---
title: "Shear correlation functions and redshift tomography"
description: "Define the two shear correlation functions, estimate their shape-noise error, weight lens planes by distance and split sources into redshift bins."
canonical_url: "https://sandboxphysics.com/stories/weak-lensing/shear-correlations-and-tomography/"
language: "en"
page_type: "story"
alternate_languages:
  zh-CN: "https://sandboxphysics.com/zh/stories/weak-lensing/shear-correlations-and-tomography/"
last_modified: "2026-11-14"
---

[Weak gravitational lensing and mass maps](https://sandboxphysics.com/stories/weak-lensing/)

# Shear correlation functions and redshift tomography

Sandbox Physics · 2026-11-14

Large-scale structure stretches every galaxy by about one percent, too little to map lump by lump. What can be measured is how similar the stretches of two galaxies are as a function of their separation. Two such correlation functions, $\xi_+$ and $\xi_-$, carry all the two-point information, and each is a Bessel transform of the convergence power spectrum. Splitting the source galaxies by redshift adds a second axis: nearer and farther galaxies weight the same lens planes differently, so their auto- and cross-correlations sort the mass by distance.

Read [turning a shear field into a mass map](https://sandboxphysics.com/stories/weak-lensing/from-shear-to-a-mass-map/) first. This chapter uses averages over pairs, the Bessel functions $J_0$ and $J_4$ as given functions, and comoving distance.

## Two correlation functions

Pick two galaxies separated by $\theta$ and project both shears onto the line joining them, as tangential and cross components. Average over all pairs at that separation:

$$
\xi_+(\theta)=\langle\gamma_{\rm t}\gamma_{\rm t}\rangle+\langle\gamma_\times\gamma_\times\rangle,\qquad \xi_-(\theta)=\langle\gamma_{\rm t}\gamma_{\rm t}\rangle-\langle\gamma_\times\gamma_\times\rangle .
$$

In complex form, with $\phi$ the direction of the separation, $\xi_+=\langle\gamma\gamma^*\rangle$ and $\xi_-=\mathrm{Re}\,\langle\gamma\gamma\,e^{-4i\phi}\rangle$. A mirror reflection flips the sign of $\gamma_\times$, so in a mirror-symmetric universe $\langle\gamma_{\rm t}\gamma_\times\rangle=0$. [Cosmology with cosmic shear observations: a review, Eqs. 31-32](https://arxiv.org/abs/1411.0115v2)

Both functions follow from the convergence power spectrum. Splitting it into an E part, made by lensing, and a B part, which lensing does not make,

$$
\xi_\pm(\theta)=\frac1{2\pi}\int_0^\infty\mathrm d\ell\,\ell\,J_{0,4}(\ell\theta)\left[P_E(\ell)\pm P_B(\ell)\right],
$$

with $J_0$ for $\xi_+$ and $J_4$ for $\xi_-$. [Cosmology with cosmic shear observations: a review, Eq. 35](https://arxiv.org/abs/1411.0115v2) $J_0$ is broad, so $\xi_+$ collects power from scales larger than $\theta$. $J_4$ starts slowly, so $\xi_-$ is dominated by scales smaller than $\theta$. The sign of $P_B$ differs between them, which is what B-mode tests exploit.

## Measuring them and the shape-noise floor

The estimator replaces shears with measured ellipticities, $\hat\xi_\pm=\sum_{ij}(\epsilon_{{\rm t},i}\epsilon_{{\rm t},j}\pm\epsilon_{\times,i}\epsilon_{\times,j})/N_{\rm pairs}$, summed over pairs whose separation falls in a bin. Intrinsic shapes are uncorrelated between different galaxies, so they average out of the mean but not out of the scatter. Each product of two noise terms has variance $\sigma_\epsilon^4$, and $\hat\xi_+$ sums two such products per pair:

$$
\sigma(\hat\xi_\pm)\approx\frac{\sqrt2\,\sigma_\epsilon^2}{\sqrt{N_{\rm pairs}}} .
$$

This is the shape-noise floor. On top of it comes sample variance: one field contains only a finite number of structures.

**Worked example.** The [reproducible tutorial](https://sandboxphysics.com/stories/weak-lensing/recompute-shear-correlations/) uses a $256'\times256'$ field with 10 galaxies per square arcminute in each redshift bin, 655 360 galaxies. Pairs separated by $6.3'$ to $10'$ number about $N\,n\,\pi(10^2-6.3^2)/2=6.2\times10^8$. With $\sigma_\epsilon=0.26$, the floor is $3.8\times10^{-6}$ per field and $7.8\times10^{-7}$ for the mean of 24 fields. The measured standard error of the mean is $1.2\times10^{-6}$ for $\xi_+$ and $8.7\times10^{-7}$ for $\xi_-$. $\xi_+$ is the broader filter and carries more sample variance.

## Which lens planes a source feels

In a flat universe, the convergence of a source at comoving distance $\chi_{\rm s}$ is a weighted integral of the density contrast $\delta$ along the line of sight:

$$
\kappa(\boldsymbol\theta)=\frac{3H_0^2\Omega_{\rm m}}{2c^2}\int_0^{\chi_{\rm s}}\mathrm d\chi\,(1+z)\,\frac{\chi(\chi_{\rm s}-\chi)}{\chi_{\rm s}}\,\delta(\chi\boldsymbol\theta,\chi).
$$

[Cosmology with cosmic shear observations: a review, Eqs. 21-24](https://arxiv.org/abs/1411.0115v2)

The weight $\chi(\chi_{\rm s}-\chi)/\chi_{\rm s}$ is a parabola. It vanishes at the observer and at the source, and peaks halfway in comoving distance. Mass behind the source gets no weight at all. For thin lens planes the integral becomes a sum with one weight $W_{ip}$ per source bin $i$ and plane $p$.

## Tomography

Split sources into redshift bins and correlate every pair of bins. If the planes are statistically independent with the same power spectrum $P$, the correlation between bins $i$ and $j$ is

$$
\xi^{ij}_\pm(\theta)=\Bigl(\sum_pW_{ip}W_{jp}\Bigr)\,\xi_\pm^{\rm plane}(\theta).
$$

**Table 4.1** Plane weights in the tutorial, flat universe with $\Omega_{\rm m}=0.3$, normalised to the largest entry.

| Source bin | Plane at $z=0.25$ | Plane at $z=0.5$ | Plane at $z=0.8$ |
| --- | --- | --- | --- |
| $z_{\rm s}=0.6$ | 0.485 | 0.289 | 0 |
| $z_{\rm s}=1.2$ | 0.654 | 1.000 | 0.922 |

The near bin does not see the plane at $z=0.8$. The tomographic weights are $0.319$ for the near bin, $2.277$ for the far bin and $0.606$ for the cross-correlation. The far bin's correlation is 7.1 times stronger. The cross-correlation is not zero, because both bins see the two nearer planes: its correlation coefficient is $0.606/\sqrt{0.319\times2.277}=0.71$. The growth of the correlations with source redshift carries information about distances and about the growth of structure.

## What tomography is sensitive to

Every step that assigns a galaxy to a bin or turns its shape into a shear now matters.

- Redshift errors. If 20 percent of the far galaxies are wrongly put in the near bin, the near-bin weight becomes $(0.833^2\times0.319)+(2\times0.833\times0.167\times0.606)+(0.167^2\times2.277)=0.453$, 42 percent above the nominal value. The tutorial's measurement agrees with this prediction and rejects the nominal one.
- Shear calibration. A multiplicative bias $m$ turns $\gamma$ into $(1+m)\gamma$ and $\xi^{ij}$ into $(1+m_i)(1+m_j)\xi^{ij}$.
- Intrinsic alignments. Physically close galaxies can be aligned by the same tidal field, adding a positive term in the same bin. The tidal field that aligns near galaxies also lenses far ones, giving a negative term that survives in cross-correlations. In a linear alignment model, Hirata and Seljak estimated that this second term can suppress the lensing power of a single broad bin by about 10 percent at $z_{\rm s}=1$ and 30 percent at $z_{\rm s}=0.5$. [Intrinsic alignment-lensing interference as a contaminant of cosmic shear](https://arxiv.org/abs/astro-ph/0406275v3)

## Lab tasks

1. In [the weak-lensing mass map Lab](https://sandboxphysics.com/labs/weak-lensing-mass-map/), drag the source plane in the side view from redshift 2.5 down to 0.4. Watch the lensing efficiency curve and the two halo points on it.
2. Load Mass behind the sources. Which halo vanishes from the mass map, and what does the efficiency plot say about it?
3. Drag the far halo to redshift 0.9 with the sources at 1.2, then to 0.3. Predict which position gives a larger true Einstein radius, then check.

## Exercises

1. Show that rotating every ellipticity by $45^\circ$, $\epsilon\to i\epsilon$, leaves $\xi_+$ unchanged and flips the sign of $\xi_-$.
2. Ignoring the factor $1+z$, show that the weight $\chi(\chi_{\rm s}-\chi)/\chi_{\rm s}$ peaks at $\chi=\chi_{\rm s}/2$, and find its peak value.
3. Using Table 4.1, check the three tomographic weights.
4. If both bins have an unknown $m=-0.02$ that is ignored, by how much is every $\xi^{ij}$ underestimated?
5. A survey has $\sigma_\epsilon=0.26$ and $5.9$ galaxies per square arcminute over $4143$ square degrees. Estimate the shape-noise floor of $\xi_+$ in a bin from $6.3'$ to $10'$, using all sources as one bin.

Answers

1. $\xi_+=\langle\gamma\gamma^*\rangle$ and $(i\gamma)(i\gamma)^*=\gamma\gamma^*$, so $\xi_+$ is unchanged. $\xi_-=\mathrm{Re}\langle\gamma\gamma e^{-4i\phi}\rangle$ and $(i\gamma)(i\gamma)=-\gamma\gamma$, so $\xi_-$ changes sign. A rotation by $45^\circ$ turns E modes into B modes.
2. $\mathrm d/\mathrm d\chi\,[\chi(\chi_{\rm s}-\chi)]=\chi_{\rm s}-2\chi=0$ at $\chi=\chi_{\rm s}/2$. The peak value is $(\chi_{\rm s}/2)^2/\chi_{\rm s}=\chi_{\rm s}/4$.
3. Near bin: $0.485^2+0.289^2=0.319$. Far bin: $0.654^2+1^2+0.922^2=2.277$. Cross: $0.485\times0.654+0.289\times1+0=0.606$.
4. $(1-0.02)^2=0.960$: every correlation is 4 percent low.
5. The area is $4143\times3600=1.49\times10^7$ square arcminutes and $N=8.8\times10^7$ galaxies. Pairs: $N\,n\,\pi(10^2-6.3^2)/2=8.8\times10^7\times5.9\times94.7=4.9\times10^{10}$. The floor is $\sqrt2\times0.26^2/\sqrt{4.9\times10^{10}}=4.3\times10^{-7}$. A real survey splits these galaxies among bins and loses pairs to masks and weights, so this is a lower limit.

The tutorial works with independent Gaussian planes and one source redshift per bin. A real analysis uses a full redshift distribution per bin, a nonlinear matter power spectrum and correlated structure along the line of sight. The [guide to the DES Year 3 analysis](https://sandboxphysics.com/stories/weak-lensing/reading-des-y3-lensing-and-clustering/) shows how one survey handles each of these.

## Try it in the Lab

[Weak-lensing mass map](https://sandboxphysics.com/labs/weak-lensing-mass-map/): Move the source plane past a lens and read how the lensing efficiency falls to zero.

## Sources

1. [Cosmology with cosmic shear observations: a review, Eqs. 31-32](https://arxiv.org/abs/1411.0115v2)
2. [Intrinsic alignment-lensing interference as a contaminant of cosmic shear](https://arxiv.org/abs/astro-ph/0406275v3)

## Keep reading

Weak gravitational lensing and mass maps · Reading path

### From galaxy shapes to a map of mass

1. [Article **How can galaxies that are barely stretched map dark matter?**](https://sandboxphysics.com/stories/weak-lensing/galaxy-shapes-weigh-dark-matter/)
2. [Article **Convergence, shear and what a galaxy shape measures**](https://sandboxphysics.com/stories/weak-lensing/convergence-shear-and-galaxy-shapes/)
3. [Article **Turning a shear field into a mass map**](https://sandboxphysics.com/stories/weak-lensing/from-shear-to-a-mass-map/)
4. [You are here **Shear correlation functions and redshift tomography**](https://sandboxphysics.com/stories/weak-lensing/shear-correlations-and-tomography/)
5. [Article **Recompute shear correlations and tomography from a simulation**](https://sandboxphysics.com/stories/weak-lensing/recompute-shear-correlations/)
6. [Article **How to read the DES Year 3 lensing and clustering analysis**](https://sandboxphysics.com/stories/weak-lensing/reading-des-y3-lensing-and-clustering/)
