---
title: "How to read TDCOSMO IV on the mass sheet and the Hubble constant"
description: "Read TDCOSMO IV section by section: the mass-sheet transform, line-of-sight mass, stellar kinematics and the hierarchical inference of the Hubble constant."
canonical_url: "https://sandboxphysics.com/stories/quasar-time-delays/reading-tdcosmo-iv/"
language: "en"
page_type: "story"
alternate_languages:
  zh-CN: "https://sandboxphysics.com/zh/stories/quasar-time-delays/reading-tdcosmo-iv/"
last_modified: "2026-11-14"
---

[Lensed quasar time delays and the Hubble constant](https://sandboxphysics.com/stories/quasar-time-delays/)

# How to read TDCOSMO IV on the mass sheet and the Hubble constant

Sandbox Physics · 2026-11-14

TDCOSMO IV asks how far the lens galaxies' mass profiles could depart from a power law and still fit the images and delays. Its answer is that the mass sheet stays almost free, so only the stars' velocity dispersion can fix it. With the seven time-delay lenses alone the paper gets $H_0=74.5^{+5.6}_{-6.1}$, a 9 percent measurement. Adding 33 SLACS lenses gives $67.4^{+4.1}_{-3.2}$, about 5 percent, but only under the assumption that the two samples come from the same galaxy population. Read it for that trade: fewer assumptions about the mass profile, a larger error, and a new assumption about the population.

This guide follows the paper section by section. Read [the mass-sheet chapter](https://sandboxphysics.com/stories/quasar-time-delays/mass-sheet-degeneracy-and-stellar-speeds/) first, and try [the recompute guide](https://sandboxphysics.com/stories/quasar-time-delays/recompute-h0-from-a-frozen-tdcosmo-slice/) if you want to handle Table 2 yourself.

## Which version to read

Birrer et al., "TDCOSMO IV: Hierarchical time-delay cosmography: joint inference of the Hubble constant and galaxy density profiles", arXiv:2007.02941. Version 1 is dated 6 July 2020, version 2 12 October 2020 and version 3 19 December 2020. The published paper is A&A 643, A165 (2020). Section, equation, table and figure numbers below come from version 3 as rendered by ar5iv; check them against the PDF of the version you hold. [TDCOSMO IV, arXiv version 3](https://arxiv.org/abs/2007.02941v3)

## Part 1: one lens and the transform, Secs. 2.1 to 2.3

Section 2.1 sets up the lens equation, the Fermat potential and the time-delay distance $D_{\Delta t}=(1+z_{\rm d})D_{\rm d}D_{\rm s}/D_{\rm ds}$ (Eqs. 1 to 9). Section 2.2 gives the stellar velocity dispersion from the Jeans equation; the key line is Eq. (17), $\sigma^2=(D_{\rm s}/D_{\rm ds})\,c^2J$, which puts the cosmology in a distance ratio and the galaxy in the dimensionless $J$. The same section names the mass-anisotropy degeneracy.

Section 2.3 is the heart of the paper. Eq. (20) is the transformed lens equation and Eq. (21) the transformed convergence. Eqs. (22) to (25) say what happens to the observables: image positions stay, the source moves to $\lambda\boldsymbol\beta$, delays scale by $\lambda$ and the velocity dispersion by $\sqrt\lambda$. Eqs. (26) to (29) say what happens to the inferences: the time-delay distance scales by $\lambda^{-1}$, so does the distance ratio from kinematics, the distance to the lens is unchanged and $H_0$ scales by $\lambda$. If you can derive Eqs. (24), (25) and (29) on paper, the rest of the paper is bookkeeping.

## Part 2: where a sheet comes from, Secs. 2.4 to 2.7

Section 2.4 treats mass along the line of sight as an external sheet, $\kappa_{\rm ext}$ (Eqs. 32 and 33). Section 2.5 splits the total transform into external and internal parts, $\lambda=(1-\kappa_{\rm s})\lambda_{\rm int}$ (Eq. 34). Only the internal part is new here; earlier work already carried $\kappa_{\rm ext}$.

Section 2.6 answers the obvious objection that an infinite sheet is unphysical. It replaces the sheet with a cored component of finite size. Figure 2 shows that imaging can constrain the core when it is small, but not when its radius is beyond about five Einstein radii. Figure 3 marks which cores are physically allowed, and Figure 4 shows the predicted velocity dispersion matches the exact-sheet formula to better than 1 percent for $\lambda_{\rm c}$ between 0.8 and 1.2. That is why the paper can use the simple algebra of Sec. 2.3 for a realistic profile.

Section 2.7 writes the likelihood of one lens as a product of four independent data sets: imaging, delays, kinematics and line-of-sight counts (Eqs. 40 to 42). Note which parameters each factor sees. The delay term sees $\lambda D_{\Delta t}$; the kinematics term sees $\lambda$ and the anisotropy; imaging sees neither.

## Part 3: the hierarchy and its test, Secs. 3 and 4

Section 3 promotes $\lambda_{\rm int}$ and the anisotropy from per-lens numbers to population distributions, with a mean, a scatter and a trend of $\lambda_{\rm int}$ with $r_{\rm eff}/\theta_{\rm E}$, the parameter $\alpha_\lambda$. A single lens cannot pin $\lambda$ down; a population can, if the population assumption holds.

Section 4 runs the method on Rung 3 of the time-delay lens modeling challenge, 16 mock lenses ray-traced through hydrodynamic simulations with a hidden $H_0=65.413$. The paper finds $66.9\pm4.2$ with its preferred anisotropy prior (Figure 5). Read the caveats in the same section: the simulations have known limits below the kiloparsec scale, so this test supports the method only down to a few percent.

## Part 4: seven lenses alone, Sec. 5

Section 5.1 and Table 2 give each lens's redshifts, radii, slope, $\kappa_{\rm ext}$ and the power-law time-delay distance. Figure 6 shows the $\kappa_{\rm ext}$ posteriors, with sample mean $0.035^{+0.021}_{-0.016}$. Section 5.2 and Figure 7 give the hierarchical result, $H_0=74.5^{+5.6}_{-6.1}$ and $\lambda_{\rm int,0}=1.02^{+0.08}_{-0.09}$. Section 5.2 says plainly why the error grew from earlier work: the profile assumption was removed and only the kinematics fix $\lambda$.

One sentence in Sec. 5.2 is easy to miss: an overall shift in $\kappa_{\rm ext}$ shared by all lenses would be absorbed by $\lambda_{\rm int}$, because the kinematics constrain the total $\lambda$.

## Part 5: borrowing SLACS, Secs. 6 and 7

Section 6 adds 33 SLACS lenses that have imaging and stellar kinematics but no time delays; nine have spatially resolved IFU spectra. Section 6.3 explains how they were selected to resemble the TDCOSMO deflectors, and Figure 8 compares the two samples parameter by parameter. Read Fig. 8 critically, because everything after it depends on that resemblance.

Section 7.2, Table 6 and Figure 12 give four combinations under the same priors.

**Table 6.1** The four results of Table 6 of TDCOSMO IV, in km s$^{-1}$ Mpc$^{-1}$.

| Data | $H_0$ | $\lambda_{\rm int,0}$ |
| --- | --- | --- |
| TDCOSMO only | $74.5^{+5.6}_{-6.1}$ | $1.02^{+0.08}_{-0.09}$ |
| TDCOSMO + SLACS IFU | $73.3\pm5.8$ | $1.00\pm0.08$ |
| TDCOSMO + SLACS SDSS | $67.4^{+4.3}_{-4.7}$ | $0.91^{+0.05}_{-0.06}$ |
| TDCOSMO + SLACS SDSS + IFU | $67.4^{+4.1}_{-3.2}$ | $0.91\pm0.04$ |

Table 6 also lists $\alpha_\lambda$, the scatters and the anisotropy. Note that the TDCOSMO + SLACS IFU row, where SLACS informs only the anisotropy, stays close to TDCOSMO only; the large shift comes with the SDSS kinematics of all 33 SLACS lenses.

## Part 6: what the authors think it means, Sec. 8

Section 8.1 interprets $\lambda_{\rm int}=0.91$ as a cored mass component on top of the power law, and Figure 17 draws it: the two profiles agree near the Einstein radius and differ by up to 50 percent well outside it.

Section 8.2 breaks down the 5 percent error of the combined result. Fixing $\lambda_{\rm int}=1$ would leave about 2 percent, so the sheet contributes about 4.5 percent. Velocity dispersion measurements contribute 3 percent, and the anisotropy prior and $\alpha_\lambda$ share the remaining 3.5 percent.

Section 8.3 lists what the validation does not cover: kinematics measurement and modeling (8.3.1), selection differences between samples (8.3.2), line-of-sight structure (8.3.3) and more flexible models (8.3.4). Section 8.4 says the next step is spatially resolved kinematics.

Section 8.5 was written after unblinding, and the paper says so. It traces the drop of about 6 units from TDCOSMO-only to TDCOSMO + SLACS to two causes: the SLACS IFU data prefer a lower anisotropy parameter, and the SLACS lenses prefer a $\lambda_{\rm int,0}$ about 8 percent lower. It then offers two explanations, more radial orbits than the prior assumed or a real difference between quad-selected TDCOSMO galaxies and mostly double SLACS galaxies, and says resolved kinematics can tell them apart. Figure 18 compares the blind measurements of the collaboration.

## Measured, assumed and inferred

Keep these apart while reading:

- **Measured:** image positions, flux ratios, time delays, velocity dispersions, galaxy counts around each lens.
- **Assumed:** a power-law profile plus a sheet-like freedom, an Osipkov–Merritt anisotropy form, flat $\Lambda$CDM, and, for the 67.4 result, a shared population for TDCOSMO and SLACS deflectors.
- **Inferred:** $\lambda_{\rm int,0}$, its trend $\alpha_\lambda$, the anisotropy distribution and $H_0$.
- **Open in this paper:** whether the two samples share a population, which Sec. 8.5 leaves to future resolved kinematics. This guide does not cover later TDCOSMO papers.

Image positions alone never appear in the list of things that fix $H_0$. The delays fix $\lambda D_{\Delta t}$, and kinematics fix $\lambda$.

## A checklist for this paper

1. Derive Eqs. (24), (25) and (29) of Sec. 2.3 yourself.
2. In Eqs. (40) to (42), mark which data set sees $\lambda$ and which sees $D_{\Delta t}$.
3. Compare Fig. 5 with the hidden challenge value and read the Sec. 4 caveats.
4. In Table 2, find the lens with the smallest $D_{\Delta t}^{\rm pl}$ error and ask whether its $\kappa_{\rm ext}$ error is also small.
5. In Table 6, separate the result that needs the same-population assumption from the one that does not.
6. Read Sec. 8.5 knowing it was written after unblinding.

In [the gravitational-lens Lab](https://sandboxphysics.com/labs/gravitational-lens-telescope/?mode=time-delay), load "Same images, hidden sheet" and tighten the stellar-speed error under the advanced settings to see the band of allowed sheets shrink, the single-lens version of Sec. 2.7.

## Sources

- Birrer et al. 2020, TDCOSMO IV, A&A 643, A165. [arXiv:2007.02941v3](https://arxiv.org/abs/2007.02941v3)
- Wong et al. 2020, H0LiCOW XIII, for the earlier six-lens result the paper compares against. [arXiv:1907.04869v2](https://arxiv.org/abs/1907.04869v2)
- Schneider and Sluse 2013, for an explicit composite lens that a power law fits but that gives a different $H_0$. [arXiv:1306.0901v2](https://arxiv.org/abs/1306.0901v2)

## Try it in the Lab

[Quasar time delays in the gravitational-lens Lab](https://sandboxphysics.com/labs/gravitational-lens-telescope/?mode=time-delay): Narrow the stellar-speed error and watch how tightly it pins the hidden sheet.

## Sources

1. [TDCOSMO IV, arXiv version 3](https://arxiv.org/abs/2007.02941v3)
2. [arXiv:1907.04869v2](https://arxiv.org/abs/1907.04869v2)
3. [arXiv:1306.0901v2](https://arxiv.org/abs/1306.0901v2)

## Keep reading

Lensed quasar time delays and the Hubble constant · Reading path

### From a delayed flare to the Hubble constant

1. [Article **How a delayed quasar flare measures the size of the universe**](https://sandboxphysics.com/stories/quasar-time-delays/one-flare-many-paths/)
2. [Article **Arrival time, the Fermat potential and the delay of an isothermal lens**](https://sandboxphysics.com/stories/quasar-time-delays/arrival-time-and-the-fermat-potential/)
3. [Article **Mass slope, shear and the time-delay distance**](https://sandboxphysics.com/stories/quasar-time-delays/mass-slope-shear-and-the-time-delay-distance/)
4. [Article **The mass-sheet degeneracy and what stellar speeds add**](https://sandboxphysics.com/stories/quasar-time-delays/mass-sheet-degeneracy-and-stellar-speeds/)
5. [Article **Recompute the Hubble constant from seven TDCOSMO lenses**](https://sandboxphysics.com/stories/quasar-time-delays/recompute-h0-from-a-frozen-tdcosmo-slice/)
6. [You are here **How to read TDCOSMO IV on the mass sheet and the Hubble constant**](https://sandboxphysics.com/stories/quasar-time-delays/reading-tdcosmo-iv/)

1 more group

### Other articles that use the same experiments

Each experiment from this article that another series also uses, followed by the articles there.

1. [Lab **Gravitational Lens Telescope**](https://sandboxphysics.com/labs/gravitational-lens-telescope/)
2. [Article **How a dark planet makes a distant star brighten**](https://sandboxphysics.com/stories/microlensing-planets/a-dark-planet-brightens-a-star/)
3. [Article **From a bent light ray to two stellar images**](https://sandboxphysics.com/stories/microlensing-planets/from-bent-light-to-two-images/)
4. [Article **What happens when a stellar disk crosses a caustic?**](https://sandboxphysics.com/stories/microlensing-planets/a-stellar-disk-crosses-a-caustic/)
5. [Article **What can a microlensing light curve tell us about its lens?**](https://sandboxphysics.com/stories/microlensing-planets/read-a-light-curve-without-weighing-too-soon/)
6. [Article **Read a dark planet’s mass from Earth and space**](https://sandboxphysics.com/stories/microlensing-planets/read-ground-space-and-roman-evidence/)
7. [Article **Fit the published OGLE-2005-BLG-390 photometry**](https://sandboxphysics.com/stories/microlensing-planets/refit-the-published-ogle390-files/)
8. [Article **A background star helped weigh an isolated black hole**](https://sandboxphysics.com/stories/astronomy-gravity/isolated-black-hole-microlensing/)
