Before this chapter, read arrival time, the Fermat potential and the isothermal lens. You need integrals in polar coordinates and a second-order Taylor expansion in two variables.
A galaxy with any mass slope
Real galaxies are not exactly isothermal. A flexible model lets the convergence fall as a power of radius,
where θ=∣θ∣ and γ′ is the slope of the three-dimensional density, ρ∝r−γ′. The isothermal sphere is γ′=2. A steeper galaxy, γ′>2, concentrates its mass towards the centre.
For a round lens, the deflection at radius θ is the mean convergence inside that radius times θ: α(θ)=(2/θ)∫0θκ(u)udu. This is the two-dimensional version of Gauss's law. Doing the integral,
At θ=θE the deflection equals the radius, so the mean convergence inside the Einstein radius is exactly 1 for every slope.
Two ways to change the slope
Keep the source fixed. Put the source at β=0.1′′ behind a lens with θE=1′′, zd=0.5 and zs=2. The model in the Lab gives a delay of 16.6 days for γ′=1.8, 2.0 and 2.2 alike. The images move: the outer one sits at 1.124′′ for γ′=1.8 and at 1.084′′ for γ′=2.2. So at fixed source the delay hardly cares about the slope.
Keep the images fixed. This is the situation an astronomer faces. The image positions are measured to a few thousandths of an arcsecond; the source position is not observed at all. Changing the slope now forces a different θE and β to keep the images in place, and the delay changes a lot.
The delay at fixed image positions
Take two images on opposite sides of a round lens, the outer one at radius R1 and the inner one at R2. Kochanek (2002) showed that their delay depends only on the image radii and the convergence in the ring between them. Define the mean convergence in that ring,
Expanding in the ring's thickness ΔR=R1−R2 over its mean radius ⟨R⟩, with a power-law density of slope η in the ring, he found
Kochanek writes this as τ1−τ2 with an overall minus sign; we quote it for the later image minus the earlier one. His η is our γ′. Kochanek 2002, Eqs. (3) to (8)
The leading term is exact for the isothermal sphere. There ⟨κ⟩=1/2 and the bracket's second term vanishes because 1−2×21=0. With R1=θE+β and R2=θE−β, 21(R12−R22)=2θEβ, the result of the previous chapter.
Near the Einstein ring a power law has 1−⟨κ⟩≈(γ′−1)/2. So at fixed images the predicted delay is about (γ′−1) times the isothermal one. For a measured delay, DΔt∝1/(γ′−1) and
Wucknitz (2002) writes the potential as ψ∝rβ, so his β is our 3−γ′. For round lens models with images near the Einstein ring and a fitted external shear he finds H0∝2−β, the same scaling. For quadruple lenses with the shear held fixed he finds H0∝(2−β)/β instead; either way a shallower profile gives a smaller H0. Wucknitz 2002, abstract and Sec. 11
Worked example. Put images at R1=1.3′′ and R2=0.75′′. For each slope, the lens equation at the two images fixes θEγ′−1=(R1+R2)/(R12−γ′+R22−γ′). The exact Fermat-potential differences are:
Kochanek's two-term series matches every row to better than 0.02 percent. A slope uncertainty of ±0.05 therefore means about ±5 percent in H0, which is why lens modellers work so hard on the slope.
A nearby galaxy: convergence and shear
Lens galaxies have neighbours. Model one as another isothermal galaxy with Einstein radius bp at distance d from the lens, in direction φp. Its potential is ψp=bp∣θ−θp∣. Near the main lens, ∣θ∣≪d, so expand to second order. The constant does nothing. The first-order term is a uniform deflection, which just shifts where we think the source is. The second-order term is
To check it, put the neighbour on the x axis. Its second derivatives at the origin are ∂x2ψp=0 and ∂y2ψp=bp/d, so ψext=(bp/2d)y2. Writing y2=21θ2(1−cos2φ) gives the form above.
The first term is a uniform convergence, a thin sheet of mass. The second is shear: it stretches images along one axis and squeezes them along the other. With shear, the source-plane line where magnification diverges, the caustic, opens into a small diamond. A source inside the diamond has four bright images; outside it has two. The Lab's default neighbour has bp=0.75′′ at d=3.4′′, so κp=γp=0.11.
From a delay to the Hubble constant
The model predicts the Fermat-potential difference Δϕ. The measured delay then gives the time-delay distance, DΔt=cΔt/Δϕ, which is TDCOSMO IV's Eq. (8). Since DΔt∝1/H0 at fixed redshifts and matter density,
TDCOSMO IV, Eqs. (8) and (9)
Worked example. For the Lab's default quad, the model predicts 15.5 days between the first and last images when H0=70, using DΔt=2972 Mpc. Suppose the measured delay were 17.0 days. Then DΔt=2972×17.0/15.5=3260 Mpc and H0=70×2972/3260=63.8 km s−1 Mpc−1.
Worked example with a real lens. TDCOSMO IV lists DΔt=1947 Mpc for RXJ1131−1231, from a power-law model, with zd=0.295 and zs=0.654. A flat universe with H0=70 and Ωm=0.3 gives 2334 Mpc for those redshifts, so this lens alone says H0=70×2334/1947=83.9 before any correction for the line of sight. The method tutorial repeats this for all seven lenses.
The matter density barely matters here. Changing Ωm from 0.3 to 0.25 or 0.35 moves the seven-lens result by less than 0.3 percent, as the tutorial shows. A time-delay distance measures mainly H0.
Lab tasks
- In the quasar time-delay Lab, switch off the nearby galaxy and set the source about 0.1′′ from the centre. Move the steepness from 1.8 to 2.2 and note what moves and what stays.
- Turn the nearby galaxy back on and drag it around the lens. Watch the diamond-shaped caustic in the source-plane panel turn with it.
- Load "Wide double, long delay" and then "Einstein cross". Explain from the caustic panel why one has two images and the other four.
Exercises
- Derive α(θ) for the power law from α=(2/θ)∫0θκudu, and check that it reduces to θE for γ′=2.
- Show that ⟨κ⟩=(R1α1−R2α2)/(R12−R22), where αi=α(Ri).
- A modeller changes the slope of a lens from 2.00 to 2.10 and keeps the images fixed. By how much does the inferred H0 change?
- A neighbour with bp=0.5′′ sits 2.5′′ from the lens. Find κp and γp.
- A lens at zd=0.5, zs=2 has a model Fermat difference of 0.30 square arcseconds and a measured delay of 22.0 days. Find DΔt and H0, using 83.2 days per square arcsecond at H0=70.
AnswersShow the answer
- ∫0θ23−γ′θEγ′−1u2−γ′du=21θEγ′−1θ3−γ′. Multiplying by 2/θ gives θEγ′−1θ2−γ′. For γ′=2 the power of θ is zero and α=θE.
- From Exercise 1's method, θα(θ)=2∫0θκudu for any round lens. Subtracting at R1 and R2 gives 2∫R2R1κudu=R1α1−R2α2; divide by R12−R22.
- H0∝γ′−1, so it rises by the factor 1.10/1.00, about 10 percent.
- κp=γp=0.5/(2×2.5)=0.1.
- The predicted delay at H0=70 is 0.30×83.2=24.96 days. The measured 22.0 days is shorter, so DΔt=2972×22.0/24.96=2620 Mpc and H0=70×24.96/22.0=79.4 km s−1 Mpc−1.
Next: the mass-sheet degeneracy and what stellar speeds add.