (ESA), Camino bajo del Castillo, s/n, Urbanizacion Villafranca del Castillo, Villanueva de la Cañada, 28692 Madrid, Spain

149 ARC Centre of Excellence for Dark Matter Particle Physics, Melbourne, Australia

150 Centre for Astrophysics & Supercomputing, Swinburne University of Technology, Hawthorn, Victoria 3122, Australia

151 Van Swinderen Institute, University of Groningen, Nijenborgh 3, 9747 AG Groningen, The Netherlands

152 Department of Physics and Astronomy, University of the Western Cape, Bellville, Cape Town, 7535, South Africa

153 Ohio University, Physics & Astronomy Department,1 Ohio University, Athens, OH 45701, USA

154 Institute for Theoretical Particle Physics and Cosmology (TTK), RWTH Aachen University, 52056 Aachen, Germany

155 INAF-Osservatorio Astrofisico di Arcetri, Largo E. Fermi 5, 50125, Firenze, Italy

156 Centro de Astrofísica da Universidade do Porto, Rua das Estrelas, 4150-762 Porto, Portugal

157 HE Space for European Space Agency (ESA), Camino bajo del Castillo, s/n, Urbanizacion Villafranca del Castillo, Villanueva de la Cañada, 28692 Madrid, Spain

158 Dipartimento di Fisica “G. Occhialini”, Università degli Studi di Milano Bicocca, Piazza della Scienza 3, 20126 Milano, Italy

159 Department of Astrophysics, University of Zurich, Winterthurerstrasse 190, 8057 Zurich, Switzerland

160 Department of Mathematics and Physics, Roma Tre University, Via della Vasca Navale 84, 00146 Rome, Italy

161 Theoretical astrophysics, Department of Physics and Astronomy, Uppsala University, Box 524, 751 20 Uppsala, Sweden

162 Minnesota Institute for Astrophysics, University of Minnesota, 116 Church St SE, Minneapolis, MN 55455, USA

163 Mathematical Institute, University of Leiden, Einsteinweg 55, 2333 CA Leiden, The Netherlands

164 Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK

165 Univ. Lille, CNRS, Centrale Lille, UMR 9189 CRIStAL, 59000 Lille, France

166 Kobayashi-Maskawa Institute for the Origin of Particles and the Universe, Nagoya University, Chikusa-ku, Nagoya, 464-8602, Japan

167 Institute for Particle Physics and Astrophysics, Dept. of Physics, ETH Zurich, Wolfgang-Pauli-Strasse 27, 8093 Zurich, Switzerland

168 Department of Astrophysical Sciences, Peyton Hall, Princeton University, Princeton, NJ 08544, USA

169 Fakultät für Physik, Universität Bielefeld, Postfach 100131, 33501 Bielefeld, Germany

170 Space physics and astronomy research unit, University of Oulu, Pentti Kaiteran katu 1, FI-90014 Oulu, Finland

171 Department of Physics and Astronomy, Lehman College of the CUNY, Bronx, NY 10468, USA

172 American Museum of Natural History, Department of Astrophysics, New York, NY 10024, USA

173 Department of Physics and Astronomy, University of British Columbia, Vancouver, BC V6T 1Z1, Canada

Ambiguous lens candidates

As we explained in Sect. 3.1, we adopted a series of objective criteria to define our sample of lenses. Nevertheless, we had to use judgement to further bring the purity of the sample to 100%. In the process, we eliminated 17 ambiguous lens candidates that we deemed not to be lenses. Figure A.1 shows their images, alongside our best-fit model. In Table A.1 we list the motivations that led us to remove them from the golden lens sample.

Completeness

Our analysis rests on the assumption that our lens sample is a representative subset of all of the golden lenses in Q1. That is, we assume that our sample is either complete or is missing lenses at random, with no correlation between the physical properties of the lenses and the probability of them being included in the sample. Assessing completeness is difficult because we lack a truly complete reference sample of lenses, in Q1 or anywhere. Our main approach is to examine how the lens finding procedure that fed our sample performed on 1) images of simulated lenses, and 2) lenses that were found directly from visual inspection. The first step in this process is to make a census of the provenance of the golden lenses. Of the 95 lenses in our sample, 80 were discovered as part of the search described by W26. Two of them were previously known: EUCL J041339.51−484111.4 was originally discovered by Storfer et al. [65], and EUCL J040457.32−492221.5 was in the sample of Inchausti et al. [38]. Of the remaining 16 golden lenses, eleven were discovered by Euclid Collaboration: Ecker et al. [24]. These lenses were excluded by W26 at the sample-building phase, as they removed any object with a match in the Gaia catalogue, as an attempt to remove stars from the sample. Two lenses were discovered by Euclid Collaboration: Rojas et al. [30], one by Euclid Collaboration: Xu et al. [33], and one, EUCL J040205.11−480808.4, was a previously known lens, discovered by Stein et al. [64]. Finally, our sample includes the lens NGC 6505, which was originally discovered by O’Riordan et al. [49], and which would have been found by the W26 search as well. In summary, the vast majority of our lenses belong to the W26 sample. Therefore, we focus primarily on that lens finding effort to assess the completeness of our sample.

W26 estimated the completeness of their lens search to be in the range 60–70%. However, that figure refers to all candidates graded B or above. As the golden lenses are a much smaller subset, we expect the completeness of the golden lens sample to be different. The lens search of W26 consisted of a three-step process. In a first step, potential lens candidates were ranked by making use of convolutional neural networks (CNNs). In the second step, the highest-scoring 78 214 lens candidates were visually inspected by citizen scientist volunteers, along with 40 000 candidates drawn randomly from the initial pool of ∼106\sim10^{6} objects (these two samples overlap). In the third and final step, expert scientists visually inspected 7362 lens candidates and assigned them the grades on which our initial selection is based.

We can evaluate the completeness of the human selection steps by analysing their performance on images of simulated lenses. These simulations were shown to experts and volunteers during the classification process, interspersed among the real images of lens candidates. Experts collectively classified 265 simulated lenses, 146 of which are golden lenses. Of these, 114 were given a grade B or above, and 32 (approximately 22%) were missed. The images of a subset of those lenses are shown in Fig. B.1. By comparison, volunteers reached a much higher completeness. Of 1861 simulated golden lenses classified by them, only 35 were rejected (less than 2%). This is not surprising, because the goal of experts is to strike a balance between completeness and purity, and the purity of their selection is much higher than that of the volunteers. Taking these numbers at face value, one might conclude that the completeness of the two human inspection steps combined is roughly 77%, with the limiting factor being the expert classification. However, caution is required. This figure is based on the implicit assumption that the simulated golden lenses are representative of the real sample of golden lenses in the sky. We have no means of verifying this assumption at this stage. In order to make progress in our understanding of the human selection function, classification data on a much larger sample of simulated golden lenses are needed. We aim to build such a data set in future lens searches.

To assess the completeness of the CNN-based selection step, we examined the random sample of images that was independently inspected by the volunteers. In the 40 000 randomly selected objects the six golden lenses. Five of these were also flagged by the CNNs, while one was missed. Therefore, our guess for the completeness of the CNN selection step is ∼83%\sim83\%. However, this estimate is highly uncertain due to the small numbers involved.

Multiplying the completeness of the CNN- and visual-selection steps that we estimated above, we obtain a value of 64% for the completeness of the W26 lens finding procedure. On the one hand, this is a lower limit of the overall completeness of our sample, because some of the lenses missed by W26 were later found by Euclid Collaboration: Rojas et al. [30], Euclid Collaboration: Bergamini et al. [22], Euclid Collaboration: Ecker et al. [24], and Euclid Collaboration: Xu et al. [33]. On the other hand, the limited amount of information available prevents us from drawing a firm conclusion on completeness at this stage. With future lens searches, we aim to both increase our completeness over the Q1 area and to better characterise the selection function of our search.

Colour-composite images of the 17 ambiguous lens candidates

Figure A.1. Colour-composite images of the 17 ambiguous lens candidates that we excluded from the golden lens sample. For each candidate, we show the corresponding lens light-subtracted image, and best-fit lens model next to it. Each image is 10′′10^{\prime\prime} on the side.

Object nameReason for discarding
EUCL J033145.30−294246.3The model predicts an extended and curved counter-image, but the observed one is compact.
EUCL J175935.63+663355.6The counter-image is likely an object in the lens plane.
EUCL J034546.25−490936.7The ring shows irregular structure that is more typical of ring galaxies.
EUCL J042246.82−481759.7The shape of the arc cannot be reproduced by a lens model.
EUCL J035343.62−474653.9The model predicts an additional image that is not observed.
EUCL J040615.79−475232.0The model cannot reproduce the position of the counter-image.
EUCL J040748.85−473802.1The ring shows irregular structure that is more typical of ring galaxies.
EUCL J040827.48−470732.1The model requires a two-pixel offset between mass and light to reproduce the images.
EUCL J033948.62−290942.9The surface brightness of the ring is too extended, compared to that predicted by the model.
EUCL J032346.78−272726.0The ring shows irregular structure that is more typical of ring galaxies.
EUCL J033016.18−261615.7The lens shows disturbed morphology, making a non-lens scenario more likely.
EUCL J180443.51+634625.4The arcs are too straight, compared to the model prediction.
EUCL J174500.66+643019.6The relative brightness of the two images is very different from that predicted by the model.
EUCL J174212.69+653500.6The ring shows irregular structure that is more typical of ring galaxies.
EUCL J181015.36+683825.4The shape of the counter-image does not match that predicted by the model.
EUCL J181126.31+682107.0The orientation of the arc is different from that predicted by the model.
EUCL J035809.69−482755.9The ring shows irregular structure that is more typical of ring galaxies.

Table A.1. Ambiguous lens candidates excluded from the golden lens sample.

Selection of simulated golden lenses classified as grade C or below

Figure B.1. Selection of simulated golden lenses that were classified as grade C or below by lensing experts.

References

  1. [1]Adams, N. J., Bowler, R. A. A., Jarvis, M. J., Häußler, B., & Lagos, C. D. P. 2021, MNRAS, 506, 4933
  2. [2]Arnouts, S., Cristiani, S., Moscardini, L., et al. 1999, MNRAS, 310, 540
  3. [3]Auger, M. W., Treu, T., Bolton, A. S., et al. 2009, ApJ, 705, 1099
  4. [4]Auger, M. W., Treu, T., Bolton, A. S., et al. 2010, ApJ, 724, 511
  5. [5]Bernardi, M., Domínguez Sánchez, H., Margalef-Bentabol, B., Nikakhtar, F., & Sheth, R. K. 2020, MNRAS, 494, 5148
  6. [6]Bertin, E. 2011, in ASP Conference Series, Vol. 442, Astronomical Data Analysis Software and Systems XX, ed. I. N. Evans, A. Accomazzi, D. J. Mink, & A. H. Rots, 435
  7. [7]Blumenthal, G. R., Faber, S. M., Flores, R., & Primack, J. R. 1986, ApJ, 301, 27
  8. [8]Bolton, A. S., Brownstein, J. R., Kochanek, C. S., et al. 2012, ApJ, 757, 82
  9. [9]Bolton, A. S., Burles, S., Koopmans, L. V. E., Treu, T., & Moustakas, L. A. 2006, ApJ, 638, 703
  10. [10]Brownstein, J. R., Bolton, A. S., Schlegel, D. J., et al. 2012, ApJ, 744, 41
  11. [11]Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000
  12. [12]Bundy, K., Leauthaud, A., Saito, S., et al. 2017, ApJ, 851, 34
  13. [13]Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1994, ApJ, 429, 582
  14. [14]Cappellari, M., Romanowsky, A. J., Brodie, J. P., et al. 2015, ApJ, 800, L21
  15. [15]Coil, A. L., Blanton, M. R., Burles, S. M., et al. 2011, ApJ, 741, 8
  16. [16]DESI Collaboration, Abdul Karim, M., Adame, A. G., et al. 2026, AJ, 171, 285
  17. [17]Despali, G., Giocoli, C., Moscardini, L., et al. 2026, A&A, 708, A47
  18. [18]Duffy, A. R., Schaye, J., Kay, S. T., et al. 2010, MNRAS, 405, 2161
  19. [19]Dutton, A. A. & Macciò, A. V. 2014, MNRAS, 441, 3359
  20. [20]Etherington, A., Nightingale, J. W., Massey, R., et al. 2022, MNRAS, 517, 3275
  21. [21]Euclid Collaboration: Aussel, H., Tereno, I., Schirmer, M., et al. 2026, A&A, 711, A1
  22. [22]Euclid Collaboration: Bergamini, P., Meneghetti, M., Acebron, A., et al. 2026, A&A, 711, A33
  23. [23]Euclid Collaboration: Cropper, M., Al-Balawhan, A., Amiaux, J., et al. 2025, A&A, 697, A2
  24. [24]Euclid Collaboration: Ecker, L. R., Fabricius, M., Seitz, S., et al. 2026, A&A, accepted, arXiv:2603.28580
  25. [25]Euclid Collaboration: Holloway, P., Verma, A., Walmsley, M., et al. 2026, A&A, 711, A30
  26. [26]Euclid Collaboration: Jahnke, K., Gillard, W., Schirmer, M., et al. 2025, A&A, 697, A3
  27. [27]Euclid Collaboration: Li, T., Collett, T. E., Walmsley, M., et al. 2026, A&A, 711, A29
  28. [28]Euclid Collaboration: Lines, N. E. P., Collett, T. E., Walmsley, M., et al. 2026, A&A, 711, A28
  29. [29]Euclid Collaboration: Mellier, Y., Abdurro'uf, Acevedo Barroso, J., et al. 2025, A&A, 697, A1
  30. [30]Euclid Collaboration: Rojas, K., Collett, T. E., Acevedo Barroso, J. A., et al. 2026, A&A, 711, A27
  31. [31]Euclid Collaboration: Romelli, E., Kümmel, M., Dole, H., et al. 2026, A&A, 711, A4
  32. [32]Euclid Collaboration: Walmsley, M., Holloway, P., Lines, N. E. P., et al. 2026, A&A, 711, A26 (W26)
  33. [33]Euclid Collaboration: Xu, X., Chen, R., Li, T., et al. 2026, A&A, 712, A207
  34. [34]Euclid Quick Release Q1. 2025, https://doi.org/10.57780/esa-2853f3b
  35. [35]Fitzpatrick, E. L. 1999, PASP, 111, 63
  36. [36]Galan, A., Van de Vyvere, L., Gomer, M., Vernardos, G., & Sluse, D. 2023, The Journal of Open Source Software, 8, 5567
  37. [37]Ilbert, O., Arnouts, S., McCracken, H. J., et al. 2006, A&A, 457, 841
  38. [38]Inchausti, J. C., Storfer, C. J., Huang, X., et al. 2026, ApJ, 1008, 102
  39. [39]Kinney, A. L., Calzetti, D., Bohlin, R. C., et al. 1996, ApJ, 467, 38
  40. [40]Koopmans, L. V. E., Treu, T., Bolton, A. S., Burles, S., & Moustakas, L. A. 2006, ApJ, 649, 599
  41. [41]Lidman, C., Tucker, B. E., Davis, T. M., et al. 2020, MNRAS, 496, 19
  42. [42]Liu, R., Sonnenfeld, A., Nipoti, C., & Li, R. 2025, A&A, 701, A181
  43. [43]Muzzin, A., Marchesini, D., Stefanon, M., et al. 2013, ApJ, 777, 18
  44. [44]Napolitano, N. R., Romanowsky, A. J., & Tortora, C. 2010, MNRAS, 405, 2351
  45. [45]Navarro, J. F., Frenk, C. S., & White, S. D. M. 1997, ApJ, 490, 493
  46. [46]Navarro, V., del Rio, S., Angel Diego, M., et al. 2024, in Space Data Management. Studies in Big Data, Vol. 141, 1–13
  47. [47]Oguri, M. 2010, PASJ, 62, 1017
  48. [48]Oldham, L., Auger, M. W., Fassnacht, C. D., et al. 2017, MNRAS, 465, 3185
  49. [49]O’Riordan, C. M., Oldham, L. J., Nersessian, A., et al. 2025, A&A, 694, A145
  50. [50]Parsa, S., Dunlop, J. S., McLure, R. J., & Mortlock, A. 2016, MNRAS, 456, 3194
  51. [51]Peirani, S., Dubois, Y., Volonteri, M., et al. 2017, MNRAS, 472, 2153
  52. [52]Ruff, A. J., Gavazzi, R., Marshall, P. J., et al. 2011, ApJ, 727, 96
  53. [53]Sahu, N., Tran, K.-V., Suyu, S. H., et al. 2024, ApJ, 970, 86
  54. [54]Schneider, P. & Sluys, D. 2013, A&A, 559, A37
  55. [55]Shajib, A. J., Treu, T., Birrer, S., & Sonnenfeld, A. 2021, MNRAS, 503, 2380
  56. [56]Shu, Y., Bolton, A. S., Brownstein, J. R., et al. 2015, ApJ, 803, 71
  57. [57]Shuntov, M., McCracken, H. J., Gavazzi, R., et al. 2022, A&A, 664, A61
  58. [58]Sonnenfeld, A. 2022a, A&A, 659, A132
  59. [59]Sonnenfeld, A. 2022b, A&A, 659, A133
  60. [60]Sonnenfeld, A. 2024, A&A, 690, A325
  61. [61]Sonnenfeld, A., Gavazzi, R., Suyu, S. H., Treu, T., & Marshall, P. J. 2013a, ApJ, 777, 97
  62. [62]Sonnenfeld, A., Li, S.-S., Despali, G., et al. 2023, A&A, 678, A4
  63. [63]Sonnenfeld, A., Treu, T., Gavazzi, R., et al. 2013b, ApJ, 777, 98
  64. [64]Stein, G., Blaum, J., Harrington, P., Medan, T., & Lukić, Z. 2022, ApJ, 932, 107
  65. [65]Storfer, C., Huang, X., Gu, A., et al. 2024, ApJS, 274, 16
  66. [66]Tan, C. Y., Shajib, A. J., Birrer, S., et al. 2024, MNRAS, 530, 1474
  67. [67]Tortora, C., Napolitano, N. R., Roy, N., et al. 2018, MNRAS, 473, 969
  68. [68]Tortora, C., Napolitano, N. R., Saglia, R. P., et al. 2014, MNRAS, 445, 162
  69. [70]Wright, A. H., Hildebrandt, H., van den Busch, J. L., et al. 2020, A&A, 640, L14
  70. [71]Zhou, R., Dey, B., Newman, J. A., et al. 2023, AJ, 165, 58

Paper details

vlnc:W_ucjhn78d
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

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