Convergence and shear are second derivatives of the same potential, so in Fourier space the shear is just the convergence rotated by twice the angle of the wave vector. Rotating back turns a measured shear field into a convergence map. The rotation also yields a second, imaginary map that lensing leaves empty, a built-in null test. Two things are lost on the way: the mean of the map, because a uniform mass sheet changes no shape, and sharpness, because the map must be smoothed to keep its noise finite.
Write the potential as a sum of plane waves, ψ(θ)=∫(2π)2d2ℓψ^(ℓ)eiℓ⋅θ. Each derivative ∂j becomes a factor iℓj, so ψ,jk→−ℓjℓkψ^. From the definitions in the previous chapter,
The factor e2iβ has modulus one. Shear and convergence therefore have the same power at every scale; only the phase differs.
Inverting: the Kaiser-Squires map
Multiply by e−2iβ to undo the rotation:
κ^E+iκ^B=ℓ2(ℓ12−ℓ22)−2iℓ1ℓ2γ^(ℓ),ℓ=0.
Kaiser and Squires wrote the same result in real space, as a convolution of the shear with a kernel that falls off as 1/θ2. Weak Gravitational Lensing, Eqs. 5.3-5.4 In practice one bins galaxy ellipticities on a grid, takes a fast Fourier transform, multiplies by the kernel above and transforms back.
For a shear field that really comes from a real convergence, the imaginary part κB is zero. Measured shapes also carry noise, imperfect corrections for the telescope's blurring and, in principle, alignments that galaxies acquire on their own. These need not be curl-free, so they can fill κB. The B map is a null test: it should look like pure noise. Discriminating weak lensing from intrinsic spin correlations using the curl-gradient decomposition
Worked example. Take a single convergence wave along x, κ=Acosℓx. Its wave vector has β=0, so e2iβ=1 and γ1=Acosℓx, γ2=0. Where κ is positive, images stretch along x, across the crest; where it is negative, along y. A ridge of mass acts like a cylindrical lens and magnifies only across itself. Turn the wave to run along y. Now β=90∘ and e2iβ=−1, so γ1=−Acosℓy: the stretch turns with the wave and again runs across the crests.
The mode that cannot be recovered
At ℓ=0 the kernel is undefined. That mode is the mean convergence of the field, and it is lost because a uniform sheet makes no shear. More precisely, shapes measure g=γ/(1−κ), and the transformation
κ→κ′=λκ+(1−λ),γ→γ′=λγ
keeps g fixed for any λ=0: γ′/(1−κ′)=λγ/[λ(1−κ)]=g. This is the mass-sheet degeneracy. Shapes alone cannot tell the original mass from a weakened copy plus a uniform sheet. The magnification changes, μ′=μ/λ2, so galaxy sizes or number counts can break it. Weak Gravitational Lensing, Eqs. 5.9-5.10
A small-field Kaiser-Squires map simply sets the mean of the field to zero. Our reproducible tutorial checks this on a noise-free isothermal halo in a 40′ field. Away from the edges, the reconstruction differs from the truth by an almost constant offset of 0.0225. The true mean convergence of the field is 0.0220. After removing the offset, the residual is a tenth of the signal.
Noise and smoothing
With galaxies at random positions, the Kaiser-Squires estimate has infinite noise unless it is smoothed. Model each shear component as white noise. Galaxies with scatter σϵ per component and density n give a noise power of σϵ2/n in each component. The inversion kernel has modulus one, so κE and κB each inherit noise power σϵ2/n. Smooth with a Gaussian of width θs, whose Fourier window is e−ℓ2θs2/2. The noise variance of the map is
σκ2=∫(2π)2d2ℓnσϵ2e−ℓ2θs2=4πθs2nσϵ2.
Worked example. With σϵ=0.26, n=20arcmin−2 and θs=1′, σκ=0.26/4π×20=0.016. A simulation of pure Gaussian noise on a fine grid gives 0.0164 in the E map and 0.0162 in the B map. A cluster with θE=20′′ has κ=0.17 at 1′, but only the parts of it wider than about 1′ survive the smoothing. Halving θs doubles the noise.
Smoothing trades resolution for noise. Edges cost more. Shear outside the field is unknown and usually set to zero, so structure just beyond the edge leaks into the map. Masks inside the field do the same. Each map pixel near an edge is less certain than the formula says.
Lab tasks
In the weak-lensing mass map Lab, compare the E-mode and B-mode maps for the default field. Does the B map show either halo?
Load Too few galaxies. Predict from the noise formula how much worse the map should be at 4 galaxies per square arcminute than at 20, then compare the peak-over-noise readout.
Load A hidden mass sheet. The shapes are the same as before. Which readouts stay fixed and which move? Look at the galaxy sizes too.
Open the advanced settings and change the smoothing scale from 0.5 to 2 arcminutes. Describe what happens to the two halos and to the noise.
Exercises
Derive γ^=e2iβκ^ from the potential, and show that ∣γ^∣2=∣κ^∣2.
Show that under κ→λκ+(1−λ), γ→λγ, the magnification becomes μ/λ2.
A convergence wave runs at 45∘: κ=Acos[ℓ(x+y)/2]. Find γ1 and γ2.
What smoothing scale gives σκ=0.01 with σϵ=0.26 and n=20arcmin−2?
Rotate every measured ellipticity by 45∘, ϵ→iϵ. What happens to the E and B maps?
AnswersShow the answer
With ψ,jk→−ℓjℓkψ^: γ^1+iγ^2=−21(ℓ12−ℓ22+2iℓ1ℓ2)ψ^=−21(ℓ1+iℓ2)2ψ^ and κ^=−21ℓ2ψ^. The ratio is (ℓ1+iℓ2)2/ℓ2=e2iβ, of modulus one.
μ′−1=(1−κ′)2−∣γ′∣2=λ2(1−κ)2−λ2∣γ∣2=λ2μ−1.
β=45∘, so e2iβ=i: γ1=0 and γ2=Acos[ℓ(x+y)/2].
θs=0.26/(0.014π×20)=1.64′.
κE+iκB is linear in γ, so it is multiplied by i: the new E map is −κB and the new B map is κE. A 45∘ error in the shape convention moves all the mass into the null map, which is why the B map catches it.