The mass-sheet degeneracy and what stellar speeds add
Sandbox Physics ·
Scale a lens's convergence by λ, add a uniform sheet 1−λ, and move the source to λβ. Every image stays exactly where it was and every flux ratio is unchanged, but all delays are multiplied by λ. Images alone therefore cannot tell you λ, and an analyst who assumes λ=1 reads H0/λ from the delays. Counts of galaxies along the line of sight and the speeds of stars inside the lens supply the missing information, because stellar speeds scale as λ rather than λ.
Before this chapter, read mass slope, shear and the time-delay distance. You need the lens equation, the Fermat potential and the idea that the deflection of a round mass is set by the mass inside the radius.
The transform
Start from a lens with deflection α(θ) and a source at β. A uniform sheet of convergence κs deflects by κsθ: by Gauss's law the mean convergence inside any circle is κs. Now build a new lens: the old one scaled by λ plus a sheet 1−λ,
κλ(θ)=λκ(θ)+(1−λ),αλ=λα+(1−λ)θ.
Put its source at λβ. Its lens equation is
λβ=θ−λα(θ)−(1−λ)θ.
The θ terms combine to λθ; divide by λ and you get back β=θ−α(θ). Every image of the old lens is an image of the new one. This is TDCOSMO IV's Eq. (20), the mass-sheet transform that Falco and collaborators described in 1985. TDCOSMO IV, Sec. 2.3
What it changes
Delays. The new Fermat potential is ϕλ=21(θ−λβ)2−λψ−21(1−λ)θ2, because the sheet's potential is 21(1−λ)θ2. Expanding the square and collecting terms,
ϕλ(θ)=λ[ϕ(θ,β)−21β2]+21λ2β2.
The last two pieces do not depend on θ, so they cancel in any difference between images. Every delay is multiplied by λ.
Brightness. The source-plane map is now λβ(θ), so its Jacobian determinant is multiplied by λ2 and every magnification is divided by λ2. We never know how bright the quasar is without the lens, so only flux ratios are measured, and those are unchanged.
The inferred distance. If the truth is the transformed lens and the analyst models the original one, the measured delays are λ times the model's. The analyst infers a time-delay distance λ times the true one, so H0inferred=H0true/λ. TDCOSMO IV states the same thing from the model's side, DΔt,λ=λ−1DΔt and H0,λ=λH0, in its Eqs. (26) and (29).
Image positions and flux ratios leave the transform untouched. The paper notes that only observables tied to the absolute size or intrinsic brightness of the source, or to the lensing potential itself, can break it.
Two physical sources of a sheet
Mass along the line of sight. Galaxies and dark matter in front of and behind the lens focus the light a little. To first order this acts like a sheet, κext, and changes the time-delay distance as
where "bkg" is the smooth-universe value (Eqs. 32 and 33). It is estimated by counting galaxies near the lens or from weak lensing of background galaxies, to a few percent per lens. For the seven TDCOSMO lenses the sample mean is ⟨κext⟩=0.035. TDCOSMO IV, Secs. 2.4 and 5.1
The galaxy's own profile. A real galaxy can depart from the power law in a way that mimics the transform, for example with an extra cored component of dark matter. TDCOSMO IV calls this the internal transform, λint, and writes the total as
λ=(1−κs)λint.
An exact sheet extends to infinity and cannot be the whole truth, but a core several arcseconds wide reproduces it over the region that matters (Sec. 8.1). Schneider and Sluse (2013) built a composite lens of stars plus dark matter that a power law fits extremely well, yet the two give values of H0 about 20 percent apart. Schneider and Sluse 2013, abstract
What stellar speeds add
The stars in the lens galaxy move with a spread of line-of-sight speeds, the velocity dispersion σ. Their motion responds to the galaxy's three-dimensional mass. The Jeans equation, which balances the stars' random motion against gravity, gives the prediction as a cosmology-dependent factor times a dimensionless integral J of the mass, light and orbit model,
Under the transform the galaxy's own mass is scaled by λ, and the speed squared needed to hold stars in orbit scales with the mass, so σλ=λσ (Eq. 25). The delays scale as λ and the speeds as λ, so measuring both pins λ down. A fractional error ϵ on σ becomes an error of about 2ϵ on λ, and therefore on H0.
There is a catch. The predicted σ also depends on how elongated the stars' orbits are, the anisotropy. Velocity dispersion alone cannot separate orbit shape from mass; this is the mass-anisotropy degeneracy. Spatially resolved spectra, which measure σ at several radii, help.
Worked examples
The hidden-sheet preset. In the Lab's Einstein cross, the last image arrives 15.49 days after the first. With λ=0.88 it arrives after 0.88×15.49=13.64 days, at the same place. The brightest image's magnification goes from 5.91 to 5.91/0.882=7.63. Assuming no sheet, H0 reads 70/0.88=79.5.
A real velocity dispersion. B1608+656 has σ=260±15 km/s (TDCOSMO IV, Sec. 5.1). That is a 5.8 percent error, so this one measurement fixes λ only to about 12 percent. This is why TDCOSMO combined many lenses and treated λint as a property of the galaxy population.
The 2020 result. TDCOSMO IV found λint=1.02−0.09+0.08 from the seven time-delay lenses alone and 0.91±0.04 after adding 33 SLACS lenses with stellar speeds. Multiplying the power-law estimate by these factors is the step from H0≈74 to H0≈67. TDCOSMO IV, Table 6
Lab tasks
In the quasar time-delay Lab, load "Einstein cross", note the four delays, then load "Same images, hidden sheet". Check that each delay is multiplied by 0.88 and that the images have not moved.
Move the hidden-sheet slider from 0.8 to 1.2 and watch the line in the Hubble-constant panel. Where does it cross the true value?
Open the advanced settings and change the stellar velocity dispersion error from 0.05 to 0.02. How wide is the band of sheets allowed by the stellar speeds now?
Exercises
Show that the flux ratio of any two images is unchanged by the transform.
The true lens has λ=1.05 and H0=70. An analyst assumes no sheet. What H0 does the analyst infer?
A lens sits in an overdense field with κext=0.06. By what factor must an H0 inferred without it be multiplied?
What fractional error on σ is needed to know λ to 3 percent from one lens?
Suppose the lensed source were a supernova whose true brightness is known. Explain why this could break the degeneracy.
AnswersShow the answer
Each magnification is divided by the same λ2, so their ratio is unchanged.
70/1.05=66.7 km s−1 Mpc−1.
DΔtlens=(1−κext)DΔtbkg, and H0∝1/DΔtbkg, so the corrected H0 is 0.94 times the uncorrected one.
δλ/λ≈2δσ/σ, so δσ/σ≈1.5 percent.
With a known true brightness, the observed brightness gives the absolute magnification μ/λ2, not only flux ratios. Combined with the model's μ, that measures λ.