betaFeedback

WorksPhysics and Astronomy

The mass density slope of early-type galaxies: newest results from the redshift distribution of strong gravitational lenses

Abstract

The mass distribution of early-type galaxies (ETGs) encodes key information about baryonic processes and dark-matter assembly. In this paper, we propose a new framework to investigate the mass-density profiles of ETGs and their evolution with redshift, based on the redshift distribution of strong gravitational lenses. We compile 200 galaxy-scale strong lensing systems with complete information, including lens and source redshifts, Einstein radii, and spectroscopic velocity dispersions. To account for the heterogeneous construction of the compiled sample, we model a lens-redshift-dependent selection function for each survey category. By combining the lens-redshift distribution with velocity dispersion function (VDF) and spherical Jeans-based mapping, we obtain population-level constraints on the total mass-density slope gamma in three lens-redshift intervals, marginalizing over the luminosity-density slope delta and the stellar orbital anisotropy beta. We find that the total mass density of ETGs remains close to isothermal at low and intermediate redshifts (γ=2.00−0.15+0.07γ=2.00^{+0.07}_{-0.15} at 0 < z_l < 0.5 and γ=1.96−0.26+0.16γ=1.96^{+0.16}_{-0.26} at 0.5 < z_l < 0.6), with a shallower effective slope at higher redshift (γ=1.81−0.21+0.17γ=1.81^{+0.17}_{-0.21} at 0.6<zl≲1.00.6<z_l\lesssim1.0). The marginalized posteriors of delta and beta remain broad, consistent with δ≃2δ\simeq2 and with isotropic orbits in all three redshift intervals. These results are consistent with the redshift trend predicted by Horizon-AGN simulations, in which active galactic nucleus (AGN) feedback, together with stellar assembly and merging, contributes to the structural evolution of ETGs. Applied to the much larger lens samples expected from future missions such as the Chinese Space Station Telescope (CSST), this method will enable tighter tests of density-slope evolution and its connection to galaxy formation processes.

Introduction

The mass distribution of galaxies has been the focus of many observational studies, as it provides a direct probe of both galaxy-formation physics and the connection between luminous and dark matter ([108, 135, 99]). On galactic scales, the inner density profile reflects the coupled evolution of baryons and dark matter, offering key insights into star-formation efficiency, feedback, halo response, and the assembly history of galaxies ([140, 106]). These processes redistribute matter within galaxies and leave observable imprints on velocity dispersion, stellar mass-to-light ratio, and dynamical structure. Radial mass profiles are also essential for connecting galaxy-scale measurements to halo modeling and structure formation in the Λ\LambdaCDM framework ([100, 68, 64]), and they provide useful tests of dark-matter physics on sub-galactic to halo scales ([19, 131, 81]). Among different galaxy populations, early-type galaxies (ETGs) are particularly well suited for such studies because their quiescent stellar populations, high central densities, and relatively regular morphologies make them favorable targets for both dynamical and strong lensing analyses ([72, 9]). The total density slope of ETGs therefore provides a compact and physically informative quantity with which to study the interplay between baryonic assembly and dark-matter structure across cosmic time. Massive ETGs are often found to have total mass-density profiles that are close to isothermal over the radial range probed by stellar dynamics and strong lensing ([29, 28]). This near-isothermal structure is commonly interpreted as the result of the coupled contribution of a centrally concentrated stellar component and a more extended dark-matter halo, shaped by baryonic assembly, feedback, quenching, and merging processes ([95, 88, 57]). Several theoretical and empirical profiles have been used to describe galaxy mass distributions. The Navarro-Frenk-White (NFW) profile, motivated by collisionless Λ\LambdaCDM simulations, provides a useful description of dark-matter halos, but the inner regions of massive galaxies are strongly affected by baryonic physics ([43, 30]). For lensing and dynamical studies of ETGs, the total mass distribution is therefore often described by an effective power-law profile, ρtot(r)∝r−γ\rho_{\mathrm{tot}}(r) \propto r^{-\gamma}, where γ=2\gamma= 2 corresponds to the isothermal case [94] (Xu et al. 2016; Li et al. 2018). This compact parametrization captures the combined stellar and dark-matter distribution over the radial range most relevant for strong lensing and provides a useful way to quantify possible structural evolution.

Several complementary observational methods have been used to measure ETG mass profiles and to test whether their effective total density slopes evolve with redshift. Stellar-dynamical analyses infer the three-dimensional mass distribution from stellar motions or other tracers, typically through Jeans or orbit-based modeling and with assumptions about the tracer distribution and orbital anisotropy [e.g. 98, 40]. These studies provide direct dynamical information, although high-quality spectroscopy, especially spatially resolved spectroscopy at high redshift, remains observationally demanding [133, 89]. Strong gravitational lensing provides an independent and complementary constraint through the projected mass enclosed within the Einstein radius, however, lensing-only inferences of density slopes are affected by degeneracies such as the mass-sheet degeneracy and the coupling between density slope and external convergence [e.g. 50, 109]. Joint lensing-and-dynamics (L&D) analyses combine these complementary observables and have provided some of the strongest constraints on individual ETG lenses [e.g. 11, 72] (Li et al. 2018).

Despite the broad agreement that massive ETGs are approximately isothermal, the reported normalization and redshift dependence of γ\gamma vary among samples and methods. Some lensing-only analyses have reported evidence for a positive trend of γ\gamma with redshift [47], whereas several joint L&D analyses have found a decreasing total density slope with redshift [103, 17, 123, 56] (Sheu et al. 2025). Dynamics-only studies of nearby ETGs have often reported little or no redshift evolution in the inferred total slope [e.g. 98, 40, 39]. However, the recent analysis of Mozumdar et al. [86], based on deep Multi Unit Spectroscopic Explorer (MUSE-DEEP) spectroscopy and the Mapping Nearby Galaxies at Apache Point Observatory (MaNGA) survey, reports a mild but ∼5σ\sim5\sigma-significant negative evolution of dγ/dz≈−0.20±0.03d\gamma/dz \approx-0.20 \pm0.03, a dynamics-only result that is closer in sign to the negative evolution reported by several lensing-dynamics studies. These differences may reflect genuine departures from a single power-law description [47], differences in the radial range probed by each technique, or sample-selection effects. Measuring the population-level redshift dependence of γ\gamma therefore remains important for understanding ETG structural evolution and for assessing the assumptions entering strong lensing statistics.

The rapid growth of strong-lens samples is opening a new regime for such measurements. Spectroscopic searches in the Sloan Digital Sky Survey (SDSS) established well-characterized samples of galaxy-scale lenses [18], while wide-field imaging surveys have produced increasingly large catalogues of lens candidates, including those from the Hyper Suprime-Cam (HSC) Subaru Strategic Program and the Ultraviolet Near-Infrared Optical Northern Survey (UNIONS) [119, 107, 110, 125]. Recent work has further demonstrated the value of combining imaging-based searches with wide-area spectroscopy, as illustrated by the spectroscopic confirmation of HSC lens candidates with the Dark Energy Spectroscopic Instrument (DESI) [115]. The first Euclid Quick Data Release alone yielded 497 galaxy-scale lens candidates over 63 deg263\,\mathrm{deg}^{2} [48], illustrating the scale of the samples that will become available from Euclid and other forthcoming wide-field surveys. Forecasts for Euclid, the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST), Roman, and the Chinese Space Station Telescope (CSST) indicate that these facilities may together deliver lens samples of order 10510^{5} systems [35, 139, 26]. The resulting samples will span different discovery channels, imaging depths, angular resolutions, and spectroscopic follow-up strategies, making the treatment of survey selection an increasingly important component of population-level lensing analyses [74].

In this work, we use strong lensing statistics to extend the lensing-dynamics connection to the population level by combining the lens-redshift distribution with the velocity-dispersion function (VDF) of the lens population [38]. Rather than performing detailed dynamical modeling for each individual lens, we apply a statistical strong lensing framework in which ensemble lensing observables are combined with VDF calibrations and a Jeans-based mapping. For our newly compiled sample of 200 galaxy-scale ETG lenses drawn from different discovery channels, we incorporate a redshift-dependent selection function into the lens-redshift likelihood and investigate the mass profiles of ETG lenses and their possible redshift dependence. The paper is organized as follows: Section 2 describes the lensing statistics and the selection function used in our analysis. Section 3 introduces the observational strong lensing sample. Section 4 presents the inferred density slopes and stability tests, based on the redshift distribution of available gravitational lenses. Section 5 summarizes the main conclusions and discussion, forecasting the performance of our methodology in future missions such as CSST. Appendix A gives the details of the selection function calculation.

Methodology

Strong lensing optical depth and the lens galaxy model

Our method utilizes the strong lensing optical depth, which describes the lensing probability density for a background source at zsz_{\mathrm{s}} being lensed by a foreground galaxy at redshift zlz_{\mathrm{l}}. Since galaxy-scale strong lensing events are rare, the lensing probability can be described in terms of an optical-depth formalism. Following the framework of Ofek et al. [91]; Cao & Zhu [25]; Cao et al. [24], we write the differential optical-depth kernel evaluated at a given Einstein radius θE\theta_{\mathrm{E}} as

dτdzl(θE,zl∣zs)=n(θE,zl)⋅(1+zl)3⋅Scr⋅c dtdzl,(1)\frac{d\tau}{dz_{\mathrm{l}}}(\theta_{\mathrm{E}},z_{\mathrm{l}}|z_{\mathrm{s}}) = n(\theta_{\mathrm{E}},z_{\mathrm{l}})\cdot(1+z_{\mathrm{l}})^3 \cdot S_{\mathrm{cr}} \cdot\frac{c\,dt}{dz_{\mathrm{l}}}, \tag*{(1)}

where n(θE,zl)n(\theta_{\mathrm{E}},z_{\mathrm{l}}) is the comoving number density of lensing galaxies characterized by Einstein radius θE\theta_{\mathrm{E}} at lens redshift zlz_{\mathrm{l}}, and the factor (1+zl)3(1+z_{\mathrm{l}})^3 converts it to the corresponding physical number density. The lensing cross section is approximated as

Scr=π(θEDl)2,(2)S_{\mathrm{cr}} = \pi(\theta_{\mathrm{E}}D_{\mathrm{l}})^2, \tag*{(2)}

where DlD_{\mathrm{l}} denotes the angular diameter distance from the observer to the lens galaxy. This term represents the characteristic lensing area associated with the observed Einstein radius. The line-of-sight path-length element is

c dtdzl=cH(zl)(1+zl),(3)\frac{c\,dt}{dz_{\mathrm{l}}} = \frac{c}{H(z_{\mathrm{l}})(1+z_{\mathrm{l}})}, \tag*{(3)}

where H(zl)H(z_l) denotes the Hubble parameter at redshift zlz_l. Throughout this work we adopt a flat Λ\LambdaCDM cosmology with Ωm=0.3\Omega_m = 0.3 and H0=70 km s−1 Mpc−1H_0 = 70\ \mathrm{km}\,\mathrm{s}^{-1}\,\mathrm{Mpc}^{-1}.

Based on the framework proposed by Koopmans [70], we adopt a spherical, scale-free model for the lens galaxy, in which the stellar luminosity density ν(r)\nu(r), total mass density ρ(r)\rho(r), and orbital anisotropy β\beta are parameterized as

ν(r)=ν0r−δ,ρ(r)=ρ0r−γ,β(r)=1−⟨σθ2⟩⟨σr2⟩.(4)\nu(r) = \nu_0 r^{-\delta}, \qquad\rho(r) = \rho_0 r^{-\gamma}, \qquad\beta(r) = 1 - \frac{\langle\sigma_\theta^2 \rangle}{\langle\sigma_r^2 \rangle}. \tag*{(4)}

Here ν0\nu_0 and ρ0\rho_0 are normalization constants, and σr\sigma_r and σθ\sigma_\theta are the radial and tangential components of the stellar velocity dispersion. In this parameterization, γ\gamma describes the total mass-density slope, δ\delta describes the stellar luminosity-density slope, and β\beta is the constant orbital anisotropy parameter. This scale-free form provides a compact phenomenological description that is well suited for statistical lensing applications.

Within this framework, the spherical Jeans equation provides the connection between the mass distribution and the stellar kinematics. Specifically, it links the stellar velocity dispersion σref\sigma_{\mathrm{ref}}, defined with respect to a reference radius RrefR_{\mathrm{ref}}, to the Einstein radius of the lens galaxy, RE=DlθER_E = D_l\theta_E. Depending on the dataset, RrefR_{\mathrm{ref}} may correspond to the spectroscopic aperture radius RapR_{\mathrm{ap}}, the effective radius ReR_e, or a standardized aperture such as Re/8R_e/8. Following Koopmans [70], the relation can be expressed as

σref2=1π(GMdynRE)1f(γ,δ,β)(RrefRE)2−γ,(5)\sigma_{\mathrm{ref}}^2 = \frac{1}{\pi}\left(\frac{G M_{\mathrm{dyn}}}{R_E}\right)\frac{1}{f(\gamma,\delta,\beta)}\left(\frac{R_{\mathrm{ref}}}{R_E}\right)^{2-\gamma}, \tag*{(5)}

where ξ≡γ+δ−2\xi\equiv\gamma+ \delta- 2 and

f(γ,δ,β)=[2π(δ−3)(ξ−3)(ξ−2β)Γ(δ/2)Γ(γ/2)Γ[(δ−1)/2]Γ[(γ−1)/2](Γ[(ξ−1)/2]Γ(ξ/2)−βΓ[(ξ+1)/2]Γ[(ξ+2)/2])]−1.(6)f(\gamma,\delta,\beta) = \left[ \frac{2\sqrt{\pi}(\delta-3)}{(\xi-3)(\xi-2\beta)} \frac{\Gamma(\delta/2)\Gamma(\gamma/2)} {\Gamma[(\delta-1)/2]\Gamma[(\gamma-1)/2]} \left( \frac{\Gamma[(\xi-1)/2]}{\Gamma(\xi/2)} -\beta\frac{\Gamma[(\xi+1)/2]}{\Gamma[(\xi+2)/2]} \right) \right]^{-1}. \tag*{(6)}

When assuming γ=δ\gamma= \delta and β=0\beta= 0, the commonly used γ\gamma-only approximation can be written as

f(γ)=−1π(5−2γ)(1−γ)3−γΓ(γ−1)Γ(γ−3/2)[Γ(γ/2−1/2)Γ(γ/2)]2.(7)f(\gamma) = -\frac{1}{\sqrt{\pi}}\frac{(5-2\gamma)(1-\gamma)}{3-\gamma}\frac{\Gamma(\gamma-1)}{\Gamma(\gamma-3/2)} \left[\frac{\Gamma(\gamma/2-1/2)}{\Gamma(\gamma/2)}\right]^2. \tag*{(7)}

For γ=2\gamma= 2 in Eq. (7), Eq. (5) reduces to the singular isothermal sphere (SIS) limit, with the velocity-dispersion relation σref2(R)=1π[GMdynRE]\sigma_{\mathrm{ref}}^2(R) = \frac{1}{\pi}\left[\frac{G M_{\mathrm{dyn}}}{R_E}\right]. We follow the same dynamical mass–Einstein radius formulation adopted in Cao et al. [22]. The dynamical mass associated with the Einstein radius, derived from Eq. (5), is expressed as

Mdyn=πGσref2RE(RERref)2−γf(γ,δ,β)=πGσref2DlθE(θEθref)2−γf(γ,δ,β),(8)\begin{aligned} M_{\mathrm{dyn}} = \frac{\pi}{G}\sigma_{\mathrm{ref}}^2 R_E\left(\frac{R_E}{R_{\mathrm{ref}}}\right)^{2-\gamma}f(\gamma,\delta,\beta) \\ &= \frac{\pi}{G}\sigma_{\mathrm{ref}}^2 D_l\theta_E\left(\frac{\theta_E}{\theta_{\mathrm{ref}}}\right)^{2-\gamma}f(\gamma,\delta,\beta), \tag*{(8)} \end{aligned}

where θref≡Rref/Dl\theta_{\mathrm{ref}} \equiv R_{\mathrm{ref}}/D_l denotes the corresponding angular reference radius. The lensing mass projected within the Einstein radius follows from the critical surface density (Meneghetti [82])

Mlens=c24GDsDlDlsθE2.(9)M_{\mathrm{lens}} = \frac{c^2}{4G}\frac{D_sD_l}{D_{ls}}\theta_E^2. \tag*{(9)}

where DsD_s and DlsD_{ls} represent the angular diameter distances from the observer to the source and from the lens to the source, respectively. The Einstein radius is a robust lensing observable and is less sensitive than the full mass model to the detailed radial parameterization of the lens potential ([50, 132, 55, 46]). This makes it suitable for a statistical analysis of heterogeneous lens samples.

From the equations above, the reference angular scale θref\theta_{\mathrm{ref}} enters Eq. (8) through the factor (Rref/RE)2−γ(R_{\mathrm{ref}}/R_E)^{2-\gamma}, which accounts for the aperture dependence of the velocity-dispersion normalization. By combining the dynamical mass in Eq. (8) with the lensing mass inferred from Eq. (9), we obtain an explicit relation between the mass-density slope γ\gamma and the Einstein radius,

θE=[4π(σrefc)2θrefγ−2DlsDsf(γ,δ,β)]1γ−1.(10)\theta_E = \left[4\pi\left(\frac{\sigma_{\mathrm{ref}}}{c}\right)^2\theta_{\mathrm{ref}}^{\gamma-2}\frac{D_{ls}}{D_s}f(\gamma,\delta,\beta)\right]^{\frac{1}{\gamma-1}}. \tag*{(10)}

Building on this relation, the characteristic Einstein radius corresponding to the characteristic VDF velocity dispersion can be expressed as

θE∗=λ(e)[4π(σ∗refc)2θrefγ−2DlsDsf(γ,δ,β)]1γ−1.(11)\theta_{E*} = \lambda(e)\left[4\pi\left(\frac{\sigma_{*{\rm ref}}}{c}\right)^2\theta_{\mathrm{ref}}^{\gamma-2}\frac{D_{ls}}{D_s}f(\gamma,\delta,\beta)\right]^{\frac{1}{\gamma-1}}. \tag*{(11)}

Here σ∗ref\sigma_{*{\rm ref}} is the characteristic velocity dispersion of the VDF evaluated at its reference radius. The factor λ(e)\lambda(e) accounts for the dynamical normalization of non-spherical galaxies (Keeton & Kochanek [69]), and ee denotes the projected ellipticity of the lens mass distribution, related to the projected minor-to-major axis ratio by q=1−eq = 1-e, consistent with the singular isothermal ellipsoid convention adopted by Oguri et al. [93]. In this work, we adopt a fixed mean ellipticity e=0.25e = 0.25, consistent with the axis-ratio distributions of ETGs observed in the SDSS survey (Bernardi et al. [13]), and approximate the three-dimensional shapes of lens galaxies as an equal-weight combination of oblate and prolate isothermal spheroids:

λ(e)=0.5λobl(e)+0.5λpro(e),(12)\lambda(e) = 0.5\lambda_{\mathrm{obl}}(e) + 0.5\lambda_{\mathrm{pro}}(e), \tag*{(12)}

with

λobl(e)≈exp⁡(0.108e+0.180e2+0.797e5),\lambda_{\mathrm{obl}}(e) \approx\exp\left(0.108\sqrt{e}+0.180e^2+0.797e^5\right),
λpro(e)≈1−0.258e+0.827e6.(13)\lambda_{\mathrm{pro}}(e) \approx1-0.258e+0.827e^6. \tag*{(13)}

Since the VDF of ETGs evolves with redshift, we adopt different sets of Schechter-function parameters (α,β′,n∗,σ∗)(\alpha,\beta',n_*,\sigma_*) for different redshift intervals, as summarized in Table 3. The VDF is written as

SourceRedshiftn∗n_\ast (×10−3 h3 Mpc−3\times10^{-3}\,h^3\,\mathrm{Mpc}^{-3})σ∗ref\sigma_{\ast\mathrm{ref}} (km s−1\mathrm{km\,s}^{-1})α\alphaβ′\beta'
Choi et al. [34]local8.0161±5161 \pm52.32±0.12.32 \pm0.12.67±0.072.67 \pm0.07
Montero-Dorta et al. [83]0.555.48±1.525.48 \pm1.52116.40±15.12116.40 \pm15.125.03±1.035.03 \pm1.032.23±0.172.23 \pm0.17
Taylor et al. [129]0.6–1.01.96−0.37+0.911.96^{+0.91}_{-0.37}183−69+50183^{+50}_{-69}2.2−1.2+2.02.2^{+2.0}_{-1.2}2.8−0.8+1.12.8^{+1.1}_{-0.8}

Table 3. Best-fit parameters of the modified Schechter function for the velocity dispersion function (VDF) adopted in this work.

dNdσref=n∗(σrefσ∗ref)αexp⁡[−(σrefσ∗ref)β′]×β′Γ(α/β′)1σref,(14)\frac{dN}{d\sigma_{\mathrm{ref}}} = n_*\left(\frac{\sigma_{\mathrm{ref}}}{\sigma_{*{\rm ref}}}\right)^\alpha\exp\left[-\left(\frac{\sigma_{\mathrm{ref}}}{\sigma_{*{\rm ref}}}\right)^{\beta'}\right] \times\frac{\beta'}{\Gamma(\alpha/\beta')}\frac{1}{\sigma_{\mathrm{ref}}}, \tag*{(14)}

which gives the comoving number density of galaxies per unit velocity dispersion. Here σref\sigma_{\mathrm{ref}} denotes the stellar velocity dispersion defined at the same aperture convention as the adopted VDF, and σ∗ref\sigma_{*{\rm ref}} is the corresponding characteristic velocity dispersion. The parameters α\alpha and β′\beta' control the low-velocity slope and the high-velocity cutoff of the VDF, respectively, while n∗n_* and σ∗ref\sigma_{*{\rm ref}} represent the characteristic comoving number density and velocity dispersion.

Because the characteristic velocity dispersions adopted in different VDF measurements are either measured within or corrected to different aperture conventions, we assign a reference angular scale θref\theta_{\mathrm{ref}} to each VDF dataset. For Choi et al. [34], stellar velocity dispersions were measured from SDSS spectra obtained with 3″-diameter fibres and corrected to a standard aperture of Re/8R_{\mathrm{e}}/8. To estimate the physical scale corresponding to this aperture correction, we use the early-type SDSS size–velocity-dispersion relation of Hyde & Bernardi [63], whose velocity dispersions are defined with the same Re/8R_{\mathrm{e}}/8 aperture convention. For σ∗=161 km s−1\sigma_{*}=161\ \mathrm{km\ s}^{-1}, their ReR_{\mathrm{e}}–σ\sigma scaling relation gives Re≃2.6 kpcR_{\mathrm{e}}\simeq2.6\ \mathrm{kpc}, and hence Re/8≃0.33 kpcR_{\mathrm{e}}/8\simeq0.33\ \mathrm{kpc}. At the representative lens redshift of the Choi et al. [34] sample, zl=0.07z_{\mathrm{l}}=0.07, this gives θref≃0.24′′\theta_{\mathrm{ref}}\simeq0.24''. For Montero-Dorta et al. [83], we use their “raw” VDF, in which stellar velocity dispersions are measured directly within the BOSS 2″-diameter spectroscopic fibre (θref=1.0′′\theta_{\mathrm{ref}}=1.0''), without any aperture correction. For Taylor et al. [129], the LEGA-C stellar velocity dispersions are measured from spatially integrated one-dimensional spectra. Unlike a central small-aperture velocity dispersion, this quantity reflects the luminosity-weighted line broadening of the stellar light collected by the slit. The measured dispersion is therefore tied to the spatial distribution of the stellar light that dominates the integrated spectrum. As a characteristic scale of this main stellar light distribution, we adopt the projected half-light radius ReR_{\mathrm{e}} as the effective radius and set θref=θRe\theta_{\mathrm{ref}}=\theta_{R_{\mathrm{e}}}. For quiescent galaxies at z∼0.8z\sim0.8, adopting a typical LEGA-C stellar mass of log⁡(M∗/M⊙)≃10.75\log(M_{*}/M_{\odot})\simeq10.75 (Straatman et al. [126]), the quiescent mass-size relation of van der Wel et al. [133] gives Re≃2.86 kpcR_{\mathrm{e}}\simeq2.86\ \mathrm{kpc}. At zl=0.8z_{\mathrm{l}}=0.8, this corresponds to θref≃0.38′′\theta_{\mathrm{ref}}\simeq0.38''. We note that the systematic uncertainty associated with different aperture conventions is expected to be limited to the ∼0.06\sim0.06 dex offset between σRe\sigma_{R_{\mathrm{e}}} and σ(Re/8)\sigma_{(R_{\mathrm{e}}/8)} reported by Taylor et al. [129], which remains smaller than the statistical uncertainties of the VDF Schechter parameters.

Returning to Eq. (1), the comoving number density n(θE,zl)n(\theta_{\mathrm{E}},z_{\mathrm{l}}) can be related to the VDF in Eq. (14) via a change of variables,

n(θE,zl) dθE=dNdσref(σref,zl) dσref.(15)n(\theta_{\mathrm{E}},z_{\mathrm{l}})\,d\theta_{\mathrm{E}}=\frac{dN}{d\sigma_{\mathrm{ref}}}(\sigma_{\mathrm{ref}},z_{\mathrm{l}})\,d\sigma_{\mathrm{ref}}. \tag*{(15)}

Using Eq. (10), the ratio σref/σ∗ref\sigma_{\mathrm{ref}}/\sigma_{*\mathrm{ref}} can be replaced by the corresponding Einstein-radius ratio,

R≡σrefσ∗ref=[λ(e)θEθE∗](γ−1)/2.(16)R\equiv\frac{\sigma_{\mathrm{ref}}}{\sigma_{*\mathrm{ref}}}=\left[\frac{\lambda(e)\theta_{\mathrm{E}}}{\theta_{\mathrm{E}*}}\right]^{(\gamma-1)/2}. \tag*{(16)}

Under this transformation, the differential optical-depth kernel evaluated at the observed Einstein radius becomes

dτdzl(zl∣zs,θE)=n∗(γ−1)β′Γ(α/β′)Dl2(1+zl)2c2H(zl)×θERαexp⁡(−Rβ′).(17)\frac{d\tau}{dz_{\mathrm{l}}}(z_{\mathrm{l}}|z_{\mathrm{s}},\theta_{\mathrm{E}})=n_{*}\frac{(\gamma-1)\beta^{\prime}}{\Gamma(\alpha/\beta^{\prime})}\frac{D_{\mathrm{l}}^{2}(1+z_{\mathrm{l}})^{2}c}{2H(z_{\mathrm{l}})}\times\theta_{\mathrm{E}}R^{\alpha}\exp(-R^{\beta^{\prime}}). \tag*{(17)}

For a given strong lensing system ii, the relative probability density of finding its lens galaxy at redshift zl,iz_{\mathrm{l},i} is given by the differential optical depth of Eq. (17) evaluated at zl,iz_{\mathrm{l},i}, normalized by its integral along the line of sight to the source. This normalization implicitly assumes that every lens system able to produce the observed configuration has the same probability of being recorded. In practice, a system that lies at 0<zl<zs0<z_{\mathrm{l}}<z_{\mathrm{s}} and produces the observed Einstein radius θE\theta_{\mathrm{E}} does not automatically enter our lens catalogue: it must first be identified as a lens and have both its lens and source redshifts measured. We therefore introduce a selection function SS, defined as the probability that

a system with given intrinsic properties is included in our compilation,

S=P(selected∣intrinsic properties),(18)S=P(\mathrm{selected}\mid\mathrm{intrinsic\ properties}), \tag*{(18)}

so that the conditional likelihood can be written as

pi(p)=p(zl,i∣θE,i,zs,i,p)=dτdzl(θE,i,zl∣zs,i)S∣zl=zl,i∫0zs,idτdzl(θE,i,zl′∣zs,i)S dzl′.(19)p_i(\mathbf{p})=p(z_{\mathrm{l},i}\mid\theta_{\mathrm{E},i},z_{\mathrm{s},i},\mathbf{p})=\frac{\left.\dfrac{d\tau}{dz_{\mathrm{l}}}(\theta_{\mathrm{E},i},z_{\mathrm{l}}\mid z_{\mathrm{s},i})S\right|_{z_{\mathrm{l}}=z_{\mathrm{l},i}}}{\int_{0}^{z_{\mathrm{s},i}}\frac{d\tau}{dz_{\mathrm{l}}}(\theta_{\mathrm{E},i},z_{\mathrm{l}}^{\prime}\mid z_{\mathrm{s},i})S\,dz_{\mathrm{l}}^{\prime}}. \tag*{(19)}

The integral in the denominator runs over the full foreground path to the source, 0<zl<zs,i0<z_{\mathrm{l}}<z_{\mathrm{s},i}, following the standard lens-redshift test. Eq. (19) therefore constrains the shape of the lens-redshift distribution at a given Einstein radius, not the absolute number of lenses.

The lens selection function

Whether a strong lens system ends up in our sample depends on how it was found. Systems with confirmed lensing features and with both zlz_{\mathrm{l}} and zsz_{\mathrm{s}} measured have entered existing samples through three main routes.

Image-based identification. Here the system is first recognized from lensing configurations such as multiple images, lensed arcs, or Einstein rings. The image-based identification has been done in two main ways. The traditional approach is to search wide-field imaging for prominent arcs or rings, which requires the background source to be bright enough, the lensed images to be well resolved, and the light of the foreground galaxy not to swamp them. Early surveys such as SL2S and LSD identified candidates in this way and then obtained deep spectroscopy of each target individually on large telescopes, from which both zlz_{\mathrm{l}} and zsz_{\mathrm{s}} were measured. More recently, wide-area surveys (e.g. Shu & Li [115]; Alves de Oliveira et al. [5]; Tran et al. [130]) have confirmed large numbers of candidates in imaging first and then cross-matching those candidates with existing SDSS or DESI spectra to obtain their redshifts and velocity dispersions.

Spectroscopic identification. Here the system is first flagged from the spectrum of the foreground galaxy. Within the magnitude limits of a systematic spectroscopic survey, a galaxy whose fibre spectrum shows emission lines from a higher-redshift object in addition to its own absorption features is selected as a candidate, provided that these lines stand out above the residual noise left once the foreground spectrum has been subtracted. Such candidates then trigger follow-up imaging, typically with HST, and are retained once the lensing configuration is confirmed. The classic SLACS, S4TM, BELLS, and BELLS GALLERY strong lensing samples were all assembled in this way.

Source-triggered discovery. In a small number of cases the discovery starts from the source rather than from the lens: a bright background object picked up in a radio or submillimetre survey, such as MG2016+112 or SDP.81, is later recognized as lensed, and the foreground galaxy is identified only afterwards, entering the sample passively and with its own properties playing no role in the discovery.

For a single survey, the selection function can be constructed in detail. Ferrami & Wyithe [52, 53], for instance, modelled the detectability of SL2S lenses by folding the detection signal-to-noise ratio, the bright-image fraction, and the seeing-limited resolution into the probability of a system being found. They showed that these effects substantially reduce the number of detectable lenses and reshape the redshift and mass distributions of the observed sample. Sonnenfeld [117] worked in a different direction, building an empirical model of the SLACS selection function to correct for the over-representation of galaxies with high velocity dispersions, and found that the mean total mass density slope of the parent early-type population differs from that of the lens sample by Δγ∼0.01\Delta\gamma\sim0.01. Both approaches rely on a complete description of the imaging and of the search algorithm of one particular survey. The sample used in our work, however, is drawn from many sources with heterogeneous selection criteria, and for most of them no selection function has been given. Compiled samples of this kind will become more common as future surveys enlarge the available catalogues. We therefore need a strategy that handles the different selection effects system by system within a common framework.

Capelo & Natarajan [27] tested the robustness of the lens-redshift test and found that lenses discovered through their background sources introduce no appreciable bias in the lens-redshift distribution, whereas for systems found through the foreground galaxy or its environment the selection function depends strongly on the lens redshift itself. This distinction between what is set by the source and what is set by the lens, together with the fact that the integral in Eq. (19) runs over zlz_l alone, leads us to write the selection function as

S=SA(θE,zs) SB(zl;θE,zs).(20)S = S_A(\theta_E, z_s)\,S_B(z_l;\theta_E,z_s). \tag*{(20)}

Here SAS_A collects the factors that stay fixed as the lens is moved along the line of sight, while SBS_B carries the dependence on zlz_l. Since θE,i\theta_{E,i} and zs,iz_{s,i} are held fixed by the observations, Eq. (20) simply states where the zlz_l dependence of each factor resides. Factors that are set by both the source and the lens are split in the same way: the detectability of a lensed image, for example, depends on the brightness of the arc as well as on the contamination from the foreground light, and only the part of it that changes with zlz_l is assigned to SBS_B.

SAS_A therefore collects the factors that depend on the background source properties and govern whether the source is detected and recognized: the brightness and angular separation of the lensed images or arc; the luminosities of the background emission lines and whether they fall within the spectral coverage of the spectrograph; the alignment of the images with respect to the fibre aperture; the success rate of the source-redshift measurement; and the lens modeling requirement that the system show an identifiable lensing configuration in the available imaging. Magnification bias also belongs here. Although it alters the effective number density of sources, at the given conditional observables (θE,zs)(\theta_E,z_s) it is determined by the source and does not change with zlz_l.

SBS_B acts through the foreground galaxy: whether it falls within the target magnitude windows of its parent spectroscopic survey; the probability that its redshift and velocity dispersion are successfully measured under a fixed exposure strategy; and the suppression of the background arc or emission lines by the light of the foreground galaxy itself. Our isolation criterion and the zlz_l-dependent part of the lens-light subtraction operate through the same apparent brightness, and are grouped here as well.

Since SAS_A does not depend on zlz_l, it can be taken outside the integral, and Eq. (19) reduces to

picorr=SA[dτdzl∣zl,iSB]SA∫0zs,idτdzl′SB dzl′=[dτdzl∣zl,iSB]∫0zs,idτdzl′SB dzl′.(21)p_i^{\mathrm{corr}} = \frac{S_A\left[\left.\frac{d\tau}{dz_l}\right|_{z_{l,i}}S_B\right]} {S_A\int_0^{z_{s,i}}\frac{d\tau}{dz'_l}S_B\,dz'_l} = \frac{\left[\left.\frac{d\tau}{dz_l}\right|_{z_{l,i}}S_B\right]} {\int_0^{z_{s,i}}\frac{d\tau}{dz'_l}S_B\,dz'_l}. \tag*{(21)}

The remaining redshift-dependent selection enters through SBS_B. For the heterogeneous compilation considered here, we model the foreground-dependent effects that can be characterized across the sample through the survey magnitude windows and foreground-light suppression. We assess the sensitivity of the inferred slopes to this model in Sect. 4.2. Based on the three discovery channels summarized above, we consider these two contributions to SBS_B in turn.

The first is the observational brightness window. Lenses identified spectroscopically enter the sample only when the foreground galaxy falls within the target-selection magnitude window [mbright,mlim][m_{\mathrm{bright}},m_{\mathrm{lim}}] of its parent survey, and the same thresholds apply to lenses found in imaging whose redshifts or velocity dispersions come from a cross-match with a spectroscopic database. This window determines both whether a system is observed and whether it has a spectrum that provides a redshift and velocity dispersion to satisfy our criteria. For lenses that were followed up with dedicated deep spectroscopy, the exposure could be tailored to each target and integrated until both redshifts were measured, so no such window applies.

The second is the suppression of the background signal by the foreground light. In spectroscopic identification the background emission line must stand above the residual noise left by the subtraction of the foreground spectrum, and that noise grows with the continuum flux of the lens galaxy. In image-based identification the lensed images must be separable from the foreground light which governs both whether the configuration is recognised as lensing and whether a candidate is convincing enough to be followed up spectroscopically. The two mechanisms are physically distinct, but act in the same direction: the brighter the foreground galaxy, the harder the background signal is to detect.

What the two categories have in common is that both act through the apparent brightness of the foreground galaxy, SBS_B is therefore governed by the apparent magnitude mlensm_{\mathrm{lens}} of the lens galaxy. As mlensm_{\mathrm{lens}} approaches or crosses either the faint-end target-selection limit mlimm_{\mathrm{lim}} or the bright-end saturation limit mbrightm_{\mathrm{bright}}, the probability that the system enters the sample drops. In turn, mlensm_{\mathrm{lens}} depends on the redshift of the lens, since it is determined by the rest-frame absolute magnitude Mlens(zl)M_{\mathrm{lens}}(z_l), the distance modulus μ(zl)\mu(z_l), the KK-correction K(zl)K(z_l), and the evolutionary correction E(zl)E(z_l),

mlens(zl)=Mlens(zl)+μ(zl)+K(zl)+E(zl).(22)m_{\mathrm{lens}}(z_l)=M_{\mathrm{lens}}(z_l)+\mu(z_l)+K(z_l)+E(z_l). \tag*{(22)}

For the absolute magnitude we follow the Faber-Jackson velocity-dispersion–magnitude relation of Bernardi et al. [14], Mlens(zl)=M∗+A(log⁡10σref(zl)−V∗)M_{\mathrm{lens}}(z_l)=M_*+A(\log_{10}\sigma_{\mathrm{ref}}(z_l)-V_*), where σref(zl)=σ∗refR(zl;γ,δ,β)\sigma_{\mathrm{ref}}(z_l)=\sigma_*^{\mathrm{ref}}R(z_l;\gamma,\delta,\beta) is the velocity dispersion the galaxy would need in order to produce the observed Einstein radius if it were placed at zlz_l, obtained from the ratio RR defined in Eq. (16). The relation is calibrated on an SDSS early-type galaxy sample, in which the joint distribution of (M,log⁡10σ)(M,\log_{10}\sigma) is described by a bivariate Gaussian. Conditioning on the velocity dispersion gives P(M∣log⁡10σ)=N(M∗+A(log⁡10σ−V∗),w2)P(M\mid\log_{10}\sigma)=\mathcal{N}(M_*+A(\log_{10}\sigma-V_*),w^2), so that the absolute magnitude is normally distributed around M∗+A(log⁡10σ−V∗)M_*+A(\log_{10}\sigma-V_*) with scatter ww. Both the slope AA and the scatter ww are set by the covariance parameters of Bernardi et al. [14], as detailed in Appendix A.1. For the remaining terms in Eq. (22), the KK-correction removes the bandpass shift caused by redshift and converts the observed magnitude to a common rest-frame band, while EE is the passive luminosity-evolution correction, which accounts for the change in the intrinsic luminosity of the galaxy as its stellar population ages with redshift. The methods, templates, and numerical values used for these three terms are given in Appendix A.2.

From (22), for a given velocity dispersion the absolute magnitude of the lens galaxy is not uniquely determined but follows a distribution. The apparent magnitude should therefore be written as mlens+ϵm_{\mathrm{lens}}+\epsilon, where ϵ∼N(0,w2)\epsilon\sim\mathcal{N}(0,w^2) and ww is the width set by the intrinsic scatter of the Faber–Jackson relation. Whether a system falls within a given magnitude window is thus not a deterministic decision but a probability. For a window mbright<m<mlimm_{\mathrm{bright}}<m<m_{\mathrm{lim}} this probability is

P(mbright<m<mlim)=Φ(mlim−mlensw)−Φ(mbright−mlensw),(23)P(m_{\mathrm{bright}}<m<m_{\mathrm{lim}})=\Phi\left(\frac{m_{\mathrm{lim}}-m_{\mathrm{lens}}}{w}\right)-\Phi\left(\frac{m_{\mathrm{bright}}-m_{\mathrm{lens}}}{w}\right), \tag*{(23)}

where Φ(⋅)\Phi(\cdot) is the standard normal cumulative distribution function. SBS_B can accordingly be written as the product of the brightness-window term and the foreground-light suppression term,

SB(zl)=[Φ(mlim−mlensw)−Φ(mbright−mlensw)][1−ϵΦ(msup−mlensw)].(24)S_B(z_l)=\left[\Phi\left(\frac{m_{\mathrm{lim}}-m_{\mathrm{lens}}}{w}\right)-\Phi\left(\frac{m_{\mathrm{bright}}-m_{\mathrm{lens}}}{w}\right)\right]\left[1-\epsilon\Phi\left(\frac{m_{\mathrm{sup}}-m_{\mathrm{lens}}}{w}\right)\right]. \tag*{(24)}

The first term gives the probability that the lens galaxy is included in the spectroscopic survey. A galaxy that is too faint is unlikely to yield a usable spectrum, while one that is too bright may be rejected because of saturation or fibre cross-talk. Both limits are set by the published specifications of each parent survey. The second term accounts for the suppression of the background signal by the foreground light, which reduces the probability that the system is recognised as a lens or that a redshift is successfully extracted. Here ϵ\epsilon is the maximum loss in detection probability caused by this effect (0≤ϵ≤10\leq\epsilon\leq1), and msupm_{\mathrm{sup}} is the brightness at which the suppression becomes appreciable. For lens magnitudes fainter than msupm_{\mathrm{sup}} the factor tends to unity, and for brighter magnitudes it tends to 1−ϵ1-\epsilon. No external calibration is available for msupm_{\mathrm{sup}}, so we place it at the 10th percentile of the predicted magnitude distribution of the lens sample in each selection category. This location corresponds to the brightest systems in each category and quantifies the largest impact that foreground suppression can have. This factor is evaluated with the predicted ii-band magnitude, as described in Appendix A.3.

Statistical inference and MCMC sampling

The normalization is performed over the full foreground path to the source, 0<zl<zs0<z_l<z_s, following the standard lens-redshift test. This form focuses the inference on the relative lens-redshift probability at the observed Einstein radius rather than on the absolute lensing rate. The total log-likelihood for the observed NlensN_{\mathrm{lens}} sample is then

ln⁡L(p)=∑i=1Nlensln⁡picorr(p),(25)\ln\mathcal{L}(\mathbf{p})=\sum_{i=1}^{N_{\mathrm{lens}}}\ln p_i^{\mathrm{corr}}(\mathbf{p}), \tag*{(25)}

where the model parameter vector is

p=(γ,δ,β).(26)\mathbf{p}=(\gamma,\delta,\beta). \tag*{(26)}

We employ a Bayesian Markov Chain Monte Carlo (MCMC) framework to explore the posterior distributions of the model

parameters. For each redshift interval and VDF choice, we use 32 walkers and 50000 steps, discarding the first 20% of each chain as burn-in. Convergence is assessed via the Gelman–Rubin statistic and the integrated autocorrelation time τ\tau; all chains satisfy R^<1.01\hat{R}<1.01 and Nkept/τ>50N_{\mathrm{kept}}/\tau>50, confirming adequate mixing and convergence. The reported constraints correspond to the median and the 16th-84th percentile credible intervals of the marginalized posterior distributions.

The analysis is performed independently in the three redshift intervals defined by the available VDF measurements. For the fiducial case, we use the central VDF parameters in each interval. To quantify the impact of VDF uncertainties and sample construction, we perform a series of targeted robustness tests to assess the stability of the inferred density slopes. The resulting posterior samples and marginalized likelihood contours serve as the basis for the statistical analysis presented in Section 4.

Observational data

Compilation of the lens sample

Our analysis is based on a newly compiled sample of 200 galaxy-scale strong gravitational lens systems. Motivated by known systematic differences between lens galaxies in group/cluster environments and isolated lenses [10], we focus on galaxy-scale early-type lenses that appear isolated based on currently available imaging and literature information. Lens galaxies reported as disky or late-type in the discovery papers or high-resolution imaging are excluded. To be included in the sample, each system must exhibit identifiable lensing features in available imaging and have the observational quantities required by our statistical framework: lens redshift zlz_l, source redshift zsz_s, Einstein radius θE\theta_E, spectroscopic aperture radius θap\theta_{\mathrm{ap}}, and aperture velocity dispersion σap\sigma_{\mathrm{ap}}. The velocity-dispersion information is used to define a kinematically characterized lens sample and to enable data-quality checks, while the baseline statistical likelihood is constructed from the conditional optical-depth formalism described in Sect. 2. We emphasize that the statistical likelihood does not use the individual σap\sigma_{\mathrm{ap}} measurements as lens-by-lens dynamical constraints; instead, the kinematic information enters the inference statistically through the VDF of the deflector population. For each system we also record the channel through which it was discovered and the spectroscopic survey from which its redshift and velocity dispersion come, since these set the selection-function parameters introduced in Sect. 2 (see Sect. 3.2).

The foundation of our sample is the compilation presented in Chen et al. [33], which contains 161 galaxy-scale strong lens systems. One of them, SL2SJ095921+020638, is listed below among the galaxy-quasar systems under its Guo et al. [58] designation, leaving 160 systems counted from this source. We further supplement the sample by drawing from the compilation of Alves de Oliveira et al. [5]. Rather than adopting their catalogue wholesale, we trace each system individually back to its original discovery reference and verify that it satisfies our selection criteria. Systems that lack a reliable primary source, show evidence of a group or cluster environment, or are identified as non-isolated are excluded. The velocity-dispersion values for these systems, which are useful for characterizing the lens galaxies and for consistency checks with dynamical analyses, are adopted from the SDSS cross-match reported in Alves de Oliveira et al. [5]. After screening, 14 systems are retained, with the accepted systems and their original references listed in Table 1. As discussed in Sect. 2, the source properties do not affect the inference. We therefore also draw galaxy-quasar systems from Guo et al. [58]. Applying the same criteria as above, we include these galaxy-quasar systems: SDSSJ0114+0722, J0203+1612, HSCJ0918−0220, J0949+4208, HSCJ1215−0058, HSCJ1220+0112, SDSSJ1524+4409, and SDSSJ1640+1932. Note that COSMOS5921+0638 in Guo et al. [58] is the same system as SL2SJ095921+020638 in Chen et al. [33]; it is counted once, in the galaxy-quasar block of Table 1, where we adopt the updated measurements of Guo et al. [58]. In addition, we include four systems from recent new surveys that satisfy the same criteria, namely 2MASXJ04035024−0239275, AGEL053724−464702, AGEL233610−020735, and HSCJ0904−0102, which extend the sample toward higher lens redshifts and larger Einstein radii.

Namezlz_lzsz_sθE\theta_E (arcsec)θap\theta_{\mathrm{ap}} (arcsec)σap\sigma_{\mathrm{ap}} (km s−1^{-1})Ref.
From Alves de Oliveira et al. [5]
J085559.0−040917.00.41902.9501.360.70281 ± 22[120]
J020241.4−064611.30.50202.7481.161.00160 ± 26[66]
J023307.1−043838.20.67091.8691.771.00386 ± 111[121]
J023817.8−054555.60.59931.7630.931.00382 ± 86[32]
J090311.6+003906.50.29973.0421.561.00243 ± 16[4]
J103121.0+302658.00.16710.7470.881.50196 ± 13[114]
J111040.4+364924.40.73302.5001.991.00354 ± 116SuGOHIa^{a}
J120623.9+001507.10.56323.1201.171.00271 ± 47[121]
J121052.5−011905.20.69982.2951.161.00334 ± 66[121]
J140929.7−011410.80.58422.3021.241.00187 ± 33[121]
J141300.1−012608.20.74902.6661.141.00243 ± 67[121]
J141815.7+015832.30.55592.1391.401.00219 ± 30[121]
J143213.3+631703.80.12310.6641.261.50194 ± 8[18]
J143607.5+494313.20.12541.2312.031.50282 ± 10[116]
Galaxy–quasar systems
SDSSJ0114+07220.40801.8280.8300.75222 ± 20[84]
J0203+16120.48802.1801.3601.00207 ± 50[75]
HSCJ0918−02200.46000.8031.2800.75360 ± 30[67]
J0949+42080.51001.2701.2300.75290 ± 50[76]
COSMOS5921+06380.55103.1400.7000.75186 ± 30[51]
HSCJ1215−00580.45912.8822.3420.75220 ± 60He et al. (2025)
HSCJ1220+01120.48761.7080.9580.75222 ± 40He et al. (2025)
SDSSJ1524+44090.32001.2100.7900.75222 ± 15[92]
SDSSJ1640+19320.19500.7782.4901.00373 ± 8[136]
From recent surveys
2MASXJ04035024−02392750.06600.1921.4701.50314 ± 5[37]
AGEL053724−4647020.35202.3441.9020.66288 ± 19[105]
AGEL233610−0207350.49422.6621.3921.00272 ± 35[105]
HSCJ0904−01020.95703.4031.2300.50259 ± 56[65]
This work (modeled from Shu et al. 2025)
HSC J0015+01370.37881.1651.3440.75272 ± 21[115]
012018+0011250.59852.9991.1010.75233 ± 24[115]
015310−0423150.28730.6480.8950.75338 ± 37[115]
085046+0039050.88442.8381.8120.75394 ± 77[115]
090938+0028420.77952.5461.0940.75387 ± 65[115]
HSC J1014+03320.67772.1651.2650.75261 ± 31[115]
123636−0035390.51020.7630.8190.75370 ± 32[115]
141649+0138220.43371.3131.1210.75256 ± 19[115]
090402+0314030.84683.3530.8540.75177 ± 26[115]
090548+0047430.85352.9120.9360.75249 ± 34[115]
HSC J1104−00520.87302.4561.3370.75455 ± 91[115]
HSC J1107−00370.81441.4111.6590.75268 ± 47[115]
120657−0102410.83002.7730.7560.75193 ± 45[115]

Table 1. Strong lens systems newly added on top of Chen et al. [33], drawn from Alves de Oliveira et al. [5], Guo et al. [58], recent surveys and our own lens modelling. Columns list the system name, lens redshift zlz_l, source redshift zsz_s, Einstein radius θE\theta_E, angular aperture radius θap\theta_{\mathrm{ap}} for velocity dispersion, aperture-averaged velocity dispersion σap\sigma_{\mathrm{ap}} with its uncertainty, and the data reference. The listed velocity dispersions characterize the spectroscopic properties of the lens galaxies and are not used as individual dynamical likelihood terms in the baseline statistical analysis.

We additionally perform independent lens modeling for systems drawn from Shu & Li [115], who cross-matched strong lens candidates identified in the Hyper Suprime-Cam Subaru Strategic Program Aihara et al. [3] with DESI DR1 spectroscopy, confirming spectroscopic redshifts for both lens and source. DESI is conducting one of the largest spectroscopic surveys to date DESI Collaboration et al. [42], targeting tens of millions of galaxies and quasars with the primary goal of constraining the cosmic expansion history. Its wide-field spectroscopic capability has also proven effective for the confirmation of strong gravitational lens systems. We apply our selection criteria to their sample and model the HSC images using PyAutoLens Nightingale et al. [90]; He et al. (2024), primarily in the ii band, which generally provides the best balance between image quality and lens-galaxy light tracing for strong lensing modeling. For a small number of systems where the ii-band imaging was of lower quality or did not yield a converged model, alternative bands were adopted: 015310-042315 in gg, 141649+013822 in rr, HSC J1107-0037 in gg, and 120657-010241 in rr. The lens light is modeled with a multi-Gaussian expansion (MGE; Emsellem et al. [45]) comprising two groups of co-centred Gaussians with logarithmically spaced widths and linearly solved intensities. The lens mass is modeled as a singular isothermal ellipsoid (SIE) with external shear, as is common in the literature, and the background source is modeled using the SersicCore light profile in PyAutoLens, which implements a softened Sérsic profile with a finite core radius to improve numerical stability. Since θE\theta_E is expected to be a model-independent lensing quantity Etherington et al. [46], the use of our lens model provides a robust estimate of the Einstein radius for the statistical analysis. Systems are retained only when the lens light subtraction is clean, the reconstructed source is physically plausible, and the lens model converges to a well-defined solution, as assessed from the residual maps and χ2\chi^2 maps presented in Figure 3. The velocity dispersions for these systems are extracted from the DESI DR1 FastSpecFit Spectral Synthesis and Emission-Line Catalogue2, which provides derived spectrophotometric quantities including stellar velocity dispersions obtained from spectral fitting. For these DESI systems, the angular aperture radius is set by the DESI fibre radius, θap=0.75′′\theta_{\mathrm{ap}} = 0.75''. These velocity dispersions and aperture radii provide a homogeneous spectroscopic characterization of the newly modeled sub-sample and are used consistently in compiling the observational quantities listed in Table 1.

In total, 13 systems pass all criteria, with modeling reconstructions, residual maps, and χ2\chi^2 maps presented in Fig. 3. The final sample comprises systems from Chen et al. [33], Alves de Oliveira et al. [5], the modelled systems from Shu & Li [115], galaxy-quasar systems from Guo et al. [58], and additional systems from recent new surveys, yielding a total of 200 systems. By combining legacy compilations with modern wide-field discoveries, our sample covers a lens redshift range of 0.025≤zl≤1.0040.025 \leq z_l \leq1.004 and an Einstein radius range of 0.36′′≤θE≤2.55′′0.36'' \leq\theta_E \leq2.55''. This broad coverage supports a systematic investigation of the total mass-density slope in ETG lenses over a wide redshift baseline. As discussed above, lens systems reached our sample through three discovery channels. Spectroscopically selected lenses, from surveys such as SLACS, BELLS, and BELLS GALLERY (n=135n = 135), are identified through the detection of emission lines from background sources in the spectra of foreground galaxies. Image-selected lenses, from surveys such as SL2S, AGEL, and SuGOHI (n=59n = 59), are identified photometrically through arc or ring signatures in wide-field imaging. Source-triggered systems (n=6n = 6) were first noticed as bright background objects and only later recognized as lensed. The distributions of zlz_l, zsz_s, and θE\theta_E are shown in Figure 1 and Figure 2.

Redshift distribution of the strong lensing sample

Figure 1. Redshift distribution of the strong lensing sample.

Einstein radius distribution of the strong lensing sample

Figure 2. Einstein radius distribution of the strong lensing sample. The histograms show lenses discovered through spectroscopy (n=135n=135), through lens imaging (n=59n=59), and through source-triggered detections (n=6n=6).

Lens modeling results for the 13 strong gravitational lens systems modeled in this work

Figure 3. Lens modeling results for the 13 strong gravitational lens systems modeled in this work. Each row shows, from left to right: the observed HSC image, the reconstructed lensed source model, the source-plane reconstruction, the normalized residual map (in units of σ\sigma), and the χ2\chi^2 map. The system name is labeled in white in the leftmost panel of each row.

Sample Selection Classes

Six rows of HSC images, source model images, source plane images, normalized residual maps, and chi-squared maps

Figure 3 (continued).

HSC image, source model image, source plane image, normalized residual map, and chi-squared map

Figure 3 (continued, PDF p. 10).

Sample Selection Classes

Having assembled the sample from the various discovery channels described above, we now specify how the selection-function parameters introduced in Sect. 2 are assigned to individual systems when computing SBS_B. The assignment follows the spectroscopic survey from which a system obtained its redshift and velocity dispersion, rather than only the way in which it was first noticed, since it is the parent spectroscopic sample that imposes the magnitude window. For systems targeted by a systematic spectroscopic survey, and for systems found in imaging whose spectroscopic quantities come from a cross-match with such a survey, we adopt the target-selection thresholds of that survey as mlimm_{\mathrm{lim}} and mbrightm_{\mathrm{bright}}. For systems followed up with dedicated spectroscopy of individual targets no such window applies, and only the foreground-suppression factor acts. For source-triggered systems the discovery does not depend on the foreground galaxy, so SB≡1S_B \equiv1. The foreground-suppression amplitude is applied uniformly to both the spectroscopically targeted and imaging-discovered categories, with a fiducial value of ε=0.2\varepsilon= 0.2. Since there is no specific external calibration for ε\varepsilon, we later vary it as a stress test of the assumed foreground-suppression strength. For the same reason, msupm_{\mathrm{sup}} is set separately for each category to assess the maximum impact of foreground suppression in that category, as described in Sect. 2. The categories and the adopted magnitude windows are summarized in Table 2.

Sample source and follow-upSelection function categorynnMagnitude selection
SDSS spectroscopic parent sampleSPEC_SDSS (MAIN, LRG, SPECIAL)98rPetro∈[14.5,19.5]r_{\mathrm{Petro}} \in[14.5, 19.5]
BOSS spectroscopic parent sampleSPEC_BOSS (CMASS, LOWZ)45rcmodel∈[16.0,19.6]r_{\mathrm{cmodel}} \in[16.0, 19.6], icmodel∈[17.5,19.9]i_{\mathrm{cmodel}} \in[17.5, 19.9]
DESI spectroscopic parent sampleSPEC_DESI13zfibre<21.60z_{\mathrm{fibre}} < 21.60
Imaging confirmation + spectroscopic follow-upIMAGING_SPEC_FOLLOWUP42None (foreground suppression only)
Source-triggered + spectroscopic follow-upSOURCE_SPEC_FOLLOWUP2None (SB≡1S_B \equiv1)

Table 2. Summary of selection-function categories. The first column lists the discovery path and follow-up confirmation of each system; the second column gives the selection-function category identifier used in this work; third column nn is the number of systems; the fourth column lists the magnitude window used for target selection in each survey for specific band. Here rPetror_{\mathrm{Petro}} is the SDSS Petrosian magnitude used for the MAIN and LRG samples; rcmodelr_{\mathrm{cmodel}} and icmodeli_{\mathrm{cmodel}} are the BOSS cmodel magnitudes used for the LOWZ and CMASS samples; and zfibrez_{\mathrm{fibre}} is the DESI fibre magnitude in the zz band, providing the faint-end limit for LRG targeting. These magnitude definitions belong to different photometric systems and are not homogenized. When computing SBS_B, each magnitude is converted to the corresponding total magnitude system to match the lens apparent magnitude predicted by the Faber–Jackson relation; details of the conversion and adopted values are given in Appendix A.§ 3. Six systems in the sample were first noticed through the background source; the four whose redshifts and velocity dispersions come from an existing spectroscopic library are assigned to the category of that library, so the last row contains only the two for which they come from dedicated observations.

The rows labelled by the SDSS parent sample include both lenses spectroscopically confirmed by SDSS itself and systems discovered in imaging whose velocity dispersions come from a cross-match with SDSS. This category comprises three targeting programmes, labelled MAIN, LRG, and SPECIAL, the last referring to the SDSS ancillary programmes. Although the MAIN (rPetro<17.77r_{\mathrm{Petro}} < 17.77) and LRG (rPetro<19.5r_{\mathrm{Petro}} < 19.5) thresholds differ, a galaxy that fails the MAIN cut can still enter the LRG sample if it satisfies the LRG threshold, and a number of systems in our sample satisfy both criteria. Some targets from the ancillary programmes are fainter than 19.5, the faintest of them exceeding this limit only marginally, with two at 19.8. We adopt for them the unified proxy threshold of the SDSS spectroscopic parent sample, rPetro<19.5r_{\mathrm{Petro}} < 19.5, and label them as SPECIAL for traceability. The bright-end limits quoted in Table 2 are the saturation limits of the corresponding surveys. The BOSS parent-sample rows similarly include lenses spectroscopically confirmed by BOSS and imaging-discovered systems with velocity dispersions from a BOSS cross-match. The BOSS spectroscopic parent sample includes galaxies targeted through either LOWZ or CMASS. We approximate their combined magnitude selection at each trial redshift by the larger of the two window probabilities.

Both the SPEC_SDSS and SPEC_BOSS categories therefore also contain imaging-discovered systems, such as some SuGOHI lenses, whose spectroscopic quantities are obtained by cross-matching. The SPEC_DESI category is dominated by systems discovered in HSC imaging, with lens redshifts and velocity dispersions taken from the DESI spectroscopic library. Since the DESI LRG target selection is expressed as a faint-end limit within a fibre aperture, no bright-end truncation is applied in this category. IMAGING_SPEC_FOLLOWUP consists of systems discovered in imaging and subsequently followed up with dedicated spectroscopy of individual targets, from which their redshifts and velocity dispersions are obtained. With no parent spectroscopic survey imposing a magnitude window, this category is subject to the foreground-suppression factor ε\varepsilon alone. SOURCE_SPEC_FOLLOWUP refers to systems that were first noticed through the background source and whose spectroscopic quantities come from dedicated observations, namely MG2016+112 and SDP.81. Their discovery does not depend on the brightness of the lens galaxy, so SB≡1S_B \equiv1, as discussed in Sect. 2. Systems whose redshifts and velocity dispersions rely instead on a cross-match with an existing spectroscopic library are assigned to the category of that library, whatever the channel through which they were first noticed.

Results and discussion

Constraints on ETG mass profiles

Using the newly compiled lens sample, we derive the posterior probability distributions of the Jeans parameters (γ,δ,β)(\gamma,\delta,\beta) in three redshift intervals, following the methodology described in Section 2. The VDF parameters adopted for the corresponding redshift intervals are summarized in Table 3. At low redshift (0<zl<0.50 < z_l < 0.5), we use the local ETG VDF from Choi et al. [34]. This provides a well-calibrated low-redshift reference for the velocity-dispersion distribution of ETGs. Its use over the low-redshift interval is also motivated by the result of Bezanson et al. [16], who found little evolution in the VDF between z∼0z \sim0 and z∼0.5z \sim0.5. For the intermediate redshift interval (0.5<zl<0.60.5 < z_l < 0.6), we adopt the directly measured VDF from Montero-Dorta et al. [83], based on BOSS spectra of massive red-sequence galaxies at z∼0.55z \sim0.55. For the high-redshift interval (0.6<zl≲1.00.6 < z_l \lesssim1.0), we use the LEGA-C VDF from Taylor et al. [129], which provides a direct spectroscopic calibration of the VDF for quiescent galaxies at z∼0.6−1.0z \sim0.6 - 1.0.

In the fiducial setup we adopt the category-specific magnitude windows of Sect. 3.2. For foreground suppression we set ε=0.2\varepsilon= 0.2, and place msupm_{\mathrm{sup}} at the 10th percentile of the predicted magnitude distribution in each category. Figure 4 shows the resulting posteriors together with a reference distribution built without any lens data in dashed grey curves, obtained by sam-

Posterior distributions of the parameters $\gamma$, $\delta$, and $\beta$

Figure 4. Posterior distributions of the parameters γ\gamma, δ\delta, and β\beta in the three redshift intervals for the fiducial VDF choice. Grey dashed lines mark the reference values (γ,δ,β)=(2,2,0)(\gamma,\delta,\beta) = (2, 2, 0). Grey dashed curves show the reference distribution obtained without any lens data. The shaded and dotted contours indicate the 68% and 95% credible regions, respectively.

pling uniformly inside the prior box (γ∈(1.5,2.5)\gamma\in(1.5, 2.5), δ∈(1,3)\delta\in(1, 3), β∈(−1,1)\beta\in(-1, 1)) and keeping the points where f(γ,δ,β)f(\gamma,\delta,\beta) of Eq. (6) is defined. Since ff is defined over only part of the box, this reference distribution is not totally flat in δ\delta. Across the three redshift intervals, the δ\delta posteriors remain broad and are centered around values close to δ≃2\delta\simeq2, without showing a significant redshift dependence. The orbital anisotropy parameter β\beta is also broadly constrained and remains statistically consistent with isotropy. These parameters are therefore treated mainly as structural degrees of freedom to be marginalized over, while we focus the physical interpretation on the total density slope γ\gamma, which is the primary tracer of the effective mass structure probed by the lens. In the low-redshift bin, which contains 140 lenses, the total density slope is constrained to γ=2.00−0.15+0.07\gamma= 2.00^{+0.07}_{-0.15}, consistent with an isothermal total mass-density profile. In the intermediate-redshift bin (0.5<zl<0.60.5 < z_l < 0.6), the sample contains 29 lenses and the posterior distributions are broader. We obtain γ=1.96−0.26+0.16\gamma= 1.96^{+0.16}_{-0.26}, again consistent with an isothermal profile and with the low-redshift result within the current uncertainties. For the high-redshift bin (0.6<zl≲1.00.6 < z_l \lesssim1.0), which contains 31 lenses, the fiducial analysis gives γ=1.81−0.21+0.17\gamma= 1.81^{+0.17}_{-0.21}. The central value of γ\gamma is lower than those of the low- and intermediate-redshift bins, suggesting a possible flattening of the total density profile at earlier cosmic times. The difference between the low- and high-redshift central values is Δγ∼0.20\Delta\gamma\sim0.20, corresponding to a significant ∼1σ\sim1\sigma when the uncertainties of the two bins are combined. We therefore interpret this result as a suggestive trend rather than a definitive detection of redshift evolution.

The three intervals draw on different discovery channels and VDF calibrations, so the selection function and the VDF do not act on them in the same way. Table 4 summarizes how γ\gamma responds to each factor. Switching on the selection function leaves γ\gamma essentially unchanged, and varying ε\varepsilon from zero to twice its fiducial value shifts γ\gamma by less than 0.03 in every interval. The magnitude window matters more. In the data, lenses become rarer towards high zlz_l, and the model can explain this in two ways. Either the massive lenses needed at large distances are intrinsically rare, which the optical-depth kernel captures and which sets in earlier for larger γ\gamma, or distant lenses are too faint to pass the magnitude limit, which SBS_B captures. A tighter window attributes more of the decline to selection and leaves less for the kernel, so the fitted γ\gamma drops. The shift is largest at low redshift, where spectroscopically selected systems with a faint-end limit dominate. The intermediate-redshift interval is dominated by BOSS systems, whose predicted magnitudes already sit close to the targeting limit, so that the contribution from the high-zlz_l side is removed as soon as the selection function is applied and tightening it further produces little additional shift. Finally, the −1.0-1.0 mag shift is included as a limiting test. It lowers γ\gamma to 1.79 and 1.70 in the low- and high-redshift intervals, indicating that only a magnitude threshold this strict could substantially weaken the negative γ\gamma–zz trend.

Treatment0.0<zl<0.50.0 < z_l < 0.5<br>(n=140n = 140)0.5<zl<0.60.5 < z_l < 0.6<br>(n=29n = 29)0.6<zl≲1.00.6 < z_l \lesssim1.0<br>(n=31n = 31)
Fiducial2.00−0.15+0.072.00^{+0.07}_{-0.15}1.96−0.26+0.161.96^{+0.16}_{-0.26}1.81−0.21+0.171.81^{+0.17}_{-0.21}
Selection function
No SBS_B2.00−0.15+0.072.00^{+0.07}_{-0.15}2.00−0.22+0.162.00^{+0.16}_{-0.22}1.80−0.19+0.151.80^{+0.15}_{-0.19}
Magnitude window −0.25-0.25 mag1.97−0.16+0.081.97^{+0.08}_{-0.16}1.98−0.27+0.171.98^{+0.17}_{-0.27}1.80−0.20+0.191.80^{+0.19}_{-0.20}
Magnitude window −0.5-0.5 mag1.93−0.17+0.121.93^{+0.12}_{-0.17}1.99−0.28+0.191.99^{+0.19}_{-0.28}1.77−0.19+0.211.77^{+0.21}_{-0.19}
Magnitude window −1.0-1.0 mag1.79−0.15+0.171.79^{+0.17}_{-0.15}1.96−0.27+0.241.96^{+0.24}_{-0.27}1.70−0.14+0.231.70^{+0.23}_{-0.14}
ε=0\varepsilon= 02.00−0.14+0.062.00^{+0.06}_{-0.14}1.96−0.26+0.161.96^{+0.16}_{-0.26}1.81−0.20+0.171.81^{+0.17}_{-0.20}
ε=0.4\varepsilon= 0.42.00−0.15+0.072.00^{+0.07}_{-0.15}1.95−0.26+0.161.95^{+0.16}_{-0.26}1.83−0.21+0.161.83^{+0.16}_{-0.21}
Velocity dispersion function
σ∗ref\sigma_{\ast\mathrm{ref}} high2.02−0.15+0.072.02^{+0.07}_{-0.15}1.98−0.29+0.191.98^{+0.19}_{-0.29}1.95−0.28+0.161.95^{+0.16}_{-0.28}
σ∗ref\sigma_{\ast\mathrm{ref}} low1.98−0.14+0.071.98^{+0.07}_{-0.14}1.91−0.23+0.141.91^{+0.14}_{-0.23}1.63−0.09+0.121.63^{+0.12}_{-0.09}
α\alpha high(4.2)——1.84−0.22+0.171.84^{+0.17}_{-0.22}
α\alpha low(1.0)——1.80−0.20+0.171.80^{+0.17}_{-0.20}
β′\beta' high(3.9)——1.67−0.12+0.141.67^{+0.14}_{-0.12}
β′\beta' low(2.0)——2.04−0.30+0.152.04^{+0.15}_{-0.30}
VDF marginalized——1.96−0.31+0.221.96^{+0.22}_{-0.31}

Table 4. Response of the total density slope γ\gamma to the analysis settings. The first row is the fiducial analysis. The magnitude-window rows shift all magnitude thresholds, including msupm_{\mathrm{sup}}, towards brighter values by the amount indicated, and the foreground-suppression rows vary ε\varepsilon away from its fiducial value of 0.2. The VDF rows vary one Schechter parameter at a time within its quoted uncertainty in the interval concerned, keeping the others fixed; the last row samples σ∗ref\sigma_{\ast\mathrm{ref}}, α\alpha and β′\beta' jointly with (γ,δ,β)(\gamma,\delta,\beta) using split-normal priors centred on the LEGA-C measurements of Taylor et al. [129]. Dashes mark tests not carried out in that interval.

The near-isothermal slopes recovered in the low- and intermediate-redshift bins are consistent with the long-standing result that massive ETG lenses have approximately isothermal total mass-density profiles over the radial range probed by strong lensing. This behavior is often interpreted as the outcome of the coupled evolution of baryons and dark matter: centrally concentrated stellar components, halo response to baryonic assembly, feedback-driven redistribution of gas, and subsequent merging can combine to produce an approximately scale-free total density profile even when neither the stellar nor dark-matter component is individually isothermal. In this sense, the agreement of the low- and intermediate-redshift bins with γ≃2\gamma\simeq2 supports the view that the lensing probes a radial range where the stellar and dark-matter contributions are closely balanced in massive ETGs. The lower central value obtained in the high-redshift bin may indicate that massive ETGs at earlier epochs occupy a somewhat different structural state from their low-redshift counterparts. A shallower effective total slope at zl≳0.6z_l \gtrsim0.6 would be qualitatively consistent with a picture in which the stellar mass distribution, dark-matter fraction within the lensing aperture, and halo response to baryonic processes evolve with time. Hydrodynamical simulations, however, do not yet agree on the direction of this evolution. In Horizon-AGN, AGN feedback produces shallower total slopes at earlier epochs [96, 97], qualitatively consistent with our high-redshift result. In IllustrisTNG, AGN feedback makes the slopes shallower between z∼2z \sim2 and z∼1z \sim1, after which they change little [137]. In Magneticum, the intrinsic slopes are steeper at higher redshift, yet the observed trend is recovered once the simulated galaxies are analysed with the observational method ([102]), which shows that such comparisons depend on how the effective slope is measured. In our work, given the current uncertainties, the negative γ−z\gamma- z trend interpretation should be regarded as qualitative. The high-redshift bin is particularly informative because it probes an epoch where both the lens population and the VDF

are less well sampled than at lower redshift. As shown in Table 3, the LEGA-C VDF has relatively large uncertainties. Moreover, as demonstrated by Taylor et al. [129], it exhibits substantial internal shape variations across 0.6<zl<1.00.6 < z_l < 1.0. We therefore examine the sensitivity of the high-redshift result to the adopted VDF normalization and shape in the following subsection.

Stability of the inferred density slopes

The fiducial constraints discussed above rely on the adopted VDF parameters and on the constructed lens sample. We therefore perform a set of diagnostic tests to examine how the inferred density slopes respond to a range of methodological and data-related variations. These tests are designed to separate several effects that can enter a statistical lensing analysis: the VDF calibration that sets the optical-depth kernel, the construction of the lens sample, and differences between the lens populations reached by different discovery channels.

We first examine the sensitivity of the inferred slopes to the adopted VDF calibration. In the low-redshift bin, varying σ∗ref\sigma_{\ast\mathrm{ref}} within its quoted 1σ1\sigma uncertainty changes γ\gamma by only ≃0.04\simeq0.04 across the full range. The effect is larger in the intermediate-redshift bin, with changes Δγ<0.1\Delta\gamma< 0.1, but remains within the statistical uncertainty of the fiducial constraint. In the high-redshift bin, the broader uncertainties of the available VDF calibration lead to a larger propagated uncertainty in the inferred density slope. We therefore perform a dedicated set of high-redshift VDF tests, summarized in Table 4. These tests show that changing the low-velocity slope α\alpha has little effect on γhigh−z\gamma_{\mathrm{high-z}}, whereas the characteristic velocity scale σ∗ref\sigma_{\ast\mathrm{ref}} and the high-velocity cutoff parameter β′\beta' have a more significant impact. This behavior is expected because the strong lensing optical depth is weighted toward high-velocity-dispersion galaxies, and the lensing cross section scales approximately as a high power of velocity dispersion. The high-velocity tail of the VDF is therefore a key ingredient in interpreting the high-redshift density slope. As an exploratory test of the combined VDF uncertainty, we also perform a VDF-marginalized run in which σ∗ref\sigma_{\ast\mathrm{ref}}, α\alpha, and β′\beta' are sampled jointly with (γ,δ,β)(\gamma,\delta,\beta) using independent split-normal priors based on the LEGA-C measurements. This gives γhigh−z=1.96−0.31+0.22\gamma_{\mathrm{high-z}} = 1.96^{+0.22}_{-0.31}, with uncertainties about 1.5 times larger than in the fixed-VDF case, and the selection function plays a visible role in this test. Without SBS_B, the same run gives γhigh−z=1.67−0.12+0.22\gamma_{\mathrm{high-z}} = 1.67^{+0.22}_{-0.12}, in which case the decline of lenses towards high zlz_l can then only be attributed to the VDF, which drives γ\gamma towards lower values. With SBS_B, part of this decline is attributed to selection and γ\gamma rises accordingly. This demonstrates the degeneracy between γ\gamma and the high-velocity tail of the VDF, which dominates the systematic uncertainty of the high-redshift slope. Thus, a larger high-redshift lens sample would allow the lens analysis to be carried out in narrower redshift intervals, closer to the Δz=0.1\Delta z = 0.1 VDF measurements of Taylor et al. [129], while maintaining sufficient statistical precision in each bin.

As a complementary diagnostic, we test a simple evolving-VDF prescription in the largest subsample, the low-redshift bin. We adopt

n∗(z)=n∗,0(1+z)P,σ∗(z)=σ∗,0(1+z)U(27)n_*(z) = n_{*,0}(1+z)^P,\qquad\sigma_*(z) = \sigma_{*,0}(1+z)^U \tag*{(27)}

with (P,U)=(−1,0.2)(P,U) = (-1,0.2). The density-evolution term P=−1P = -1 represents a mild decline in the comoving number density of ETGs with increasing redshift, consistent with the evolutionary trends reported by Faber et al. [49] and Brown et al. [20]. The velocity-dispersion evolution term is motivated by studies suggesting that the velocity dispersions of early-type galaxies increase with redshift, for example σ∝(1+z)0.19\sigma\propto(1+z)^{0.19} at fixed stellar mass and σ∝(1+z)0.41\sigma\propto(1+z)^{0.41} at fixed dynamical mass [104]. Since other statistical studies have reported weak or insignificant evolution in σ∗\sigma_*, we adopt U=0.2U = 0.2 as a moderate positive evolution model rather than as a unique description of the VDF evolution. In our conditional likelihood, the absolute normalization of the VDF largely cancels between the numerator and the redshift-normalization integral. This test therefore mainly probes the impact of a mild redshift evolution in the characteristic velocity dispersion σ∗ref\sigma_{*\mathrm{ref}}. For the low-redshift bin, this evolving-σ∗ref\sigma_{*\mathrm{ref}} prescription gives γ=2.06−0.12+0.07\gamma= 2.06^{+0.07}_{-0.12}, compared with the fiducial value γ=2.00−0.15+0.07\gamma= 2.00^{+0.07}_{-0.15}. The shift is therefore Δγ≃0.06\Delta\gamma\simeq0.06, comparable to the upper 1σ1\sigma uncertainty of the fiducial constraint. Although VDF evolution represents a non-negligible systematic effect for precision measurements of γ(z)\gamma(z), the qualitative conclusion that the low-redshift population is close to isothermal remains unchanged. For a more rigorous treatment, a full joint model of VDF and density-slope evolution should be pursued in future work.

We then examine the impact of sample construction. Repeating the analysis with only the original 160-lens sample gives γ=1.99−0.15+0.07\gamma= 1.99^{+0.07}_{-0.15}, 1.94−0.27+0.191.94^{+0.19}_{-0.27}, and 1.91−0.22+0.121.91^{+0.12}_{-0.22} in the low-, intermediate- and high-redshift bins, respectively. Thus, the original sample already favors near-isothermal slopes at low and intermediate redshifts and a lower central value at high redshift. The expanded 200-lens sample mainly affects the high-redshift interval, where the number of lenses increases from 18 to 31, shifting the high-redshift central value from γ≃1.91\gamma\simeq1.91 to γ≃1.81\gamma\simeq1.81 and thereby excluding the isothermal γ=2\gamma= 2 at the 1σ1\sigma level. This indicates that the newly added systems increase the leverage on the high-redshift population and strengthen the apparent high-redshift flattening, while leaving the low- and intermediate-redshift results essentially unchanged. As a complementary spectroscopic quality check, we remove systems with fractional velocity-dispersion uncertainties larger than 20%. This cut removes systems with less precise spectroscopic characterization and tests whether the inferred slopes are driven by this subset of objects. This reduces the sample from 200 to 184 lenses. The inferred slopes remain unchanged within uncertainty in the three redshift bins, indicating that the main constraints are not driven by the subset with the poorly constrained velocity dispersions.

Another important inference concerns whether lenses discovered through different strategies share a common population-level density slope. Different lens-search strategies naturally favor different regions of lens and source parameter space. Spectroscopic searches, such as those used in SLACS-like samples, preferentially identify foreground galaxies with strong absorption features and detectable background emission lines within a fixed fibre aperture, and therefore tend to select massive, relatively low-redshift lens galaxies with favorable fibre-source alignment. Image-selected searches, in contrast, are more directly sensitive to resolved arcs and multiple images, and can extend more efficiently to larger Einstein radii, higher source redshifts and lens redshifts, depending on imaging depth, angular resolution, and lens-galaxy subtraction. Our selection model accounts for the foreground-brightness dependence of each strategy, while the two strategies also reach lens galaxies of different mass and environment. Any dependence of the density slope on these properties could therefore appear as a difference in the inferred population-level γ\gamma between the two subsamples. As shown in Figure 5, the image-selected and spectroscopically selected subsamples populate the redshift and space differently. In the low-redshift bin, the spectroscopically selected subsample gives γ=2.01−0.13+0.06\gamma= 2.01^{+0.06}_{-0.13}, while the image-selected subsample gives γ=1.85−0.24+0.20\gamma= 1.85^{+0.20}_{-0.24}. The difference in central value is within 1σ1\sigma, and the image-selected low-redshift subsample is substantially smaller and has a broader posterior. In the intermediate-redshift bin, both discovery channels are consistent with an isothermal slope within their larger uncertainties. In the high-redshift bin, the image-selected sample gives γ=1.86−0.22+0.14\gamma= 1.86^{+0.14}_{-0.22}, while the spectroscopically selected subsample gives γ=1.92−0.24+0.13\gamma= 1.92^{+0.13}_{-0.24}. The latter result is reported for completeness because the spectroscopically selected high-redshift subsample contains only six systems. The high-redshift constraint is therefore dominated in sample size by the image-selected population, but both currently available spectroscopically selected and image-selected systems indicate a similar slope in the high-redshift range.

Density-slope constraints for image-selected and spectroscopically selected lenses

Figure 5. Density-slope constraints for image-selected and spectroscopically selected lenses in the three redshift intervals. The comparison tests whether the lens populations reached by the two discovery channels share a common population-level density slope. The sample sizes for each subsample are indicated at the top. The dashed line and shaded region mark the isothermal reference value γ=2\gamma= 2. Error bars denote the 16th-84th percentile credible intervals.

Taken together, the tests presented in this section demonstrate that the qualitative picture, a near-isothermal low-redshift population and a shallower high-redshift slope, is stable across a range of data-quality cuts, discovery channels, and modeling choices. The quantitative depth of the high-redshift flattening, however, remains sensitive to the assumed high-redshift VDF, particularly its high-velocity tail, and to the size and composition of the high-redshift lens sample, underscoring the need for both improved VDF measurements and larger high-redshift lens samples in future work.

Implications for ETG density-profile evolution

The redshift evolution of the total mass-density slope of massive early-type galaxies has been investigated using a variety of lensing, dynamical, and joint lensing-dynamics approaches. Dynamical studies of nearby ETGs often infer slightly steeper inner total slopes, with typical values around γ≃2.2\gamma\simeq2.2 [29, 98] (Li et al. 2019), whereas strong lensing and joint lensing-dynamics studies of massive lenses commonly find slopes close to isothermal over the radial range probed by the Einstein radius. The reported normalization and possible redshift dependence of the slope, however, vary among different studies. This diversity is expected because different analyses Notes. For Tan et al. [128], the quoted anisotropy ⟨β⟩\langle\beta\rangle is derived from a single joint fit to the full SLACS+SL2S+BELLS sample of 77 lenses, and is applied to both subsamples listed. For the two Mozumdar et al. [86] entries, the ratio σθ/σr\sigma_\theta/\sigma_r is assumed to be constant within each galaxy and is fitted independently for each galaxy; the quoted U[R(q),2]\mathcal{U}[R(q), 2] and N(1.0,0.07)\mathcal{N}(1.0, 0.07) are the priors on this constant. Here qq is the intrinsic axial ratio and R(q)=0.3+0.7qR(q) = \sqrt{0.3 + 0.7q}. Superscript labels (e.g., 1^{1}, 2^{2}) denote distinct analysis configurations reported within the same work, for example measurements obtained from different subsamples or from alternative modeling choices.

probe different samples, radial ranges, selection functions, and modelling assumptions. Table 5 and Figure 6 place our fiducial constraints in the context of previous measurements. The low- and intermediate-redshift results from this work, γ=2.00−0.15+0.07\gamma= 2.00^{+0.07}_{-0.15} and γ=1.96−0.26+0.16\gamma= 1.96^{+0.16}_{-0.26}, are consistent with the slopes reported at similar redshifts by many studies of massive ETG lenses. The high-redshift result from this work, γ=1.81−0.21+0.17\gamma= 1.81^{+0.17}_{-0.21}, lies on the shallower side of the literature distribution, extending the statistical comparison into a redshift range where existing lens constraints remain sparse. The direction of this negative trend in γ\gamma with redshift is consistent with a recent dynamics-only analysis [86], which used spatially resolved MUSE-DEEP stellar kinematics together with a local MaNGA comparison sample and reported ⟨γ⟩=2.26±0.01\langle\gamma\rangle= 2.26 \pm0.01 at ⟨z⟩=0.04\langle z\rangle= 0.04 and ⟨γ⟩=2.19±0.01\langle\gamma\rangle= 2.19 \pm0.01 at ⟨z⟩=0.44\langle z\rangle= 0.44, corresponding to dγ/dz≃−0.20±0.03\mathrm{d}\gamma/\mathrm{d}z \simeq-0.20 \pm0.03. The higher absolute normalization of γ\gamma relative to the slopes inferred here and in many lensing studies likely reflects differences in galaxy samples, radial ranges, and modelling assumptions. At z≲0.55z \lesssim0.55, Derkenne et al. [40] found no significant evolution of the dynamically inferred slope (γ=2.11±0.03\gamma= 2.11 \pm0.03 for 90 ETGs at 0.29<z<0.550.29 < z < 0.55), consistent with the near-constant slopes of our low- and intermediate-redshift bins; the decline in our results appears mainly in the 0.6<zl≲1.00.6 < z_l \lesssim1.0 interval, beyond the redshift range covered by current dynamical samples. Together, these comparisons highlight the importance of method-dependent slope definitions and radial sensitivity.

Comparison of total mass-density slopes for different samples

Figure 6. Comparison of total mass-density slopes ⟨γ⟩\langle\gamma\rangle for different samples. Blue, orange, and green symbols represent joint lensing-dynamics, dynamics-only, and lensing-only studies, respectively. Black points show the three redshift subsamples from this work. Error bars indicate 1σ1\sigma uncertainties. To highlight differences in the underlying lens samples, the marker fill style encodes the lens-selection method: open symbols denote studies based on spectroscopically selected lenses (e.g. SLACS, BELLS, BELLS GALLERY), filled symbols denote studies based on image-selected lenses (e.g. SL2S, AGEL), and half-filled symbols denote studies that combine both selection types within a single sample (e.g. SLACS+SL2S+BELLS). Dynamics-only studies do not rely on imaging and are plotted with open symbols by default. Note that several literature measurements share nearly identical median redshifts, a small horizontal jitter of ±0.005\pm0.005 in redshift was applied to avoid overlapping.

The scatter among literature measurements also reflects differences in the radial range probed and the observables used by each method, as summarized in Table 5. Our analysis connects the Einstein radius to a reference stellar velocity dispersion through a Jeans-based mapping, with the kinematic information entering statistically through the adopted VDF. Unlike studies that fit a slope to each lens, however, we infer a single population-level slope from the distribution of lens redshifts conditioned on the observed Einstein radii and source redshifts. Lensing-only analyses constrain the projected mass distribution within the Einstein radius, but the inferred density slope depends on the adopted mass model and priors and is subject to the mass-sheet degeneracy. Dynamics-only studies, in contrast, measure luminosity-weighted velocity moments of normal galaxies, typically out to the effective radius or beyond, so that the inferred slopes depend on the stellar orbital structure, the adopted light model, and the radial range covered by the kinematic data.

StudySample SizeRedshift (median)⟨γ⟩\langle\gamma\rangleLight ProfileAnisotropy
This work140∼0.23\sim0.232.00−0.15+0.072.00^{+0.07}_{-0.15}parametric power-law (δ\delta free)⟨β⟩=−0.27−0.43+0.64\langle\beta\rangle= -0.27^{+0.64}_{-0.43}
This work29∼0.55\sim0.551.96−0.26+0.161.96^{+0.16}_{-0.26}parametric power-law (δ\delta free)⟨β⟩=−0.32−0.42+0.62\langle\beta\rangle= -0.32^{+0.62}_{-0.42}
This work31∼0.73\sim0.731.81−0.21+0.171.81^{+0.17}_{-0.21}parametric power-law (δ\delta free)⟨β⟩=−0.16−0.58+0.70\langle\beta\rangle= -0.16^{+0.70}_{-0.58}
Combined lensing and dynamics studies
Tan et al. [128]1^{1}24∼0.5\sim0.52.08±0.062.08 \pm0.06double Sérsic profile⟨β⟩=0.53−0.45+0.27\langle\beta\rangle= 0.53^{+0.27}_{-0.45}
Tan et al. [128]2^{2}34∼0.2\sim0.22.04±0.042.04 \pm0.04double Sérsic profile⟨β⟩=0.53−0.45+0.27\langle\beta\rangle= 0.53^{+0.27}_{-0.45}
Sheu et al. (2025)56∼0.35\sim0.352.24±0.142.24 \pm0.14single Sérsic profile⟨β⟩=−0.20−0.19+0.21\langle\beta\rangle= -0.20^{+0.21}_{-0.19}
Cao et al. [23]1^{1}118∼0.29\sim0.292.035±0.0132.035 \pm0.013parametric power-law (δ\delta free)β\beta prior: 0.18±0.130.18 \pm0.13
Cao et al. [23]2^{2}118∼0.29\sim0.292.132±0.0552.132 \pm0.055γ=δ\gamma= \deltaβ=0\beta= 0
Sahu et al. [105]7∼0.57\sim0.571.91±0.081.91 \pm0.08double Sérsic profileβ=0\beta= 0
Auger et al. [10]73∼0.2\sim0.22.078±0.0272.078 \pm0.027de Vaucouleurs profileβ=0\beta= 0
Koopmans et al. [71]58∼0.2\sim0.22.085−0.018+0.0252.085^{+0.025}_{-0.018}Hernquistβ=0\beta= 0
Koopmans et al. [72]15∼0.2\sim0.22.01−0.03+0.022.01^{+0.02}_{-0.03}Hernquistβ=0\beta= 0
Sonnenfeld [117]59∼0.19\sim0.191.99±0.031.99 \pm0.03no explicit light fit (ReR_{\mathrm{e}}-weighted)β=0\beta= 0
Barnabè et al. [11]16∼0.16\sim0.162.074−0.041+0.0432.074^{+0.043}_{-0.041}non-parametric HST lightβ=0\beta= 0
Lyskova et al. [80]15∼0.16\sim0.162.1±0.052.1 \pm0.05Sérsic/de Vaucouleursβ\beta-insensitive radius (no β\beta fit)
Etherington et al. [47]1^{1}123∼0.3\sim0.32.024−0.021+0.0192.024^{+0.019}_{-0.021}de Vaucouleurs profileβ\beta from source L&D research
Dynamics-only studies
Mozumdar et al. [86]1^{1}277∼0.044\sim0.0442.260±0.0132.260 \pm0.013MGE light modelσθ/σr∼U[R(q),2]\sigma_\theta/\sigma_r \sim\mathcal{U}[R(q),2]
Mozumdar et al. [86]2^{2}163∼0.44\sim0.442.190±0.0132.190 \pm0.013MGE light modelσθ/σr∼N(1.0,0.07)\sigma_\theta/\sigma_r \sim\mathcal{N}(1.0,0.07)
Derkenne et al. [39]1^{1}64∼0.35\sim0.352.01±0.042.01 \pm0.04MGE light modelβ\beta free (constant per galaxy)
Derkenne et al. [39]2^{2}28∼0.3\sim0.32.23±0.042.23 \pm0.04MGE light modelβ\beta free (constant per galaxy)
Derkenne et al. [39]3^{3}150∼0\sim02.25±0.022.25 \pm0.02MGE light modelβ\beta free (constant per galaxy)
Bellstedt et al. [12]21∼0\sim02.24±0.052.24 \pm0.05MGE light modelβ\beta free (constant per galaxy)
Poci et al. [98]258∼0\sim02.193±0.0162.193 \pm0.016MGE light + (M/L)⋆_\star from pPXFβ\beta free (constant per galaxy)
Lensing-only studies
Sahu et al. [105]2^{2}7∼0.57\sim0.571.95±0.091.95 \pm0.09N/AN/A
Etherington et al. [47]2^{2}57∼0.3\sim0.32.095−0.029+0.0302.095^{+0.030}_{-0.029}N/AN/A

Table 5. Total mass-density slope constraints from previous studies and this work

Such samples are free of lens-finding biases, but are selected through the surface-brightness and signal-to-noise requirements of spatially resolved spectroscopy. Their higher values of γ\gamma may partly reflect differences in radial coverage and slope definition. Comparisons of redshift trends also require accounting for changes in sample properties, selection, and spatial resolution with redshift. Joint lensing-dynamics analyses combine these complementary constraints and partly break their respective degeneracies. They provide a useful comparison for our results, although differences in data quality, spatial coverage, and the treatment of the stellar light distribution, aperture corrections, orbital anisotropy, and symmetry assumptions should be considered when comparing the inferred slopes.

In our framework, the stellar luminosity density is described by an effective power-law slope, ν(r)∝r−δ\nu(r) \propto r^{-\delta}. This parametrization is commonly used in analytical Jeans-based lensing calculations (e.g., [72, 111, 23, 56]). While individual lens light profiles are observable, a single power law yields only one effective value: for Sérsic profiles with 2≲n≲62 \lesssim n \lesssim6, the logarithmic slope of the deprojected luminosity density increases from ∼2.0\sim2.0–2.62.6 at 0.5Re0.5R_{\mathrm{e}} to ∼2.6\sim2.6–2.82.8 at ReR_{\mathrm{e}}. We therefore treat δ\delta as a nuisance parameter of the Jeans mapping and marginalize over 1≤δ≤31 \le\delta\le3, which covers this range. Orbital anisotropy is another important source of methodological variation among published studies. Many joint lensing-dynamics analyses adopt simplified treatments of the anisotropy, often fixing β=0\beta= 0 or imposing an informative prior. In our fiducial analysis, β\beta is allowed to vary and is marginalized over together with γ\gamma and δ\delta. The resulting β\beta posteriors are broad and statistically consistent with isotropy in all three redshift intervals. As a check against the commonly adopted isotropic assumption, we also repeated the analysis with β=0\beta= 0. This changes the inferred density slopes by only Δγ≲0.05\Delta\gamma\lesssim0.05 relative to the fiducial case in all redshift bins and does not alter the redshift ordering of the results. Thus, within the current statistical precision, the qualitative comparison with previous studies is not driven by whether the orbital anisotropy is fixed to zero or marginalized over. The importance of anisotropy and symmetry assumptions has also been emphasized in more detailed dynamical studies. For example, Barnabè et al. [11] used axisymmetric, two-integral dynamical models of 16 SLACS lenses and obtained density-slope estimates systematically lower by ∼0.05\sim0.05 than those inferred by Auger et al. [10] using a spherical and isotropic Jeans model. As discussed by Sonnenfeld [118], this difference is comparable to or larger than the statistical uncertainty on the mean SLACS slope, illustrating the relevance of dynamical modelling assumptions. More recently, Tan et al. [128] compared different anisotropy prescriptions, including spatially constant and Osipkov–Merritt forms, and showed that the inferred consistency with a single power-law total profile can depend on the assumed β(r)\beta(r) model. Simulations further suggest that β\beta may evolve with galaxy mass and redshift, with more tangentially biased orbits at higher redshift and more radially biased orbits at lower redshift (Xu et al. 2017). Our β\beta posteriors are too broad to test this directly, and future spatially resolved kinematics will be needed to constrain anisotropy and density-slope evolution simultaneously. Physically, the near-isothermal slopes found at low and intermediate redshift are commonly interpreted as a manifestation of the bulge-halo conspiracy. The stellar component of massive ETGs is centrally concentrated, whereas the dark-matter halo is intrinsically shallower over the same radial range. Their combined contribution can therefore produce a nearly power-law total density profile with γ≃2\gamma\simeq2 around the Einstein radius, even though neither component is individually isothermal. This balance can be shaped by baryonic contraction, stellar mass assembly, feedback-regulated gas removal, and subsequent merging. The lower central value of γ\gamma in the high-redshift bin may point to a different balance between stellar and dark-matter contributions at earlier epochs. A shallower effective total slope can arise if the stellar component is less centrally dominant within the lensing aperture, if the dark-matter fraction within the Einstein radius is higher, or if the halo response to baryonic processes differs from that at later times. In galaxy-formation models, the outcome of these processes depends strongly on how stellar and AGN feedback are implemented, and strong-lens slopes are sensitive to this choice. Mukherjee et al. [87] showed that only some of the feedback models in the EAGLE simulations reproduced the distribution of total slopes at the Einstein radius observed for SLACS and BELLS lenses. The redshift evolution of the slope offers a further test. Several cosmological hydrodynamical simulations yield nearly isothermal slopes for massive ETGs at low redshift, but differ in how these slopes evolve towards z∼1z \sim1. In Horizon-AGN, the total density slope steepens from z=2z = 2 to z=0z = 0, and AGN feedback is needed to bring the slopes close to the observed values ([96, 97]). In IllustrisTNG, the ETG population selected at each redshift shows nearly constant slopes below z≈1z \approx1 with a mild steepening towards z=0z = 0 ([137]). In Magneticum and Illustris, by contrast, the slopes are steeper at higher redshift and become shallower towards z=0z = 0 ([102] (Xu et al. 2017)), which Remus et al. [102] attribute to the dominance of gas-poor mergers at late times. Since these predictions diverge increasingly with redshift, our high-redshift bin probes the range in which the models are best distinguished. The shallower slope obtained at 0.6<z<1.00.6 < z < 1.0 follows the direction predicted by Horizon-AGN and has the same sign as the redshift dependence found for SL2S lenses (Sonnenfeld et al. [123]). Such a steepening with cosmic time supports a modest contribution from dissipative processes, since purely dry mergers make the density profile shallower (Sonnenfeld et al. [122]). Improved calibration of the high-velocity tail of the VDF at z≳0.6z \gtrsim0.6 and larger high-redshift lens samples will be needed to determine the amplitude of this evolution and to distinguish among these formation scenarios.

Conclusion

In this work, we compiled a sample of 200 galaxy-scale strong lensing systems and applied a statistical strong lensing framework to investigate the effective mass structure of early-type galaxy (ETG) lenses. By combining the lens-redshift distribution with VDF calibrations and a spherical Jeans-based mapping between the Einstein radius and the velocity-dispersion scale, we constrained the total mass-density slope γ\gamma in three redshift intervals, marginalizing over the luminosity-density slope δ\delta and the orbital anisotropy parameter β\beta. Under the fiducial VDF assumptions, the low- and intermediate-redshift bins yield total density slopes close to the isothermal value, with γ=2.00−0.15+0.07\gamma= 2.00^{+0.07}_{-0.15} at 0<zl<0.50 < z_l < 0.5 and γ=1.96−0.26+0.16\gamma= 1.96^{+0.16}_{-0.26} at 0.5<zl<0.60.5 < z_l < 0.6. For the high-redshift bin, 0.6<zl≤1.00.6 < z_l \leq1.0, we obtain γ=1.81−0.21+0.17\gamma= 1.81^{+0.17}_{-0.21}, which excludes the isothermal value at the 1σ1\sigma level. The lower high-redshift value suggests a flattening of the effective total density profile at earlier cosmic epochs. This trend is strengthened by the expanded sample: relative to the original 160-lens sample, the number of lenses at 0.6<zl≤1.00.6 < z_l \leq1.0 increases from 18 to 31, and the high-redshift slope deviates from isothermal.

Our sample is compiled from surveys with different selection criteria. We therefore incorporated a selection function into the lens-redshift likelihood, modelled separately for each system class (Sect. 3.2). As shown in Eq. (21), our likelihood is conditioned on the observed Einstein radius and source redshift, so the inference depends only on SB(zl)S_B(z_l). Including SBS_B has little effect on the fiducial slopes, and the most influential factor is the magnitude window: tightening the thresholds affects γ\gamma differently in the three redshift intervals, and only the −1.0-1.0 mag stress test substantially weakens the redshift trend (Table 4). The role of SBS_B is most evident when the VDF parameters are marginalized over: it then raises the high-redshift slope from γ≃1.67−0.12+0.22\gamma\simeq1.67^{+0.22}_{-0.12} to 1.96−0.31+0.221.96^{+0.22}_{-0.31}, since part of the decline in lens numbers towards high zlz_l is attributed to selection rather than to the VDF.

The inferred slopes remain stable in most of our diagnostic tests (Sect. 4.2). Fixing β=0\beta= 0 changes the inferred density slopes by only Δγ≲0.05\Delta\gamma\lesssim0.05 and does not alter the qualitative redshift ordering. The results are also stable against removing systems with large fractional velocity-dispersion uncertainties, and spectroscopically and image-selected lenses yield consistent slopes in all three intervals (Figure 5). The dominant systematic uncertainty arises from the VDF, especially at high redshift. In particular, the high-redshift result is sensitive to the characteristic velocity dispersion and the high-velocity cutoff of the LEGA-C VDF, which shift the slope by amounts comparable to its statistical uncertainty. Thus, the present analysis identifies high-redshift VDF calibration as a key requirement for turning the trend found here into a more precise measurement of ETG density-profile evolution.

The near-isothermal slopes recovered at low and intermediate redshift are consistent with the established picture that massive ETG lenses have approximately isothermal total mass-density profiles over the radial range probed by strong lensing. The lower central value at high redshift follows the direction of the evolution found for SL2S lenses Sonnenfeld et al. [123], the dynamics-only results based on MUSE-DEEP and MaNGA ETGs Mozumdar et al. [86], and the prediction of Horizon-AGN Peirani et al. [97]. In a two-component picture, a shallower effective total slope could reflect a smaller relative stellar contribution within the Einstein radius at earlier epochs. Subsequent central mass growth would then move the profile toward the near-isothermal structure seen at lower redshift. Gas-poor mergers and dissipative growth can affect this balance differently: the former tend to add stars at larger radii, whereas the latter can increase the central stellar concentration. A modest contribution from dissipative growth below z∼1z \sim1 is therefore compatible with the trend found here (Sect. 4.3).

The number of known galaxy-scale strong lenses is now increasing rapidly. The Euclid Quick Data Release alone yielded about 500 lens candidates over 63 deg2^{2}, and the first Euclid data release is expected to provide several thousand high-confidence candidates Euclid Collaboration et al. [48]. Ground-based searches in the DESI Legacy Imaging Surveys and UNIONS have identified large numbers of further candidates Storfer et al. [124, 125], and Euclid, the Rubin LSST, which began its ten-year survey in 2026, and the recently launched Roman Space Telescope are together expected to increase the number of known lenses to ≳105\gtrsim10^{5} Collett [35]; ); Wedig et al. [138]. Spectroscopic programmes such as DESI and 4MOST will supply lens and source redshifts for large numbers of these systems Huang et al. [62]; ); Collett et al. [36]. In addition, CSST is expected to detect approximately 10510^{5} galaxy-galaxy strong lensing systems in simulations Cao et al. [26]. Its optical multi-band imaging will be complementary to Euclid, LSST, and Roman for lens identification and photometric characterization, extending strong-lens studies toward broader ranges of lens mass, image separation, and redshift. Compared with ground-based surveys, the space-based angular resolution of CSST is particularly valuable for resolving smaller-separation galaxy-galaxy lenses, while its ugriz photometry can improve the characterization of both lens and source galaxies. CSST slitless spectroscopy can provide useful redshift and emission-line information over a wide field, while precise stellar velocity dispersions for detailed dynamical analyses will require higher-resolution spectroscopy or cross-matches with surveys such as SDSS, DESI, and future spectroscopic facilities. The simulated CSST strong lensing catalogue of Cao et al. [26] has already been used in several follow-up studies, demonstrating its value as a testbed for future lens searches and modeling pipelines. For example, Li et al. (2024) applied the mock CSST images to train a machine-learning lens finder, while Cao et al. (2025) used them to benchmark a fast lens-modeling algorithm for large survey samples. In the context of the present work, we focus on the high-quality ”golden” subset with signal-to-noise ratios exceeding 100 in stacked observations, corresponding to roughly 4,000 systems in the CSST mock catalogue Cao et al. [26]. A simplified forecast based on this subset suggests that the typical statistical uncertainty in γ\gamma could be reduced by a factor of order ∼5\sim5 relative to current samples. More importantly, such samples would enable narrower redshift binning, improved tests of discovery-

channel dependence, and better calibration of the high-redshift VDF.

The statistical approach developed in our paper is suited to future samples. Applying the method to new surveys will require a characterization of their lens-redshift-dependent selection, including the magnitude limits and other foreground-dependent effects relevant to each discovery and follow-up strategy. Lenses from different programmes can then be analysed within the same framework. This approach provides a promising route toward population-level measurements of ETG mass-structure evolution across cosmic time.

Acknowledgements. We thank the anonymous referee for constructive comments that improved this paper. This work is supported by Beijing Natural Science Foundation No. 1242021; the National Natural Science Foundation of China (Nos. 12021003, 12203009, 12433001); and the Strategic Priority Research Program of the Chinese Academy of Sciences, Grant No. XDB23000000. We acknowledge the use of PyAutoLens Nightingale et al. [90]; He et al. (2024), emcee Foreman-Mackey et al. [54], and AstroPy Astropy Collaboration et al. [8, 7, 6]. This research used data from DESI Data Release 1 DESI Collaboration et al. [41] and its FastSpecFit value-added catalogue Moustakas et al. [85], and images from the third public data release of the HSC Subaru Strategic Program Aihara et al. [2]. This work also made use of data from SDSS and SDSS-III.

BandM∗M_* (mag)V∗V_* (dex)σM\sigma_M (mag)σV\sigma_V (dex)ρVM\rho_{VM}AA (mag dex−1^{-1})ww (mag)
rr−21.15-21.152.2002.2000.8410.8410.1110.111−0.774-0.774−5.864-5.8640.5330.533
ii−21.49-21.492.2012.2010.8510.8510.1100.110−0.781-0.781−6.042-6.0420.5310.531
zz−21.83-21.832.2002.2000.8450.8450.1100.110−0.782-0.782−6.007-6.0070.5270.527

Table A.1. Parameters of the conditional relation between velocity dispersion and absolute magnitude. M∗M_*, V∗V_*, σM\sigma_M, σV\sigma_V, and ρVM\rho_{VM} are taken from Table 1 of Bernardi et al. [14], and AA and ww are computed from them using Eqs. (A.1) and (A.2).

Calculation of the selection function SBS_B

This appendix describes how the selection function SBS_B of Eq. (24) is evaluated. For each system and each trial lens redshift zlz_l, the reference velocity dispersion σref\sigma_{\rm ref} required to reproduce the observed Einstein radius (Eq. (16)) sets the expected absolute magnitude of the lens galaxy through the Faber-Jackson relation. Once distance, the redshifting of the spectrum, and stellar population evolution are taken into account, this absolute magnitude gives the mean total apparent magnitude mlensm_{\rm lens} of Eq. (22), from which both factors of SBS_B are evaluated. The calibration of the absolute magnitude is described in Appendix A.1, and the computation of the distance modulus μ\mu, the KK-correction, and the evolution correction EE in Appendix A.2. Appendix A.3 treats the magnitude windows and the foreground suppression. The tabulated corrections are available at https://github.com/Jiang-Xin-yue/lens-statistics.

Absolute magnitude at fixed velocity dispersion

The absolute magnitude MlensM_{\rm lens} in Eq. (22) is the mean absolute magnitude at fixed velocity dispersion. We calibrate it with the analysis of Bernardi et al. [14], who modelled the joint distribution of absolute magnitude, effective radius, and velocity dispersion of about 9000 SDSS early-type galaxies as a multivariate Gaussian. Only the part of this distribution that relates MM to V=log⁡10[σ/(km s−1)]V = \log_{10}[\sigma/(\mathrm{km}\,\mathrm{s}^{-1})] enters our calculation. Conditioning the Gaussian on VV gives a mean absolute magnitude

⟨M∣V⟩=M∗+A(V−V∗),A=ρVMσMσV,(28)\langle M \mid V\rangle= M_* + A(V - V_*), \qquad A = \rho_{VM}\frac{\sigma_M}{\sigma_V}, \tag*{(28)}

with an intrinsic scatter

w=σM1−ρVM2,(29)w = \sigma_M\sqrt{1-\rho_{VM}^{2}}, \tag*{(29)}

where M∗M_* and V∗V_* are the mean values of the calibrating distribution, σM\sigma_M and σV\sigma_V are the marginal dispersions in MM and VV, and ρVM\rho_{VM} is their correlation coefficient. The slope AA thus describes how the mean absolute magnitude changes with velocity dispersion. The scatter ww quantifies the spread in luminosity that remains among galaxies of the same velocity dispersion.

Table A.1 lists the calibration parameters adopted for different bands, together with the resulting AA and ww. The SDSS magnitude window is calibrated with the rr-band relation, the BOSS LOWZ and CMASS windows with the rr- and ii-band relations, respectively, and the DESI window with the zz-band relation. The conditional relation above therefore supplies the mean absolute magnitude and scatter required for the apparent-magnitude prediction.

Distance modulus, KK-correction, and luminosity evolution

M∗M_* obtained from Appendix A.1 refers to a calibration at zero redshift, and we correct the predicted absolute magnitude for distance, redshifting, and evolution at the lens redshift below. The distance modulus is

μ(zl)=5log⁡10[DL(zl)/10 pc],\mu(z_l) = 5\log_{10}[D_L(z_l)/10\,\mathrm{pc}],

where DLD_L is the luminosity distance in the flat Λ\LambdaCDM cosmology adopted in Sect. 2. The KK-correction and the evolution correction EE are calculated from the same age-dependent model spectra. We use the simple stellar population (SSP) models of Bruzual & Charlot [21], which combine stellar spectra along evolutionary tracks to predict the integrated light of a population formed in a single burst. The stellar spectra are supplied by the STELIB library (Le Borgne et al. [73]), while a Chabrier initial mass function (Chabrier [31]) sets the relative numbers of stars formed at different masses. We adopt solar metallicity and assume a single burst of star formation at zf=2z_f = 2 (e.g. van Dokkum & van der Marel [134]; Bernardi et al. [15]). The age of the stellar population observed at redshift zz is then t(z)=tU(z)−tU(zf)t(z) = t_U(z) - t_U(z_f), where tU(z)t_U(z) is the age of the Universe at redshift zz.

For AB magnitudes and a photon-counting detector, the same-band KK-correction is written as (Hogg et al. [61])

K(z)=−2.5log⁡10[11+z∫λℓλ[λ/(1+z);t(z)]T(λ) dλ∫λℓλ[λ;t(z)]T(λ) dλ],(30)K(z) = -2.5\log_{10}\left[\frac{\frac{1}{1+z}\int\lambda\ell_\lambda[\lambda/(1+z);t(z)]T(\lambda)\,\mathrm{d}\lambda}{\int\lambda\ell_\lambda[\lambda;t(z)]T(\lambda)\,\mathrm{d}\lambda}\right], \tag*{(30)}

where ℓλ(λ;t)\ell_\lambda(\lambda;t) is the rest-frame spectral luminosity density at age tt. T(λ)T(\lambda) is the total response of filter, optics, and detector, for which we adopt the SDSS rr- and ii-band response functions (Doi et al. [44]) and the DECam zz-band bandpasses used in DES Y3 (Sevilla-Noarbe et al. [112]). The numerator and denominator give the signal through the same filter from the redshifted and the rest-frame spectrum at age t(z)t(z), respectively.

Since M∗M_* in Table A.1 refers to zero redshift, the evolution correction compares the luminosity in the same rest-frame band at ages t(z)t(z) and t(0)t(0),

E(z)=−2.5log⁡10{LT[t(z)]LT[t(0)]},LT(t)=∫λℓλ(λ;t)T(λ) dλ,(31)E(z) = -2.5\log_{10}\left\{\frac{L_T[t(z)]}{L_T[t(0)]}\right\}, \qquad L_T(t) = \int\lambda\ell_\lambda(\lambda;t)T(\lambda)\,\mathrm{d}\lambda, \tag*{(31)}

where LTL_T is the band-weighted luminosity at age tt. Stellar populations fade as they evolve passively, so lens galaxies at higher redshift are intrinsically brighter than at z=0z = 0, which gives E(z)<0E(z) < 0.

The normalization integral in Eq. (19) extends to zsz_s, beyond the highest lens redshift in our sample (zl=1.004z_l = 1.004). Above z=1.05z = 1.05, where the population age approaches zero towards zfz_f, we hold the age fixed and continue to evaluate μ\mu and KK at the actual redshift. Tabulated values of μ\mu, KK, and EE are provided in the repository.

Survey magnitude limits and foreground suppression

Since the predicted mlensm_{\rm lens} is a total magnitude, the target-selection limits of each parent spectroscopic sample must be expressed in terms of total magnitudes before the two can be compared. We describe how the survey-specific magnitude definitions are treated for SDSS, BOSS, and DESI in turn.

SDSS defines its limits in Petrosian magnitudes, which recover about 80% of the total flux of a de Vaucouleurs profile (Strauss et al. [127]), so that

mPetro−mtotal=−2.5log⁡10(0.8)≃0.24.(32)m_{\rm Petro} - m_{\rm total} = -2.5\log_{10}(0.8) \simeq0.24. \tag*{(32)}

Shifting both limits of the adopted Petrosian window, 14.5<r<19.514.5 < r < 19.5, by −0.24-0.24 mag therefore gives a total-magnitude window of 14.26<r<19.2614.26 < r < 19.26. We apply this window to all SDSS systems, whose assignment to the parent sample is described in Sect. 3.2.2.4 BOSS, in contrast, selects its targets on cmodel magnitudes. These estimate the total model flux from a combined fit of de Vaucouleurs and exponential profiles ([1]) and can therefore be used directly as total magnitudes. Since the BOSS galaxy sample comprises both the LOWZ and the CMASS selections, we adopt the LOWZ window 16.0<r<19.616.0 < r < 19.6 and the CMASS window 17.5<i<19.917.5 < i < 19.9 ([101]). Each window is compared with the predicted magnitude distribution in its band, and the larger of the two resulting probabilities is taken as the window factor for SPEC_BOSS systems.

The DESI threshold, zfibre<21.6z_{\mathrm{fibre}} < 21.6 ([143]), instead applies to the flux within the fibre. Converting it into a total-magnitude limit requires the fraction η\eta of the galaxy light that enters the fibre, which depends on the angular size of the lens galaxy and hence on zlz_l and σref\sigma_{\mathrm{ref}}. We obtain this size from the size-velocity dispersion relation of Bernardi et al. [14] and compute η\eta as the fraction of a de Vaucouleurs profile of each lens, convolved with a Gaussian of 1′′1^{\prime\prime} FWHM, that falls within the 0.75′′0.75^{\prime\prime} fibre radius, following the fibre-flux convention of the Legacy Surveys5. Since the fibre collects only part of the light, the DESI threshold corresponds to an effective total-magnitude limit

mlim,z=21.6+2.5log⁡10η,(33)m_{\mathrm{lim},z} = 21.6 + 2.5 \log_{10} \eta, \tag*{(33)}

which we apply to the SPEC_DESI category without a bright-end cut. The size-relation parameters and the aperture-correction table are provided in the repository.

For IMAGING_SPEC_FOLLOWUP systems, which were observed with dedicated spectroscopic follow-up rather than drawn from a magnitude-limited sample, the window factor is set to unity, and for SOURCE_SPEC_FOLLOWUP systems SB≡1S_B \equiv1 (Sect. 3.2). The foreground-suppression factor of Eq. (24) is evaluated with the predicted total ii-band magnitude of the lens galaxy, the ii-band scatter ww from Table A.1, and the fiducial amplitude ε=0.2\varepsilon= 0.2. We set msupm_{\mathrm{sup}} to the 10th percentile of the ii-band magnitudes predicted from the measured velocity dispersions and redshifts. These percentiles are computed once from the full sample, separately for subsets of the categories in Table 2 defined by targeting programme and discovery route, and the adopted values are provided in the repository. The product of the magnitude window and the foreground suppression is the selection function SBS_B that enters the conditional likelihood of the main text.

References

  1. [1]Abazajian, K., Adelman-McCarthy, J. K., Agüeros, M. A., et al. 2004, AJ, 128, 502
  2. [2]Aihara, H., AlSayyad, Y., Ando, M., et al. 2022, PASJ, 74, 247
  3. [3]Aihara, H., Arimoto, N., Armstrong, R., et al. 2018, PASJ, 70, S4
  4. [4]ALMA Partnership, Vlahakis, C., Hunter, T. R., et al. 2015, ApJ, 808, L4
  5. [5]Alves de Oliveira, R., França, J. P. C., & Makler, M. 2025, arXiv e-prints, arXiv:2509.090798
  6. [6]Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935, 167
  7. [7]Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., et al. 2018, AJ, 156, 123
  8. [8]Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33
  9. [9]Auger, M. W., Treu, T., Bolton, A. S., et al. 2009, ApJ, 705, 1099
  10. [10]Auger, M. W., Treu, T., Bolton, A. S., et al. 2010, ApJ, 724, 511
  11. [11]Barnabè, M., Czoske, O., Koopmans, L. V. E., Treu, T., & Bolton, A. S. 2011, MNRAS, 415, 2215
  12. [12]Bellstedt, S., Forbes, D. A., Romanowsky, A. J., et al. 2018, MNRAS, 476, 4543
  13. [13]Bernardi, M., Shankar, F., Hyde, J. B., et al. 2010, MNRAS, 404, 2087
  14. [14]Bernardi, M., Sheth, R. K., Annis, J., et al. 2003a, AJ, 125, 1849
  15. [15]Bernardi, M., Sheth, R. K., Annis, J., et al. 2003b, AJ, 125, 1866
  16. [16]Bezanson, R., van Dokkum, P. G., Franx, M., et al. 2011, ApJ, 737, L31
  17. [17]Bolton, A. S., Brownstein, J. R., Kochanek, C. S., et al. 2012, ApJ, 757, 82
  18. [18]Bolton, A. S., Burles, S., Koopmans, L. V. E., et al. 2008, ApJ, 682, 964
  19. [19]Bose, S., Hellwing, W. A., Frenk, C. S., et al. 2017, MNRAS, 464, 4520
  20. [20]Brown, M. J. I., Dey, A., Jannuzi, B. T., et al. 2007, ApJ, 654, 858
  21. [21]Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000
  22. [22]Cao, S., Biesiada, M., Gavazzi, R., Prie, P. Kowska, A., & Zhu, Z.-H. 2015, ApJ, 806, 185
  23. [23]Cao, S., Biesiada, M., Yao, M., & Zhu, Z.-H. 2016, MNRAS, 461, 2192
  24. [24]Cao, S., Covone, G., & Zhu, Z.-H. 2012, ApJ, 755, 31
  25. [25]Cao, S. & Zhu, Z.-H. 2012, A&A, 538, A43
  26. [26]Cao, X., Li, R., Li, N., et al. 2024, MNRAS, 533, 1960
  27. [27]Capelo, P. R. & Natarajan, P. 2007, New Journal of Physics, 9, 445
  28. [28]Cappellari, M. 2016, ARA&A, 54, 597
  29. [29]Cappellari, M., Romanowsky, A. J., Brodie, J. P., et al. 2015, ApJ, 804, L21
  30. [30]Cataldi, P., Pedrosa, S. E., Tissera, P. B., & Artale, M. C. 2021, MNRAS, 501, 5679
  31. [31]Chabrier, G. 2003, PASP, 115, 763
  32. [32]Chan, J. H. H., Wong, K. C., Ding, X., et al. 2024, MNRAS, 527, 6253
  33. [33]Chen, Y., Li, R., Shu, Y., & Cao, X. 2019, MNRAS, 488, 3745
  34. [34]Choi, Y.-Y., Park, C., & Vogeley, M. S. 2007, ApJ, 658, 884
  35. [35]Collett, T. E. 2015, ApJ, 811, 20
  36. [36]Collett, T. E., Sonnenfeld, A., Frohmaier, C., et al. 2023, The Messenger, 190, 49
  37. [37]Collier, W. P., Smith, R. J., & Lucey, J. R. 2018, MNRAS, 478, 1595
  38. [38]Davis, A. N., Huterer, D., & Krauss, L. M. 2003, MNRAS, 344, 1029
  39. [39]Derkenne, C., McDermid, R. M., Poci, A., et al. 2023, MNRAS, 522, 3602
  40. [40]Derkenne, C., McDermid, R. M., Poci, A., et al. 2021, MNRAS, 506, 3691
  41. [41]DESI Collaboration, Abdul Karim, M., Adame, A. G., et al. 2026, AJ, 171, 285
  42. [42]DESI Collaboration, Aghamousa, A., Aguilar, J., et al. 2016, arXiv e-prints, arXiv:1611.00036
  43. [43]Di Cintio, A., Brook, C. B., Macciò, A. V., et al. 2014, MNRAS, 437, 415
  44. [44]Doi, M., Tanaka, M., Fukugita, M., et al. 2010, AJ, 139, 1628
  45. [45]Emsellem, E., Monnet, G., & Bacon, R. 1994, A&A, 285, 723
  46. [46]Etherington, A., Nightingale, J. W., Massey, R., et al. 2022, MNRAS, 517, 3275
  47. [47]Etherington, A., Nightingale, J. W., Massey, R., et al. 2023, MNRAS, 521, 6005
  48. [48]Euclid Collaboration, Walmsley, M., Holloway, P., et al. 2026, A&A, 711, A26
  49. [49]Faber, S. M., Willmer, C. N. A., Wolf, C., et al. 2007, ApJ, 665, 265
  50. [50]Falco, E. E., Gorenstein, M. V., & Shapiro, I. I. 1985, ApJ, 289, L1
  51. [51]Faure, C., Kneib, J.-P., Covone, G., et al. 2008, ApJS, 176, 19
  52. [52]Ferrami, G. & Wyithe, J. S. B. 2024, MNRAS, 532, 1832
  53. [53]Ferrami, G. & Wyithe, J. S. B. 2025, MNRAS, 537, 779
  54. [54]Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125, 306
  55. [55]Galan, A., Vernardos, G., Minor, Q., et al. 2024, A&A, 692, A87
  56. [56]Geng, S., Grespan, M., Thuruthipilly, H., et al. 2025, A&A, 694, A196
  57. [57]Guo, Q. & White, S. D. M. 2008, MNRAS, 384, 2
  58. [58]Guo, Z., Chen, Y., Shu, Y., et al. 2026, MNRAS, 549, stag1029
  59. [61]Hogg, D. W., Baldry, I. K., Blanton, M. R., & Eisenstein, D. J. 2002, arXiv e-prints [arXiv:astro-ph/0210394]DOI
  60. [62]Huang, X., Inchausti, J. C., Storfer, C. J., et al. 2025, arXiv e-prints, arXiv:2509.18089arxiv.org/abs/2509.18089
  61. [63]Hyde, J. B. & Bernardi, M. 2009, MNRAS, 394, 1978
  62. [64]Jackson, R. A., Kaviraj, S., Yi, S. K., et al. 2024, MNRAS, 528, 1655
  63. [65]Jaelani, A. T., More, A., Sonnenfeld, A., et al. 2020, MNRAS, 494, 3156
  64. [66]Jaelani, A. T., More, A., Wong, K. C., et al. 2024, MNRAS, 535, 1625
  65. [67]Jaelani, A. T., Rusu, C. E., Kayo, I., et al. 2021, MNRAS, 502, 1487
  66. [68]Katz, H., Lelli, F., McGaugh, S. S., et al. 2017, MNRAS, 466, 1648
  67. [69]Keeton, C. R. & Kochanek, C. S. 1998, ApJ, 495, 157
  68. [70]Koopmans, L. V. E. 2006, in EAS Publications Series, Vol. 20, EAS Publications Series, ed. G. A. Mamon, F. Combes, C. D. Deffayet, & B. Fort (EDP), 161–166
  69. [71]Koopmans, L. V. E., Bolton, A., Treu, T., et al. 2009, ApJ, 703, L51
  70. [72]Koopmans, L. V. E., Treu, T., Bolton, A. S., Burles, S., & Moustakas, L. A. 2006, ApJ, 649, 599
  71. [73]Le Borgne, J.-F., Bruzual, G., Pelló, R., et al. 2003, A&A, 402, 433
  72. [74]Lemon, C., Courbin, F., More, A., et al. 2024, Space Sci. Rev., 220, 23
  73. [75]Lemon, C. A., Auger, M. W., & McMahon, R. G. 2019, MNRAS, 483, 4242
  74. [76]Lemon, C. A., Auger, M. W., McMahon, R. G., & Ostrovski, F. 2018, MNRAS, 479, 5060
  75. [80]Lyskova, N., Churazov, E., & Naab, T. 2018, MNRAS, 475, 2403
  76. [81]Mastromarino, C., Despali, G., Moscardini, L., et al. 2023, MNRAS, 524, 1515
  77. [82]Meneghetti, M. 2021, Introduction to Gravitational Lensing; With Python Examples, Vol. 956
  78. [83]Montero-Dorta, A. D., Bolton, A. S., & Shu, Y. 2017, MNRAS, 468, 47
  79. [84]More, A., Oguri, M., Kayo, I., et al. 2016, MNRAS, 456, 1595
  80. [85]Moustakas, J., Buhler, J., Scholte, D., Dey, B., & Khederlarian, A. 2023, Fast-SpecFit: Fast spectral synthesis and emission-line fitting of DESI spectra, Astrophysics Source Code Library, record ascl:2308.005
  81. [86]Mozumdar, P., Cappellari, M., Fassnacht, C. D., & Treu, T. 2026, arXiv e-prints, arXiv:2602.10347arxiv.org/abs/2602.10347
  82. [87]Mukherjee, S., Koopmans, L. V. E., Metcalf, R. B., et al. 2021, MNRAS, 504, 3455
  83. [88]Naab, T., Johansson, P. H., Ostriker, J. P., & Efstathiou, G. 2007, ApJ, 658, 710
  84. [89]Newman, A. B., Belli, S., Ellis, R. S., & Patel, S. G. 2018, ApJ, 862, 126
  85. [90]Nightingale, J., Hayes, R., Kelly, A., et al. 2021, The Journal of Open Source Software, 6, 2825
  86. [91]Ofek, E. O., Rix, H.-W., & Maoz, D. 2003, MNRAS, 343, 639
  87. [92]Oguri, M., Inada, N., Clocchiatti, A., et al. 2008, AJ, 135, 520
  88. [93]Oguri, M., Inada, N., Strauss, M. A., et al. 2012, AJ, 143, 120
  89. [94]O’Riordan, C. M., Warren, S. J., & Mortlock, D. J. 2019, MNRAS, 487, 5143
  90. [95]Oser, L., OStriker, J. P., Naab, T., Johansson, P. H., & Burkert, A. 2010, ApJ, 725, 2312
  91. [96]Peirani, S., Dubois, Y., Volonteri, M., et al. 2017, MNRAS, 472, 2153
  92. [97]Peirani, S., Sonnenfeld, A., Gavazzi, R., et al. 2019, MNRAS, 483, 4615
  93. [98]Poci, A., Cappellari, M., & McDermid, R. M. 2017, MNRAS, 467, 1397
  94. [99]Primack, J. R. 2024, Annual Review of Nuclear and Particle Science, 74, 173
  95. [100]Read, J. I., Iorio, G., Agertz, O., & Fraternali, F. 2017, MNRAS, 467, 2019
  96. [101]Reid, B., Ho, S., Padmanabhan, N., et al. 2016, MNRAS, 455, 1553
  97. [102]Remus, R.-S., Dolag, K., Naab, T., et al. 2017, MNRAS, 464, 3742
  98. [103]Ruff, A. J., Gavazzi, R., Marshall, P. J., et al. 2011, ApJ, 727, 96
  99. [104]Saglia, R. P., Sánchez-Blázquez, P., Bender, R., et al. 2010, A&A, 524, A6
  100. [105]Sahu, N., Tran, K.-V., Suyu, S. H., et al. 2024, ApJ, 970, 86
  101. [106]Santini, P., Maiolino, R., Magnelli, B., et al. 2014, A&A, 562, A30
  102. [107]Savary, E., Rojas, K., Maus, M., et al. 2022, A&A, 666, A1
  103. [108]Schaller, M., Frenk, C. S., Bower, R. G., et al. 2015, MNRAS, 451, 1247
  104. [109]Schneider, P., Ehlers, J., & Falco, E. E. 1992, Gravitational Lenses
  105. [110]Schuldt, S., Cañameras, R., Andika, I. T., et al. 2025, A&A, 693, A291
  106. [111]Schwab, J., Bolton, A. S., & Rappaport, S. A. 2010, ApJ, 708, 750
  107. [112]Sevilla-Noarbe, I., Bechtol, K., Carrasco Kind, M., et al. 2021, ApJS, 254, 24
  108. [114]Shu, Y., Brownstein, J. R., Bolton, A. S., et al. 2017, ApJ, 851, 48
  109. [115]Shu, Y. & Li, S. 2025, Science China Physics, Mechanics, and Astronomy, 68, 129511
  110. [116]Smith, R. J., Collier, W. P., Ozaki, S., & Lucey, J. R. 2020, MNRAS, 493, L33
  111. [117]Sonnenfeld, A. 2024, A&A, 690, A325
  112. [118]Sonnenfeld, A. 2025, A&A, 697, A95
  113. [119]Sonnenfeld, A., Chan, J. H. H., Shu, Y., et al. 2018, PASJ, 70, S29
  114. [120]Sonnenfeld, A., Gavazzi, R., Suyu, S. H., Treu, T., & Marshall, P. J. 2013a, ApJ, 777, 97
  115. [121]Sonnenfeld, A., Jaelani, A. T., Chan, J., et al. 2019, A&A, 630, A71
  116. [122]Sonnenfeld, A., Nipoti, C., & Treu, T. 2014, ApJ, 786, 89
  117. [123]Sonnenfeld, A., Treu, T., Gavazzi, R., et al. 2013b, ApJ, 777, 98
  118. [124]Storfer, C., Huang, X., Gu, A., et al. 2024, ApJS, 274, 16
  119. [125]Storfer, C. J., Magnier, E. A., Huang, X., et al. 2026, ApJ, 999, 1
  120. [126]Straatman, C. M. S., van der Wel, A., Bezanson, R., et al. 2018, ApJS, 239, 27
  121. [127]Strauss, M. A., Weinberg, D. H., Lupton, R. H., et al. 2002, AJ, 124, 1810
  122. [128]Tan, C. Y., Shajib, A. J., Birrer, S., et al. 2024, MNRAS, 530, 1474
  123. [129]Taylor, L., Bezanson, R., van der Wel, A., et al. 2022, ApJ, 939, 90
  124. [130]Tran, K.-V. H., Harshan, A., Glazebrook, K., et al. 2022, AJ, 164, 148
  125. [131]Tulin, S. & Yu, H.-B. 2018, Phys. Rep., 730, 1
  126. [132]Unruh, S., Schneider, P., & Sluse, D. 2017, A&A, 601, A77
  127. [133]van der Wel, A., Franx, M., van Dokkum, P. G., et al. 2014, ApJ, 788, 28
  128. [134]van Dokkum, P. G. & van der Marel, R. P. 2007, ApJ, 655, 30
  129. [135]Vogelsberger, M., Marinacci, F., Torrey, P., & Puchwein, E. 2020, Nature Reviews Physics, 2, 42
  130. [136]Wang, L., Shu, Y., Li, R., et al. 2017, MNRAS, 468, 3757
  131. [137]Wang, Y., Vogelsberger, M., Xu, D., et al. 2019, MNRAS, 490, 5722
  132. [138]Wedig, B., Daylan, T., Birrer, S., et al. 2025, ApJ, 986, 42
  133. [139]Weiner, C., Serjeant, S., & Sedgwick, C. 2020, Research Notes of the American Astronomical Society, 4, 190
  134. [140]White, S. D. M. & Frenk, C. S. 1991, ApJ, 379, 52
  135. [143]Zhou, R., Dey, B., Newman, J. A., et al. 2023, AJ, 165, 58

Paper details

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

views · 0

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