Skip to main content
Sandbox Physics
Redshift-space distortions

Growth or a distance error?

Sandbox Physics ·

A wrong distance scale and real galaxy motion both make the map anisotropic, and over a narrow range of scales they can look alike. They are told apart by how the anisotropy depends on scale and angle: infall gives a fixed ratio of quadrupole to monopole, random motion damps the line-of-sight power more and more at high wavenumber, and a distance error follows the slope of the power spectrum. A fit that allows all three recovers the growth rate on average, but one small survey volume can still be fooled.

This chapter follows Multipoles of a squashed map. You need first-order Taylor expansion and the multipoles from that chapter. By the end you should be able to derive how a distance error changes the power spectrum, show why a small error mimics infall, and read the Lab's three-way fit, including its failure cases.

What an assumed distance does

To place galaxies in a box, a survey converts redshift into distance with an assumed expansion history. Suppose the assumed distances along the line of sight are ss times the true ones, while distances across the sky are right. The parameter ss is the Lab's distance stretch. Real analyses let the two directions err separately, and the overall scale shift then moves features in wavenumber. Only the ratio changes the shape of the map, so one parameter is enough here.

In the assumed coordinates every line-of-sight separation is ss times too long, so every line-of-sight wavenumber is ss times too short: k∥true=s k∥k_\parallel^{\mathrm{true}}=s\,k_\parallel. The true wavenumber and angle that correspond to an observed (k,μ)(k,\mu) are

kt=k1+(s2−1)μ2,μt=sμ1+(s2−1)μ2.k_{\mathrm t}=k\sqrt{1+(s^2-1)\mu^2},\qquad\mu_{\mathrm t}=\frac{s\mu}{\sqrt{1+(s^2-1)\mu^2}}.

Power has units of volume, and the assumed volume is ss times the true one, so

Pobs(k,μ)=s Ps(kt,μt).P_{\mathrm{obs}}(k,\mu)=s\,P_s\bigl(k_{\mathrm t},\mu_{\mathrm t}\bigr).

Ballinger, Peacock and Heavens wrote the same mapping with a flattening factor FF. For the stretch defined here, F=1/sF=1/s. Ballinger, Peacock and Heavens 1996, §3

Why a small error looks like infall

Take no motion at all and a small stretch, s=1+ϵs=1+\epsilon. Then kt2≈k2(1+2ϵμ2)k_{\mathrm t}^2\approx k^2(1+2\epsilon\mu^2). Near any kk the power spectrum behaves locally like knk^n, with local slope n=dln⁡P/dln⁡kn=d\ln P/d\ln k, so P(kt)≈P(k)(1+nϵμ2)P(k_{\mathrm t})\approx P(k)(1+n\epsilon\mu^2). To first order,

Pobs(k,μ)P(k)≈1+ϵ+nϵμ2.\frac{P_{\mathrm{obs}}(k,\mu)}{P(k)}\approx1+\epsilon+n\epsilon\mu^2.

Compare the Kaiser form (1+βμ2)2≈1+2βμ2(1+\beta\mu^2)^2\approx1+2\beta\mu^2. The μ2\mu^2 terms match when

βa=nϵ2=−n2(F−1).\beta_{\mathrm a}=\frac{n\epsilon}{2}=-\frac{n}{2}(F-1).

This is equation 21 of Ballinger, Peacock and Heavens. On the scales used here the slope is negative: the course reference finds n=−1.32n=-1.32 at k=0.08k=0.08 and n=−1.93n=-1.93 at k=0.15 h Mpc−1k=0.15\,h\,\mathrm{Mpc^{-1}}. So distances that are too long, ϵ>0\epsilon>0, mimic negative infall, which looks like fingers. Distances that are too short mimic extra growth. The course reference checks the formula against the full mapping at s=1.01s=1.01 and finds agreement to better than 0.7% at both wavenumbers.

Three fingerprints

The three causes depend on scale and angle in different ways.

Table 4.1 How each cause changes the measured multipoles.
CauseMonopole over real-space powerQuadrupole as kk growsAngle dependence
Coherent infall1+2β/3+β2/51+2\beta/3+\beta^2/5, constantconstant ratio to monopolefixed μ2\mu^2 and μ4\mu^4 mix
Random motionbelow one, fallingnegative, growing quadraticallyall orders in μ2\mu^2
Distance stretchabout ssfollows the local slope n(k)n(k)set by ss and nn

The table assumes you can divide by the real-space power, as the Lab does with its synthetic galaxies. Ballinger, Peacock and Heavens point to the angle dependence: growth produces a characteristic ratio between the μ2\mu^2 and μ4\mu^4 terms, which a geometric distortion in general does not share. They also warn that the differences are subtle and that a large survey is needed to measure the two effects separately. Ballinger, Peacock and Heavens 1996, §3

Worked example: a distance error read as fingers

Choose Wrong distance: stretched. The galaxies do not move at all; only the assumed distances are 12% too long. The Lab measures a monopole of 1.0761.076 and a quadrupole of −0.067-0.067 in the first bin, with the quadrupole falling to −0.136-0.136 by k=0.13 h Mpc−1k=0.13\,h\,\mathrm{Mpc^{-1}}.

The joint fit over growth, random motion and stretch returns s=1.115s=1.115, no growth and no random motion, close to the truth. Now hold the stretch at one, which is what you do if you trust the distances. The best fit becomes f=0f=0 with σv=211 km s−1\sigma_v=211\ \mathrm{km\,s^{-1}}: random motion that does not exist. Its squared residual is 0.055, about 200 times worse than the free fit, which is the warning sign.

The opposite error works the same way. With distances 10% too short, holding the stretch at one gives a growth rate of 0.100.10 from galaxies that stand still.

How well does the fit work?

A single box can mislead, so the course reference repeats the analysis on 12 independent synthetic boxes, each 160 h−1Mpc160\,h^{-1}\mathrm{Mpc} on a side, with known settings.

Table 4.2 Mean and box-to-box scatter of the joint fit over 12 synthetic boxes.
InjectedRecovered growth rateRecovered stretchRecovered σv\sigma_v
f=0.76f=0.76, σv=300\sigma_v=300, s=1s=10.77±0.460.77\pm0.461.02±0.061.02\pm0.06312±107312\pm107
f=0.76f=0.76, no random motion0.77±0.450.77\pm0.451.02±0.061.02\pm0.0691±13691\pm136
f=0f=0, σv=400\sigma_v=4000.02±0.010.02\pm0.010.996±0.0090.996\pm0.009403±8403\pm8
f=0f=0, s=1.12s=1.12001.115±0.0011.115\pm0.00119±1019\pm10

Velocities are in km s−1\mathrm{km\,s^{-1}} and the numbers after ±\pm are the scatter between boxes. Averaged over boxes, the growth rate is recovered without visible bias. One box is a different story: the growth-rate scatter of 0.460.46 is more than half the true value, because growth and stretch trade off. Seed 7 returns f=2.0f=2.0 at the edge of the fit grid with σv=597 km s−1\sigma_v=597\ \mathrm{km\,s^{-1}}. When the scatter of the other 11 boxes is used as the error bar for the held-out one, the true growth rate lies within one error bar in 83% of boxes and within two in 92%.

The stretch has a small systematic offset: an injected 1.121.12 comes back as 1.1151.115. That offset is larger than the box-to-box scatter of 0.0010.001, so a naive error bar covers the truth in only 8% of boxes. It comes from the finite fit grid and the bin centres.

Failure cases

The course reference keeps four failures alongside the successes.

  1. A distance error fitted with distances held fixed is read as random motion, as in the worked example.
  2. Inverting the Kaiser quadrupole-to-monopole ratio on the largest scales, without a damping term, gives f=0.68f=0.68 for an injected 0.760.76, because the fingers lower the quadrupole even there. The joint fit gives 0.720.72.
  3. Fitting coherent infall out to k=0.36 h Mpc−1k=0.36\,h\,\mathrm{Mpc^{-1}} absorbs the breakdown of the linear model into a false stretch, s=0.965s=0.965. Inside the fixed limit of 0.23 h Mpc−10.23\,h\,\mathrm{Mpc^{-1}} the stretch is 1.0001.000.
  4. One box can miss the growth rate badly, as seed 7 shows.

From the Lab to a real survey

Real surveys cannot divide by the real-space power of the same galaxies. They fit the amplitude and shape of the redshift-space power, so the bias bb and the growth rate ff become entangled with the overall clustering amplitude. Percival and White showed that the anisotropy constrains the combination fσ8f\sigma_8, the growth rate times the amplitude of matter fluctuations, independently of galaxy bias on large scales. That is why survey papers quote fσ8f\sigma_8 rather than ff or β\beta. Percival and White 2009, abstract and §2

A measured squashing has to be checked against a distance error, against random motion and against the breakdown of the linear model before it is read as a growth rate.

Lab tasks

  1. Choose Wrong distance: stretched. Read the joint fit and the fit with distances fixed, and compare the random motion each one reports.
  2. Choose Wrong distance: squashed and find the false growth rate in the fixed-distance readout.
  3. Start from the default and set the stretch to 1.051.05. Note how the fitted growth rate and stretch move together.
  4. Choose Faster growth and check whether the fit separates it from a squashing distance error.

Exercises

  1. Use βa=nϵ/2\beta_{\mathrm a}=n\epsilon/2 with n=−1.5n=-1.5 and s=1.04s=1.04. What apparent β\beta does the distance error produce?
  2. Why is the Kaiser quadrupole-to-monopole ratio independent of kk, while the distance stretch is not?
  3. Why can the monopole tell a stretch from random motion in the Lab, but not in a real survey?
  4. The growth-rate scatter per box is 0.4550.455. If scatter falls as one over the square root of the volume, how much volume gives a scatter of 0.050.05?
Answers

For the first exercise, βa=−1.5×0.04/2=−0.030\beta_{\mathrm a}=-1.5\times0.04/2=-0.030. The exact relation n(s2−1)/4n(s^2-1)/4 gives −0.031-0.031. A 4% distance error is equivalent to a growth rate change of about 0.060.06 for b=2b=2.

For the second, the Kaiser factors are pure numbers that multiply P(k)P(k), so they cancel in the ratio. A stretch moves power between wavenumbers, so its effect depends on the local slope n(k)n(k), which changes with scale.

For the third, a stretch multiplies the power by ss and random motion divides it by a damping factor, so the monopole moves in opposite directions. The Lab divides by the known real-space power. A survey does not know the real-space power, so the overall level of the monopole is absorbed into the bias and the clustering amplitude.

For the fourth, the volume must grow by (0.455/0.05)2=83(0.455/0.05)^2=83, to about 3.4×108 h−3Mpc33.4\times10^8\,h^{-3}\mathrm{Mpc^3}, a cube about 700 h−1Mpc700\,h^{-1}\mathrm{Mpc} on a side. This assumes the same fit range and the same galaxy density.

Sources

  1. Ballinger, Peacock and Heavens 1996, §3
  2. Percival and White 2009, abstract and §2

Keep reading

What this page links to directly

See this series on the content map →

Redshift-space distortions · Reading path

From a moving galaxy to a growth measurement

  1. ArticleWhy do galaxy maps point at us?
  2. ArticleHow a redshift moves a galaxy
  3. ArticleMultipoles of a squashed map
  4. You are hereGrowth or a distance error?
  5. ArticleMeasure multipoles from a frozen catalogue
  6. ArticleHow Guzzo et al. measured growth from redshift distortions in 2008
Something broken, confusing, or missing on this page?Tell us
What kind of feedback?

Sent with this page’s address, its Lab ID, the site language and your device type. Please leave out personal details. Feedback is kept for up to 12 months.Privacy policy