---
title: "Reading the GWTC-3 dark-siren cosmology paper"
description: "Separate what the GWTC-3 analysis measured from what its population model assumed, with catalog incompleteness and lensing alongside."
canonical_url: "https://sandboxphysics.com/stories/standard-sirens/reading-the-gwtc3-dark-siren-cosmology-paper/"
language: "en"
page_type: "story"
alternate_languages:
  zh-CN: "https://sandboxphysics.com/zh/stories/standard-sirens/reading-the-gwtc3-dark-siren-cosmology-paper/"
last_modified: "2026-11-14"
---

[Gravitational-wave standard sirens](https://sandboxphysics.com/stories/standard-sirens/)

# Reading the GWTC-3 dark-siren cosmology paper

Sandbox Physics · 2026-11-14

The GWTC-3 cosmology paper combines 47 events with GW170817 and reports $H_0=68^{+8}_{-6}\,\mathrm{km\,s^{-1}\,Mpc^{-1}}$ from galaxy catalogs and $68^{+12}_{-8}$ from black-hole masses alone. Its most important statement is a warning: the catalog result is driven mostly by what the authors assumed about the black-hole mass distribution, and only one event, GW190814, carries information of its own. Missing galaxies and gravitational lensing are discussed but are not the leading problem at this sensitivity.

[Open the standard sirens Lab](https://sandboxphysics.com/labs/gravitational-wave-standard-sirens/)

This guide reads the paper section by section and labels each claim as a measurement, a model-dependent result, a forecast or an open question. It assumes [dark sirens, galaxy catalogs and the selection normalisation](https://sandboxphysics.com/stories/standard-sirens/dark-sirens-and-selection-effects/).

## What we read and how to label it

We read the full text of arXiv version 2, dated 19 November 2021. We did not check whether a journal version differs. [LIGO, Virgo and KAGRA, Constraints on the cosmic expansion history from GWTC-3](https://arxiv.org/abs/2111.03604v2)

**Table 6.1** Claims used in this guide and their status.

| Claim | Source | Status |
| --- | --- | --- |
| $H_0=70.0^{+12.0}_{-8.0}$ from GW170817 and NGC 4993 | Abbott et al. 2017, Nature 551, 85 | measurement, peer reviewed |
| $H_0=68^{+8}_{-6}$, catalogs plus GW170817 | GWTC-3 cosmology, Sec. 4.2 | result conditional on a fixed black-hole population |
| $H_0=68^{+12}_{-8}$, masses plus GW170817 | GWTC-3 cosmology, abstract and Sec. 4.1 | result conditional on a mass scale that does not evolve with redshift |
| A completeness correction keeps catalog results unbiased | Gray et al., mock data | tested on simulations only |
| Lensing adds about 5 to 10 percent per distant event | Holz and Hughes 2005 | forecast |
| About 2 percent within five years | Chen, Fishbach and Holz | forecast; we did not check whether it has been met |
| How to separate population and cosmology | GWTC-3 cosmology, Sec. 6 | open question |

## The data: 47 events and two ways to get a redshift

The analysis keeps events with network signal-to-noise ratio above $11$: 42 binary black holes, the neutron-star mergers GW170817 and GW190425, two neutron star and black hole mergers and the asymmetric binary GW190814. Each event gives a luminosity distance. The redshift comes from one of two places. The first uses the redshifted masses: if source masses pile up at a known scale, the observed $(1+z)\mathcal M$ reveals $z$. The second is the galaxy catalog sum, using GLADE+ in the infrared K band and the blue $B_J$ band. The catalog method uses the gwcosmo code of Gray et al. [LIGO, Virgo and KAGRA, Constraints on the cosmic expansion history from GWTC-3, Secs. 2 and 3](https://arxiv.org/abs/2111.03604v2)

Read Table 1 of the paper alongside this. For each event it lists the localisation volume, the number of catalog galaxies inside it and the probability that the host is in GLADE+. For most black-hole events the volume is large and the probability is small, so the catalog contributes little.

## Section 4.2 and Figure 9: what the catalog adds

Using all dark sirens with K-band galaxies, the paper finds $67^{+13}_{-12}$. With an empty catalog, so that only the population assumptions act, it finds $67^{+14}_{-13}$. Those two results are nearly the same width. Adding GW170817 gives $68^{+8}_{-6}$, and most of the improvement comes from the bright siren. GW190814 alone gives $67^{+46}_{-28}$, but only when galaxies are weighted by luminosity. [LIGO, Virgo and KAGRA, Constraints on the cosmic expansion history from GWTC-3, Sec. 4.2](https://arxiv.org/abs/2111.03604v2)

So the careful reading of $68^{+8}_{-6}$ is: GW170817, plus a dark-siren contribution whose shape is set mainly by the assumed mass distribution, plus one informative dark event.

## Section 5.1 and Figure 11: the mass scale moves the answer

The parameter most correlated with $H_0$ is $\mu_g$, the mass where the excess of black holes is centred. The authors fix $\mu_g$ at values spaced by about $2.5\,\mathrm M_\odot$ around its median of $32.55\,\mathrm M_\odot$. A larger $\mu_g$ needs the events to be nearer in redshift, and the posterior moves to low $H_0$; a smaller $\mu_g$ moves it high. This is the mass and redshift degeneracy of [the earlier chapter](https://sandboxphysics.com/stories/standard-sirens/waveform-amplitude-and-luminosity-distance/), now acting on a whole population. [LIGO, Virgo and KAGRA, Constraints on the cosmic expansion history from GWTC-3, Sec. 5.1](https://arxiv.org/abs/2111.03604v2)

The abstract describes the mass distribution as peaking around $34\,\mathrm M_\odot$, while Sec. 5.1 uses the fitted parameter $\mu_g$ with median $32.55\,\mathrm M_\odot$. These describe related but different things: the shape of the whole distribution, and one parameter of the fitted model.

## Missing galaxies

Gray et al. built the out-of-catalog term used here and tested it on mock data. They split the likelihood into the host being in the catalog or not, with the probability set by the catalog's magnitude limit and a luminosity function. With 249 simulated binary neutron stars, catalogs 75, 50 and 25 percent complete gave 2.21, 2.48 and 3.20 percent precision, against 1.84 percent for a complete catalog. In a more realistic case with luminosity weighting and a 50 percent complete catalog, Table II gives 4.48 percent, while the abstract quotes 4.4 percent for the same setup. [Gray et al., Cosmological inference using gravitational wave standard sirens: a mock data analysis, Table II](https://arxiv.org/abs/1908.06050v4)

That test used a simplified universe: no galaxy clustering, no redshift uncertainties and a catalog about three times denser than reality. It shows the correction is unbiased under those conditions, not that real incompleteness is solved. Our own failure case in the [method guide](https://sandboxphysics.com/stories/standard-sirens/recomputing-bright-and-dark-siren-h0/) shows what happens with no correction at all: the posteriors become over-confident and wrong.

## Lensing

Matter along the path magnifies or dims a gravitational wave just as it does light. Holz and Hughes estimated that, for sources near $z=2$, lensing gives a mean distance error of about $0.005$ and a scatter of about $0.09$, and called the resulting 5 to 10 percent per event insurmountable for single high-redshift sources. [Holz and Hughes, Using gravitational-wave standard sirens, Sec. 4](https://arxiv.org/abs/astro-ph/0504616v1) The GWTC-3 paper argues that weak lensing adds only a few percent of variance per siren, well below the roughly 20 percent distance error of its best events, and ignores it. It argues that contamination by strong lensing is unlikely in its sample and ignores that too. [LIGO, Virgo and KAGRA, Constraints on the cosmic expansion history from GWTC-3, Sec. 5.3](https://arxiv.org/abs/2111.03604v2)

Two notational points matter when reading both papers. Holz and Hughes write the scatter as $\sqrt{\langle(\Delta D_L/D_L)^2\rangle}\simeq0.09$ and call it a variance, although the expression is a root mean square. And they write the measured distance as $D_L/\sqrt\mu$ with $\mu$ the flux magnification, while the GWTC-3 paper writes $\tilde D_L=D_L/\mu$ with $\mu$ described as changing the strain amplitude. In this guide we use $D_L/\sqrt\mu$ with $\mu$ the flux magnification.

## What remains open

The paper itself names the main open problem: dark-siren results are dominated by population assumptions. Ways forward it lists include better-localised events, deeper catalogs, cross-correlation with galaxy surveys, other mass features and neutron-star tidal effects. None of these is a result yet. Whether sirens will settle the disagreement between distance-ladder and early-universe values of $H_0$ was also still open in the papers we read.

## Lab tasks

1. Use the crowded catalog preset, then drag the candidate count from $200$ down to $10$. Compare how far the dark-siren posterior narrows with the catalog-versus-empty comparison of Sec. 4.2.
2. Drag two or three impostor galaxies in the cone to the binary's distance. Watch the dark posterior grow competing peaks while the bright posterior stays put.
3. Raise the detector sensitivity in the advanced group until the event is loud. Which posterior gains more, bright or dark?

## Exercises with explained answers

1. The catalog result $67^{+13}_{-12}$ and the empty-catalog result $67^{+14}_{-13}$ are nearly identical. What does that say about the galaxy information?
2. If the true $\mu_g$ were $35\,\mathrm M_\odot$ rather than $32.55\,\mathrm M_\odot$, which way would the population-driven $H_0$ move?
3. A lensed source has flux magnification $\mu=1.1$. What distance error does it cause?
4. Why is Gray et al.'s 4.48 percent precision with a 50 percent complete catalog a forecast rather than a measurement?

Answers

1. Adding galaxies barely narrows the posterior, so for these poorly localised black-hole events the information comes from the population model, as the paper states.
2. A larger mass scale requires smaller redshifts for the same detector-frame masses, so the posterior moves to lower $H_0$, as in Fig. 11.
3. The measured distance is $D_L/\sqrt{1.1}=0.953D_L$, an underestimate of 4.7 percent.
4. It comes from simulated events in a simulated universe without clustering or redshift errors, and assumes O2-like sensitivity. It shows what the method can do under those conditions.

## Try it in the Lab

Use the crowded catalog preset and drag the candidate count down to 10, then up to 200; describe how the dark-siren posterior changes when the host competes with more galaxies. [Open the standard sirens Lab](https://sandboxphysics.com/labs/gravitational-wave-standard-sirens/)

## Sources

1. [LIGO, Virgo and KAGRA, Constraints on the cosmic expansion history from GWTC-3](https://arxiv.org/abs/2111.03604v2)
2. [Gray et al., Cosmological inference using gravitational wave standard sirens: a mock data analysis, Table II](https://arxiv.org/abs/1908.06050v4)
3. [Holz and Hughes, Using gravitational-wave standard sirens, Sec. 4](https://arxiv.org/abs/astro-ph/0504616v1)

## Keep reading

Gravitational-wave standard sirens · Reading path

### From a wave amplitude to the Hubble constant

1. [Article **How can a gravitational wave tell us how far away it came from?**](https://sandboxphysics.com/stories/standard-sirens/measuring-distance-with-gravitational-waves/)
2. [Article **Waveform amplitude, orbit tilt and luminosity distance**](https://sandboxphysics.com/stories/standard-sirens/waveform-amplitude-and-luminosity-distance/)
3. [Article **From one distance and one redshift to the Hubble constant**](https://sandboxphysics.com/stories/standard-sirens/from-distance-to-hubble-constant/)
4. [Article **Dark sirens, galaxy catalogs and the selection normalisation**](https://sandboxphysics.com/stories/standard-sirens/dark-sirens-and-selection-effects/)
5. [Article **Recomputing a bright-siren and a dark-siren Hubble constant**](https://sandboxphysics.com/stories/standard-sirens/recomputing-bright-and-dark-siren-h0/)
6. [You are here **Reading the GWTC-3 dark-siren cosmology paper**](https://sandboxphysics.com/stories/standard-sirens/reading-the-gwtc3-dark-siren-cosmology-paper/)
