Strong-Lensing Mass and Time-Delay Tests

A multiply imaged variable source provides image morphology, magnification information, and differential arrival times through one lens. Those observables can test the relation between light deflection, stellar dynamics, and cosmological distance, but they do not uniquely determine the radial mass profile. The mass-sheet and related source-position degeneracies are the central inference problem. The assessment below is current through 2026-08-19.

Required background. Geodesics and Jacobi Fields supplies null-ray focusing and arrival-time geometry; Probability and Statistical Inference supplies joint likelihoods, priors, and hierarchical marginalization.

Helpful background. Weak-Field Gravity supplies the lensing and dynamical potentials; Relativistic Cosmology supplies angular-diameter distances and line-of-sight structure.

Lens map and arrival-time surface

For a thin lens, write the dimensionless lens equation and Fermat potential as

$$\boldsymbol{\beta} =\boldsymbol{\theta}-\boldsymbol{\nabla}\psi(\boldsymbol{\theta}), \qquad \phi(\boldsymbol{\theta},\boldsymbol{\beta}) =\frac12\lvert\boldsymbol{\theta}-\boldsymbol{\beta}\rvert^2 -\psi(\boldsymbol{\theta}).$$

The angles $\boldsymbol\theta$ and $\boldsymbol\beta$ locate the image and unlensed source relative to the lens. The angular gradient of the dimensionless projected potential $\psi$ is the reduced deflection. Images are stationary points of $\phi$, since $\boldsymbol\nabla_\theta\phi=0$ recovers the lens equation. They need not all be minima: saddles and maxima also produce images. The delay between images $A$ and $B$ is

$$\Delta t_{AB} =D_{\Delta t} \left[\phi(\boldsymbol{\theta}_A,\boldsymbol{\beta}) -\phi(\boldsymbol{\theta}_B,\boldsymbol{\beta})\right], \qquad D_{\Delta t}=(1+z_d)\frac{D_dD_s}{D_{ds}},$$

with $c=1$ and angular-diameter distances in the assumed background cosmology. The projected potential $\psi$ derives from the Weyl combination $\Phi+\Psi$; stellar kinematics respond chiefly to $\Phi$. Comparing them can test slip only if both arise from one three-dimensional lens and the same distance model.

The mass-sheet transformation

$$\kappa_\lambda(\boldsymbol{\theta}) =\lambda\kappa(\boldsymbol{\theta})+(1-\lambda), \qquad \boldsymbol{\beta}_\lambda=\lambda\boldsymbol{\beta},$$

uses convergence $\kappa=\tfrac12\nabla_\theta^2\psi$ and a nonzero constant $\lambda$. It preserves the relative image configurations of one source plane while scaling Fermat-potential differences by $\lambda$. To see why, transform the potential as

$$\psi_\lambda(\boldsymbol\theta) =\lambda\psi(\boldsymbol\theta) +\frac{1-\lambda}{2}\lvert\boldsymbol\theta\rvert^2.$$

Its lens equation gives $\boldsymbol\theta-\boldsymbol\nabla\psi_\lambda=\lambda\boldsymbol\beta$. Rescaling the unknown source brightness distribution produces the same image pixels. The Jacobian scales by $\lambda$, so signed magnifications change as $\mu_\lambda=\mu/\lambda^2$ while their ratios stay fixed. Absolute source size or luminosity, if independently known, can supply extra information. The exact transformation here assumes a single lens and source plane; multiple source redshifts or lens planes require their own consistency conditions. Imaging of an unknown single source alone therefore cannot determine the absolute delay distance. More general source-position transformations and flexible radial profiles produce related ambiguities Schneider and Sluse 2013, §2.1 and §§3–5.

What must be observed jointly

A defensible dataset pins the calibrated imaging exposures and point-spread function; the lens and source-light masks; the monitoring light curves and time-delay estimator; stellar spectra or spatially resolved kinematics; and galaxy counts, redshifts, weak shear, or simulations for the environment and line of sight. Selection begins when a survey recognizes a lens and a variable source, continues through monitoring cadence and image-separation cuts, and ends with quality decisions about arcs, delays, and spectra. That process defines the population to which a hierarchical conclusion applies.

Kinematics do not automatically remove the degeneracy. A line-of-sight velocity-dispersion profile depends on the stellar light deprojection, orbital anisotropy, aperture and seeing, dark and luminous density profiles, and the dynamical potential. External convergence rescales the lens inference but may be weakly visible in the main-lens dynamics. Microlensing perturbs quasar flux and sometimes delays; subhalos and line-of-sight perturbers perturb images. These effects require distinct nuisance parameters rather than one “mass-model error.”

The joint posterior can be written schematically as

$$p(\vartheta,\eta\mid d)\propto \mathcal L_{\rm img}\mathcal L_{\Delta t} \mathcal L_{\rm kin}\mathcal L_{\rm env} p(\vartheta,\eta),$$

where $\vartheta$ contains distance or gravity parameters and $\eta$ contains lens profile, source pixels, anisotropy, point-spread response, external convergence, microlensing, and calibration. Each factor must be normalized under its declared data representation, and shared nuisance parameters must propagate covariance. The TDCOSMO IV analysis demonstrates how population information on galaxy density profiles changes the time-delay inference Birrer et al. 2020, §§2–6. TDCOSMO V frames the additional data and sample design needed to control this flexibility Birrer and Treu 2021, §§2–5.

Results and limitations through 2026-08-19

Evidence or position	What the data constrain	Limitation or scope
Extended arcs plus accurate delays	Strongly constrain angular structure and Fermat-potential differences	Mass-sheet and source-position families retain radial freedom
Spatially resolved stellar kinematics	Add sensitivity to the three-dimensional dynamical potential	Anisotropy, deprojection, seeing, and profile assumptions remain
Environment and line-of-sight observations	Inform external convergence and perturber populations	Incomplete redshifts and simulation dependence leave residual covariance

One conservative position reports a conditional distance or slip constraint after broad lens-population marginalization. A competing position adopts a restricted radial family and obtains tighter precision; that precision is persuasive only if the family predicts independent kinematic and environmental data. Negative evidence against a deformation within one lens model does not exclude it under every allowed source-position transformation.

The adversarial check is to generate a synthetic lens, apply an explicit mass-sheet transformation, and verify that the imaging likelihood stays invariant while the delay inference moves. Adding resolved kinematics and line-of-sight data should recover the injected physical solution with calibrated coverage across alternative mass and source bases. Lens modeling in two independent software pipelines is useful, but agreement is not independent if both impose the same radial prior. The gravity interpretation remains unresolved and model-dependent wherever the transformed models survive all available observables.

What could change the conclusion. The interpretation could change when new time-delay, point-spread, resolved-kinematic, environment, or lens-selection products become available, or when an analysis demonstrates robust injection recovery across mass-sheet, source-position, anisotropy, and line-of-sight alternatives.

Exercise

Under a mass-sheet transformation with $\lambda=0.9$, an image model fits the same relative arc structure. How does its predicted time delay change, and why can imaging alone not identify the error?

Solution

The Fermat-potential difference scales by $\lambda$, so at fixed distance the predicted delay becomes $0.9$ times the original value. Equivalently, fitting an unchanged observed delay would rescale the inferred time-delay distance by $1/\lambda$. The transformed source is also smaller by $\lambda$, while the relative image geometry remains compatible. Therefore imaging alone cannot choose between the models; independent kinematics and environment information are needed, with their anisotropy and external-convergence uncertainties propagated.

At fixed dimensionless background cosmology, $D_{\Delta t}\propto H_0^{-1}$, so the fitted Hubble constant would become $H_{0,\lambda}=0.9H_0$. An apparently excellent image fit can therefore coexist with a ten-percent distance-scale bias. This illustration is algebraic, not a measured bias in a particular survey.

References

• Birrer, S., and Treu, T. (2021), “TDCOSMO V: Strategies for Precise and Accurate Measurements of the Hubble Constant with Strong Lensing,” Astronomy & Astrophysics 649, A61. DOI.
• Birrer, S., et al. (2020), “TDCOSMO IV: Hierarchical Time-Delay Cosmography—Joint Inference of the Hubble Constant and Galaxy Density Profiles,” Astronomy & Astrophysics 643, A165. DOI.
• Schneider, P., and Sluse, D. (2013), “Mass-Sheet Degeneracy, Power-Law Models and External Convergence,” Astronomy & Astrophysics 559, A37. DOI.