Cosmic Shear

The idea in one sentence

Every photon you receive from a distant galaxy has travelled through gigaparsecs of intervening structure. Along the way, the gravitational potential of every overdensity it passed has bent its trajectory by a tiny amount. Bundles of nearby rays — that is, the rays that together form a single galaxy image — are bent slightly differently, so the image arrives at your detector subtly squashed and rotated compared with the galaxy's intrinsic shape. That percent-level distortion is called cosmic shear. It is weak gravitational lensing applied not to a special foreground cluster but to the universe at large.

The amplitude per galaxy is hopeless: a 1-3% ellipticity change is undetectable against the 30%-level intrinsic ellipticity scatter of real galaxies. But the distortions are coherent — neighbouring galaxies on the sky have been lensed by overlapping mass distributions and therefore share a common shear. Average a million galaxies in a patch of sky and the intrinsic randomness averages down, leaving the lensing signal exposed. Cosmic shear is statistics; it is mass measurement by carefully counting the shape of crowds.

A measurement four teams made at once

The theoretical possibility was clear from the 1960s. Kristian and Sachs (1966) and Gunn (1967) showed that any matter distribution along the line of sight bends image rays, in principle distorting background galaxies. Blandford and Narayan, then Miralda-Escudé, then Kaiser, then Bartelmann and Schneider laid out the modern formalism in the late 1980s and early 1990s — including the prediction that a CDM universe should produce roughly percent-level shear on degree scales.

The detection took a decade. Galaxy shape measurement is excruciating: the point-spread function of any real telescope is itself slightly elliptical and varies across the field, contaminating measured galaxy ellipticities. The required precision is parts per thousand. The breakthrough came in 2000, when four independent teams — Bacon, Refregier and Ellis using the William Herschel Telescope; Kaiser, Wilson and Luppino at Mauna Kea; Wittman, Tyson and collaborators using the CTIO 4m; and van Waerbeke, Mellier and the CFHT team — all reported coherent shear signals on degree scales at the few-percent level. The signal was within a factor of two of CDM predictions. Cosmology had a new probe.

The shear field, the convergence, the power spectrum

At each point on the sky the gravitational lensing of background galaxies is described by a 2×2 distortion matrix. Decompose it into a convergence κ (an isotropic magnification, related to projected mass) and a complex shear γ = γ_1 + iγ_2 (anisotropic stretching). For weak lensing, κ and γ are linear in the projected potential:

κ(θ) = (3/2) Ω_m (H_0/c)² ∫₀^χ_s  W(χ, χ_s) [δ(χθ, χ) / a(χ)] dχ
γ_i  = ∂_i ∂_j ψ                          (i, j project potential ψ)

Here δ is the matter overdensity, a(χ) the scale factor at comoving distance χ, and W(χ, χ_s) is the lensing kernel — the geometric weight given to a lens at χ for a source at χ_s. The kernel peaks roughly halfway between the observer and the source. The shear signal is therefore most sensitive to structure at z ≈ 0.3-0.7 for source galaxies at z ≈ 1.

What you measure is not γ at a point — that is unobservable for a single galaxy — but its two-point correlation function. Define the projected shear correlations ξ_+(θ) and ξ_-(θ):

ξ_±(θ) = ⟨ γ_t γ_t ⟩ ± ⟨ γ_× γ_× ⟩       (averaged over pairs of galaxies separated by θ)

γ_t and γ_× are the tangential and cross components relative to the line joining each pair. The full information lives in the convergence power spectrum C_κ(ℓ), obtained as a weighted line-of-sight integral of the matter power spectrum P(k):

C_κ(ℓ) = ∫₀^χ_s [W(χ)² / χ²] P(ℓ/χ, z(χ)) dχ

Modern surveys fit ξ_±, C_κ, or the band-power equivalent simultaneously across multiple redshift bins and report the joint cosmological parameter constraints.

S_8 — the one number cosmic shear measures best

Inside the lensing kernel, the matter power spectrum P(k, z) is roughly proportional to σ_8², the rms density fluctuation on 8 h⁻¹ Mpc scales, with a normalisation that also depends on Ω_m. Weak lensing data are not separately sensitive to σ_8 and Ω_m — a denser, smoother universe produces about the same shear as a sparser, more clumpy one. The combination that lies along the short axis of the joint constraint ellipse is

S_8 ≡ σ_8 (Ω_m / 0.3)^(1/2)

This is the parameter every modern shear paper headlines. Roughly speaking, S_8 sets the amplitude of structure today on ~10 Mpc scales. A higher S_8 means a more clumpy, lensing-effective universe.

Three eras of cosmic shear surveys

Survey sensitivity scales with the number of galaxies with usable shape measurements and the area covered. The field has moved through three eras.

Survey	Era	Area	N_gal usable	S_8 reported
CFHTLenS	2012	154 deg²	~ 4.7 × 10⁶	0.74 ± 0.04
KiDS-450	2017	450 deg²	~ 1.5 × 10⁷	0.74 ± 0.04
DES Y1	2018	1321 deg²	~ 2.6 × 10⁷	0.78 ± 0.02
KiDS-1000	2021	1006 deg²	~ 2.1 × 10⁷	0.759 ± 0.024
HSC Y3	2023	416 deg²	~ 2.5 × 10⁷	0.776 ± 0.033
DES Y3 (3×2pt)	2022	4143 deg²	~ 1.0 × 10⁸	0.776 ± 0.017
Euclid Wide	~ 2026 first data	14 000 deg²	~ 1.5 × 10⁹	(forecast σ ≈ 0.003)
LSST / Rubin Y10	~ 2034	18 000 deg²	~ 3 × 10⁹	(forecast σ ≈ 0.003)

The takeaway: three independent ground-based experiments — KiDS, DES, HSC — agree on S_8 ≈ 0.76 with errors of a few percent. The forthcoming space and Rubin samples will tighten that order-of-magnitude further.

The S_8 tension with the CMB

Run Planck's measurement of the CMB at z ≈ 1100 forward through ΛCDM to today, and you get a prediction for the amplitude of large-scale structure now. That prediction is

S_8 (Planck 2018, ΛCDM) = 0.834 ± 0.016

The lensing surveys all sit roughly 2-3σ low compared with that:

S_8 (KiDS-1000)       = 0.759 ± 0.024     ~ 2.7σ low
S_8 (DES Y3 3×2pt)    = 0.776 ± 0.017     ~ 2.4σ low
S_8 (HSC Y3)          = 0.776 ± 0.033     ~ 1.7σ low

Individually, none would be alarming. Together — three independent shear pipelines, different telescopes, different photo-z calibrations, all pointing the same direction — they motivate the "S_8 tension". The late-time universe appears measurably less clumpy than CMB-extrapolated ΛCDM predicts.

Possible explanations break into three buckets. First, systematics in shear: residual PSF anisotropy, mis-calibrated photo-z's, or under-modelled intrinsic alignments could all pull S_8 low. Second, baryonic feedback: hydrodynamic effects (AGN feedback, supernovae) suppress small-scale matter clustering by 5-15% at k ≈ 1 h/Mpc, biasing inferred S_8 if unaccounted for; modern pipelines marginalise over baryonic feedback nuisance parameters. Third, new physics: massive neutrinos, dark sector interactions, modified gravity, or warm dark matter could all genuinely suppress late-time structure relative to ΛCDM.

Tomographic shear — the third dimension

The lensing kernel for a source at z = 1 differs from the kernel for a source at z = 0.3: the latter sees less intervening structure. By splitting the galaxy sample into multiple redshift bins (using photometric redshifts inferred from broad-band colours) and measuring the correlation function within each bin and between pairs of bins, you recover not just a 2D snapshot of projected mass but its 3D evolution.

For N tomographic bins, you have N(N+1)/2 independent ξ_+ functions and the same number of ξ_-. KiDS uses 5 bins; DES uses 4; HSC uses 4; LSST Y10 will use 10 or more. Tomography sharpens cosmological inference dramatically: σ_8 and Ω_m can be separately constrained, the equation of state w of dark energy is teased out via its effect on the growth rate of structure with redshift, and the sum of neutrino masses ∑m_ν has a measurable impact on small-scale power suppression.

The three big systematics

Cosmic shear is statistics about percent-level distortions of galaxy shapes. Every step from photons to S_8 must be controlled at the parts-per-thousand level. Three systematics dominate the modern analyses.

• Galaxy shape measurement. The point-spread function of any telescope is itself slightly anisotropic and varies across the focal plane. Bright stars in the same images are used to model the PSF; the galaxy shapes are then deconvolved. Modern pipelines (metacalibration, lensfit, Bayesian Fourier-domain methods) reach shear calibration biases of ~10⁻³. Stage-IV surveys need 10⁻⁴.
• Intrinsic alignments (IAs). Galaxies are not intrinsically randomly oriented: tidal forces during their formation align them coherently with the large-scale density gradient, mimicking lensing shear. The "II" (intrinsic-intrinsic) and "GI" (gravitational-intrinsic) IA correlations bias S_8 at the 5-15% level if ignored. The NLA (non-linear alignment) model and the TATT (tidal alignment + tidal torquing) extension fit IA amplitudes simultaneously with the cosmological signal. Bright red galaxies show stronger IAs than blue spirals; modern analyses split samples by colour to constrain the IA contribution.
• Photo-z calibration. Photometric redshifts have typical biases of σ_z / (1+z) ~ 0.02-0.05 and small but non-negligible mean offsets. A 0.01 shift in the mean redshift of a bin moves the lensing kernel and biases S_8 at the percent level. Spectroscopic samples (zCOSMOS, VIPERS, GAMA, DESI's bright sample) anchor the photo-z distributions; cross-correlation with spectroscopic samples ("clustering redshifts") provides an independent calibration.

Stage IV — Rubin, Euclid, Roman

The current Stage III surveys (KiDS, DES, HSC) measure shear over a few thousand square degrees with ~10⁸ galaxies. The Stage IV surveys coming online now and in the late 2020s push to 10⁹-galaxy samples over a substantial fraction of the sky.

• Vera Rubin Observatory / LSST. 8.4m wide-field optical telescope on Cerro Pachón, Chile. Survey first light expected 2026-2027. 10-year survey will produce six-band ugrizy imaging of 18,000 deg² to r ≈ 27, with ~3 × 10⁹ usable galaxy shapes. The DESC weak-lensing collaboration's forecast S_8 precision is ~0.003, an order of magnitude beyond KiDS-1000.
• Euclid. ESA space telescope, launched July 2023, primary mission lifetime 6 years. 14,000 deg² wide survey in optical (visible) and near-infrared (Y, J, H). Sub-arcsecond space-based PSF removes the dominant ground-based systematic. ~1.5 × 10⁹ usable galaxy shapes. First data release containing cosmic-shear analyses expected 2026.
• Nancy Grace Roman Space Telescope. NASA mission, launch ~2027. Deep near-infrared survey complementing Euclid; particularly powerful for the high-z (z > 1) shear regime where ground-based seeing degrades shape measurement most.

The combination of these surveys will push S_8 errors below 0.5%. The current 2-3σ tension will either become a 10σ measurement of new physics or evaporate as systematics are better controlled. Either outcome reshapes cosmology.

Cosmic shear in context

Cosmic shear is one of several "late-time" probes of the matter distribution, alongside galaxy clustering, redshift-space distortions, cluster counts, and the integrated Sachs-Wolfe effect. Modern analyses combine them — DES Y3's "3×2pt" likelihood, for example, fits cosmic shear, galaxy clustering, and galaxy-galaxy lensing simultaneously. The strength of combined analyses is that systematics that affect one observable (e.g. galaxy bias) drop out of others, breaking degeneracies and cross-checking results.

Crucially, shear is most directly sensitive to total matter — dark plus baryonic — not just galaxies. This is why shear is sometimes called "weighing the dark sector": it does not depend on assumptions about how galaxies trace the underlying mass.

Common pitfalls

• Treating shear like strong lensing. A single sheared galaxy tells you nothing — the intrinsic ellipticity drowns the 1-3% signal. Cosmic shear is statistics, not imaging. The science output is correlation functions and power spectra, not pretty arcs.
• Conflating σ_8 with S_8. Shear measures S_8 ≡ σ_8 (Ω_m/0.3)^(1/2) much more tightly than it measures σ_8 alone. Reports quoting only σ_8 obscure the precision; reports quoting only S_8 hide which σ_8-Ω_m combination is being constrained. Both should appear with their covariance.
• Ignoring intrinsic alignments. Pre-2010 analyses sometimes treated galaxy shapes as intrinsically uncorrelated. Tidal alignment of red galaxies produces 5-15% spurious shear signal; ignoring it biases S_8 high (II) or low (GI). Every modern pipeline marginalises over an NLA or TATT model.
• Trusting one survey. A 2σ shift in any single experiment is unremarkable. The S_8 tension is interesting precisely because three independent ground-based surveys, with different telescopes and pipelines, agree. Wait for at least two independent confirmations before believing any single shear result.
• Forgetting the baryonic correction. Pure-dark-matter N-body simulations overpredict P(k) at k ≳ 1 h/Mpc by 5-15% because they omit AGN feedback. If you fit ΛCDM to shear data without marginalising over baryonic feedback you will infer a low S_8. Modern analyses use halo-model corrections, BCM (baryonification), or hydrodynamic emulators.