From one distance and one redshift to the Hubble constant
Sandbox Physics ·
A bright siren measures the Hubble constant by dividing the host galaxy's Hubble-flow velocity by the gravitational-wave distance. The posterior is broad for one event because the distance is uncertain by 10 to 20 percent, and nearby events add a second error from the galaxy's own motion. For GW170817 this gave 70.0−8.0+12.0kms−1Mpc−1. This chapter writes the inference out, works through that event, and shows why the redshift must come from outside the waveform.
For nearby galaxies the recession velocity grows in proportion to distance,
vH=H0d.
At larger redshift the luminosity distance in a flat universe with matter fraction Ωm is
DL(z)=H0c(1+z)∫0zΩm(1+z′)3+1−Ωmdz′,
and for small z it expands as DL≈(cz/H0)[1+(1−q0)z/2], with deceleration parameter q0=23Ωm−1. For Ωm=0.3, q0=−0.55.
Table 3.1 Luminosity distance at H0=70kms−1Mpc−1 and Ωm=0.3: the full integral, checked against the astropy library, and the two-term series.
Redshift z
Integral
Series
cz/H0
0.01
43.16Mpc
43.16Mpc
42.83Mpc
0.03
131.44Mpc
131.48Mpc
128.49Mpc
0.07
315.66Mpc
316.06Mpc
299.79Mpc
Every distance scales as 1/H0. A siren measures DL, so if the redshift is known, H0 follows. Within 100Mpc the plain Hubble law is good to better than 2 percent, much smaller than the distance error of one event.
Writing the inference for one bright siren
A galaxy's measured velocity is its Hubble-flow velocity plus a peculiar velocity vp caused by nearby mass. Write the gravitational-wave data as x and the measured velocity as v. With a prior p(d)∝d2 on distance, the posterior is
where N(v;H0d,σv) is a normal density centred on H0d and σv combines the velocity measurement and the peculiar-velocity uncertainty. Abbott et al. write the same model with vp as a separate parameter and add a selection term Ns(H0). They show that Ns is constant when the prior is placed on distance and only the gravitational-wave threshold matters. If the prior is placed on redshift instead, Ns∝H03 and cancels an equal factor elsewhere, giving the same posterior. Abbott et al., A gravitational-wave standard siren measurement of the Hubble constant, Methods, Eqs. 7 to 12
The integral over cosι is where the tilt and distance trade of the previous chapter enters. The Lab draws the integrand as the likelihood map, and the bright-siren curve is this posterior on a grid.
Peculiar velocities
A peculiar velocity of a few hundred kms−1 is a fixed error in velocity, so its fractional effect on H0 is σv/(H0d). It matters most for the nearest events.
Worked example 1. At d=40Mpc and H0=70, vH=2800kms−1. A 150kms−1 uncertainty is 5.4 percent. At d=14Mpc, as in the Lab's peculiar-velocity preset, vH=980kms−1 and a spread of 450kms−1 is 46 percent: the velocity error now outweighs the distance error from the wave.
Worked example: GW170817
The gravitational-wave distance was d=43.8−6.9+2.9Mpc. NGC 4993 belongs to a group whose recession velocity relative to the cosmic microwave background is 3327±72kms−1. The authors subtract a coherent bulk flow of 310kms−1 and, being cautious about how well that flow is modelled, assign it an uncertainty of 150kms−1 instead of the formal 69kms−1. That gives vH=3017±166kms−1. Abbott et al., A gravitational-wave standard siren measurement of the Hubble constant
A quick estimate divides: 3017/43.8=68.9. Using the ends of the distance interval, 3017/46.7=64.6 and 3017/36.9=81.8. The interval is lopsided toward high values because the distance interval is lopsided toward short distances. The full posterior, with prior p(H0)∝1/H0, has its maximum at 70.0, a minimal 68.3 percent interval from 62 to 82, a mean of 78 and a standard deviation of 15. The distance interval came mainly from the tilt: the data allow a face-on orbit that is farther away.
The redshift must come from outside
The previous chapter showed that the waveform reveals (1+z)M and DL. The scaling behind that is exact in general relativity, so no improvement in detectors removes it. Cutler and Flanagan, Gravitational waves from merging compact binaries, Eqs. 18 and 19 A siren therefore needs one of three outside facts: the redshift of an identified host, a statistical redshift from every galaxy that could be the host, or a known feature in the source-frame mass distribution. The first is a bright siren. The second and third are the subject of the next chapter.
For GW170817 at z≈0.01, the redshifted chirp mass is 1 percent heavier than the true one, which hardly matters. At z=0.1 the difference is 10 percent, and treating the redshifted mass as the source mass gives the wrong astrophysics.
How the error shrinks with more events
Each bright siren gives an independent posterior, and the product narrows roughly as 1/N. Chen, Fishbach and Holz simulated a three-detector network and found a fractional uncertainty of about 15%/N for N binary neutron stars with counterparts. Chen, Fishbach and Holz, A two per cent Hubble constant measurement from standard sirens within five years That is a forecast from simulations, not a measurement. It predicts that about 56 such events would reach 2 percent.
Lab tasks
Compare the kilonova host and peculiar-velocity presets. Explain from Worked example 1 why the bright-siren posterior widens although the distance band does not.
In the kilonova host preset, press New noise several times. Record how far the bright-siren median moves and compare it with the readout's interval.
Open the advanced group and raise the peculiar-velocity spread from 200 to 600kms−1. At which binary distance does the bright-siren width stop depending on the spread?
Exercises with explained answers
A siren at d=120±12Mpc has a host with vH=8400±250kms−1. Estimate H0 and its uncertainty, treating both errors as small and independent.
At what distance does a velocity uncertainty of 200kms−1 equal a 10 percent distance error, for H0=70?
Using the 15%/N forecast, how many bright sirens would give 5 percent?
GW170817's quick estimate 68.9 differs from the published maximum of 70.0. Name two reasons a ratio of central values need not equal the posterior maximum.
AnswersShow the answer
H0=8400/120=70.0. The fractional errors are 10% and 3.0%, so the total is 0.102+0.0302=10.4%, about ±7.3kms−1Mpc−1.
Equality needs σv/(H0d)=0.10, so d=200/(0.10×70)=28.6Mpc. Closer than that, the velocity error dominates a 10 percent siren.
15/N=5 gives N=9.
The distance posterior is skewed, so the mean of 1/d is not 1/dˉ. The prior p(H0)∝1/H0 and the volume prior on d also shift the peak, and the full model integrates over every tilt and distance.
Try it in the Lab
Compare the kilonova host and peculiar-velocity presets: the distance band is similar, but the bright-siren posterior widens. Explain why from the velocity spread. Open the standard sirens Lab