The σ8 Parameter and Amplitude of Clustering

The matter power spectrum $P(k)$ is the master statistic of cosmic structure, but it bundles two statements together: a shape and an amplitude. The shape is handed down by the primordial power spectrum and reworked by the transfer function — recalled by name from aq-astro-primordial-power-spectrum and aq-astro-transfer-function-matter-power-spectrum — fixing the relative power on each scale. What it cannot fix is the overall loudness: how much structure the Universe actually contains. That single missing number must be pinned to an observation, and the parameter that carries it is $\sigma_8$ — the RMS of the linearly evolved matter density contrast, smoothed over spheres of comoving radius $8\,h^{-1}\,\mathrm{Mpc}$ . Three ideas organize everything below. First, $\sigma_8$ is not a fluctuation "at 8 Mpc" but a window-weighted integral over all of $P(k)$ , so it inherits the whole $\Lambda$ CDM parameter set — a scalar standing in for an entire curve. Second, because linear growth is scale-independent for sub-horizon matter (from aq-astro-linear-growth-of-structure), the amplitude on every comoving scale evolves by one common factor, so $\sigma_8(z)=\sigma_8\,D(z)/D(0)$ becomes a clock timing the assembly of structure. Third, low-redshift probes cannot separate amplitude from matter density and instead measure $S_8=\sigma_8\sqrt{\Omega_m/0.3}$ — the axis along which the " $S_8$ tension" between the Planck CMB and weak-lensing surveys lives. This article is the amplitude head of the structure-formation chapter: it takes the shape as given and asks what the normalization means and why one number does so much work.

Key takeaways

$\sigma_8$ is the RMS fractional matter density fluctuation inside randomly placed spheres of comoving radius $8\,h^{-1}\,\mathrm{Mpc}$ , computed from the linear power spectrum extrapolated to $z=0$ .
It fixes the overall amplitude (normalization) of $P(k)$ , the piece the primordial spectrum and transfer function leave undetermined.
Mathematically it is a top-hat-window-weighted integral of $P(k)$ , so its value folds in $\Omega_m$ , $h$ , the baryon fraction, and the tilt $n_s$ — a scalar summarizing a curve.
Its measured value $\sigma_8\approx0.81$ (Planck 2018) is a pillar of the $\Lambda$ CDM concordance model and a key growth-of-structure observable.
Weak lensing and cluster counts constrain the degenerate combination $S_8\equiv\sigma_8\sqrt{\Omega_m/0.3}$ ; a mild $2$ – $3\sigma$ tension separates the CMB and low-redshift determinations.
Because linear growth is scale-independent, $\sigma_8(z)=\sigma_8\,D(z)/D(0)$ , making $\sigma_8$ a clean tracer of the growth factor and thus a test of gravity and dark energy.
Key idea: $\sigma_8$ is the RMS linear matter fluctuation in $8\,h^{-1}\,\mathrm{Mpc}$ spheres — a single scalar that fixes the amplitude of $P(k)$ and, via $\sigma_8(z)\propto D(z)$ , quantifies how much structure has grown.

Where this lives today

$\sigma_8$ and $S_8$ are the working currency of the growth-of-structure programme, and the mild disagreement between their high- and low-redshift determinations is one of the sharpest open questions in precision cosmology. On the high-redshift side, Planck fixes the primordial amplitude from the CMB and extrapolates through linear growth to predict $\sigma_8(0)=0.811$ , with independent CMB-lensing amplitudes from ACT and SPT agreeing. On the low-redshift side, cosmic-shear surveys — KiDS-1000, DES Y3, and Hyper Suprime-Cam — measure $S_8$ from the coherent distortion of billions of galaxy shapes and consistently prefer values a few percent lower, the $S_8$ tension. Cluster abundances (SZ catalogues from ACT and SPT, eROSITA X-ray samples, optical richness) probe $\sigma_8$ through the exponential sensitivity of the halo mass function, while redshift-space distortions from DESI and eBOSS pin the growth combination $f\sigma_8(z)$ epoch by epoch. The next decade belongs to Euclid, the Vera C. Rubin Observatory's LSST, and the Nancy Grace Roman Space Telescope, whose percent-level $S_8$ measurements will decide whether the tension is a hint of new physics — modified gravity, dynamical dark energy, a non-standard neutrino sector — or an artefact of intrinsic-alignment and baryonic-feedback systematics in the lensing analyses.

The physical picture

Imagine dropping millions of spheres of radius $8\,h^{-1}\,\mathrm{Mpc}$ at random across the cosmos and weighing the matter in each. Most come out near the cosmic mean, but some land on filaments and clusters and hold a surplus, while others fall in voids and hold a deficit. Record the fractional departure from the mean — the smoothed density contrast $\delta_R$ — for every sphere, and you build a distribution whose width, the standard deviation, is $\sigma_8$ . The measured value $\sigma_8\approx0.81$ says a typical sphere departs from the mean by about $80\%$ : right at the threshold where linear theory begins to fail and collapse turns mildly nonlinear. That is no accident — the radius was chosen historically precisely because bright galaxies fluctuate by order unity there.

Figure

Throw 8 h⁻¹ Mpc spheres at random across the cosmic density field: most enclose near the mean mass, some sit on overdensities, some fall in voids. σ₈ = 0.81 is simply the standard deviation of the resulting fractional-mass distribution δ_R, extrapolated to z = 0 by linear theory

Because clustering grows with cosmic time, $\sigma_8$ is also a clock. The same comoving spheres held far smaller fluctuations in the past, and in linear theory every scale grows by one common factor $D(z)$ , so $\sigma_8(z)=\sigma_8\,D(z)/D(0)$ falls smoothly toward higher redshift. Measuring how the amplitude climbs — from cluster abundances, redshift-space distortions, or lensing — tests whether structure assembles at the rate general relativity plus dark energy predict. But there is a catch that dominates the modern story: the observables cannot cleanly separate "more amplitude" from "more matter," because a denser or a lumpier universe both lens background galaxies more strongly. The data therefore constrain the combination $S_8=\sigma_8\sqrt{\Omega_m/0.3}$ , and it is along this direction that today's most-discussed disagreement in structure formation appears — the CMB predicting a slightly clumpier low-redshift universe than the lensing surveys seem to find.

Figure

The σ₈–Ωₘ plane and the S₈ tension: Planck 2018 CMB gives S₈ = σ₈√(Ωₘ/0.3) ≈ 0.83, while low-redshift weak-lensing surveys KiDS-1000 and DES Y3 prefer S₈ ≈ 0.77 — a persistent 2 to 3σ offset along the banana-shaped degeneracy σ₈√(Ωₘ/0.3) = const

The core equations

Everything begins with the variance of the smoothed field. To turn the pointwise contrast $\delta(\mathbf{x})$ into a statement about mass in spheres, convolve it with a normalized window of radius $R$ and take the variance; the convolution theorem and the definition of $P(k)$ collapse the real-space average into a single one-dimensional integral over the power spectrum, weighted by the squared window.

\sigma_R^2 = \langle \delta_R^2(\mathbf{x}) \rangle = \frac{1}{2\pi^2}\int_0^\infty P(k)\,\tilde{W}^2(kR)\,k^2\,\mathrm{d}k \qquad (1)

where: $\sigma_R^2$ — variance of the contrast smoothed on scale $R$ ; $\delta_R(\mathbf{x})$ — density field smoothed with a window of radius $R$ ; $P(k)$ — linear matter power spectrum at $z=0$ ; $\tilde{W}(kR)$ — Fourier transform of the real-space window; $k$ — comoving wavenumber. Read it as: the variance is $P(k)$ summed over all modes, each down-weighted by how much the sphere averages it out. The window defining the "standard" $\sigma_8$ is a real-space top hat — a uniform sphere. Its Fourier transform is the spherical analogue of a sinc function: unity at $k=0$ , oscillating and decaying for $kR\gtrsim1$ , which makes $\sigma_R$ a low-pass filter that keeps large-scale power and discards small-scale power.

\tilde{W}(x) = \frac{3}{x^3}\left[\sin x - x\cos x\right], \qquad x = kR \qquad (2)

where: $\tilde{W}(x)$ — Fourier transform of the real-space top-hat sphere of radius $R$ ; $x=kR$ — dimensionless argument. Read it as: $\tilde{W}(0)=1$ (a very large sphere counts every mode fully), while $\tilde{W}$ passes through zero at $x\approx4.49$ and beyond, killing modes smaller than the sphere. Fixing the smoothing radius to the historical value $8\,h^{-1}\,\mathrm{Mpc}$ turns the family $\sigma_R$ into the single number $\sigma_8$ . The scale is a convention, not a fundamental length, but a well-chosen one: it sits at the linear-to-nonlinear boundary, small enough to measure yet large enough for linear theory to be a sensible extrapolation.

\sigma_8 \equiv \sigma_R \big|_{R = 8\,h^{-1}\,\mathrm{Mpc}} \qquad (3)

where: $\sigma_8$ — RMS linear matter fluctuation in $8\,h^{-1}\,\mathrm{Mpc}$ spheres at $z=0$ ; $h$ — dimensionless Hubble parameter, $H_0=100\,h\,\mathrm{km\,s^{-1}\,Mpc^{-1}}$ . Read it as: one evaluation of equation (1) at one radius, quoted as an RMS rather than a variance. Substituting the decomposed power spectrum makes explicit that $\sigma_8$ is not independent — it inherits both the primordial amplitude and the transfer-function shape, and so couples to the entire parameter set.

\sigma_8^2 \propto A_s \int_0^\infty k^{\,n_s+2}\,T^2(k)\,\tilde{W}^2(kR)\,\mathrm{d}k \qquad (4)

where: $A_s$ — primordial scalar amplitude; $n_s$ — scalar spectral index ( $\approx0.965$ ); $T(k)$ — matter transfer function (depends on $\Omega_m h^2$ and the baryon fraction). Read it as: $\sigma_8$ rises as $\sqrt{A_s}$ , while changing $\Omega_m h^2$ reshapes the integrand through $T(k)$ — the origin of the $\sigma_8$ – $\Omega_m$ degeneracy. Because a higher amplitude and a higher matter density boost the same lensing and cluster signals, low-redshift surveys measure a rotated combination rather than $\sigma_8$ alone — the standard choice, calibrated to decorrelate the two for weak lensing, is $S_8$ .

S_8 \equiv \sigma_8\left(\frac{\Omega_m}{0.3}\right)^{0.5} \qquad (5)

where: $S_8$ — weak-lensing amplitude along the well-constrained direction; $\Omega_m$ — present-day total matter density parameter. Read it as: the quantity a shear survey actually pins down; quoting $\sigma_8$ alone from lensing overstates the precision. In linear theory every sub-horizon matter mode grows by one scale-independent function of time, so the whole amplitude — and hence $\sigma_8$ — simply scales with the growth factor, making $\sigma_8(z)$ a direct readout of the growth history.

\sigma_8(z) = \sigma_8\,\frac{D(z)}{D(0)} \qquad (6)

where: $D(z)$ — linear growth factor, normalized to $D(0)$ today; $\sigma_8(z)$ — RMS amplitude at redshift $z$ . Read it as: rewind the clock and every comoving sphere was smoother by exactly $D(z)/D(0)$ , independent of scale. Finally, the same information lives in real space as the two-point correlation function — the Fourier partner of $P(k)$ , the excess probability of a tracer pair at separation $r$ , and the statistic that historically fixed the $8\,h^{-1}\,\mathrm{Mpc}$ scale.

\xi(r) = \frac{1}{2\pi^2}\int_0^\infty P(k)\,\frac{\sin(kr)}{kr}\,k^2\,\mathrm{d}k \qquad (7)

where: $\xi(r)$ — real-space two-point correlation function; $r$ — comoving pair separation. Read it as: the same power spectrum viewed through a different kernel; $\sigma_8$ and $\xi(r)$ are two windows onto one field, related by name to aq-astro-galaxy-correlation-function.

Derivation AThe sigma-eight variance integral

The definition of $\sigma_8$ is an integral, and the goal here is to see exactly where that integral comes from rather than accept it as a formula. We begin with the raw density contrast $\delta(\mathbf{x})$ , smooth it with a normalized top-hat sphere of radius $R$ , and ask for the variance of the smoothed field over the ensemble. Using the convolution theorem and the definition of the power spectrum as the variance per mode, the real-space average collapses into a one-dimensional integral of $P(k)$ weighted by the squared window — the master formula that, evaluated at $R=8\,h^{-1}\,\mathrm{Mpc}$ , is $\sigma_8^2$ .

Step 1. Define the smoothed field as a convolution of the contrast with a normalized window, $\delta_R(\mathbf{x}) = \int \mathrm{d}^3x'\, W(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}')$ , where $\int W\,\mathrm{d}^3x = 1$ . Why: $\sigma_8$ is about mass inside spheres, so we first replace the pointwise field by its spherical average — exactly what convolution with a top hat does.

Step 2. Transform to Fourier space, where the convolution becomes a product, $\tilde{\delta}_R(\mathbf{k}) = \tilde{W}(kR)\,\tilde{\delta}(\mathbf{k})$ . Why: smoothing is a low-pass filter — each Fourier mode is rescaled by the window's transform, with high- $k$ modes suppressed.

Step 3. Form the ensemble variance and insert the power-spectrum definition $\langle\tilde{\delta}(\mathbf{k})\tilde{\delta}^*(\mathbf{k}')\rangle=(2\pi)^3\delta_D(\mathbf{k}-\mathbf{k}')P(k)$ , giving $\langle \delta_R^2 \rangle = \int \frac{\mathrm{d}^3k}{(2\pi)^3}\,|\tilde{W}(kR)|^2\,P(k)$ . Why: $P(k)$ is the variance carried per mode, so squaring and averaging the smoothed field produces the window-weighted power.

Step 4. Use statistical isotropy to do the angular integral, $\mathrm{d}^3k \to 4\pi k^2\,\mathrm{d}k$ , leaving $\sigma_R^2 = \frac{1}{2\pi^2}\int_0^\infty P(k)\,\tilde{W}^2(kR)\,k^2\,\mathrm{d}k$ . Why: isotropy reduces the 3D integral to a single radial one, with the $4\pi/(2\pi)^3 = 1/(2\pi^2)$ prefactor.

Step 5. Set $R = 8\,h^{-1}\,\mathrm{Mpc}$ to define $\sigma_8^2$ . Why: the historical radius where bright galaxies fluctuate by order unity fixes the one number that normalizes $P(k)$ , turning a functional statistic into a reportable scalar. *Note: the prefactor $1/(2\pi^2)$ and the $k^2$ measure are dimensionally consistent with $\langle\delta^2\rangle$ being dimensionless, as required.*

Go deeper · Derivation B · 1 minFourier transform of the top-hat sphere

Derivation BFourier transform of the top-hat sphere

The variance integral is only useful once we know the window it contains, so the goal here is to compute the Fourier transform of a uniform sphere from scratch and confirm the compact closed form $\tilde{W}(x)=3(\sin x - x\cos x)/x^3$ . We start from the normalized real-space top hat, do the angular part of the transform to produce a sinc kernel, and finish the radial integral by parts, checking that a very large sphere ( $x\to0$ ) counts every mode with unit weight as it must.

Step 1. Write the normalized real-space top hat, $W(r) = \frac{3}{4\pi R^3}$ for $r<R$ and $0$ otherwise, so $\int W\,\mathrm{d}^3x = 1$ . Why: the sphere must average the field, not add to it, so its integral is one — setting the leading factor $3/4\pi R^3$ that survives to the answer.

Step 2. Take the transform and do the angular integral, $\tilde{W}(k) = \int_{r<R}\frac{3}{4\pi R^3}\,e^{i\mathbf{k}\cdot\mathbf{r}}\,\mathrm{d}^3r$ with $\int \mathrm{d}\Omega\,e^{i\mathbf{k}\cdot\mathbf{r}} = 4\pi\,\frac{\sin(kr)}{kr}$ . Why: for an isotropic window only the angle-averaged plane wave survives, and that average $\sin(kr)/(kr)$ is the origin of $\tilde{W}$ 's oscillations.

Step 3. Do the radial integral by parts, $\tilde{W}(k) = \frac{3}{R^3}\int_0^R \frac{\sin(kr)}{kr}\,r^2\,\mathrm{d}r = \frac{3}{x^3}(\sin x - x\cos x)$ with $x=kR$ . Why: this closed form makes $\sigma_8$ computable in one line; $\sin x - x\cos x$ vanishing at $x\approx4.49$ sets the window's small-scale cutoff.

Step 4. Check the large-sphere limit, $\tilde{W}(x)\to 1$ as $x\to0$ (expand $\sin x - x\cos x \approx x^3/3$ ). Why: a window far larger than any wavelength must reproduce the unsmoothed field, counting each mode with full weight — and it does. *Note: $\tilde{W}(0)=1$ and the first zero falls at $kR\approx4.493$ , consistent with the numerical value used throughout.*

Go deeper · Derivation C · 1 minGrowth-factor scaling of sigma-eight

Derivation CGrowth-factor scaling of sigma-eight

The final structural fact is that $\sigma_8$ evolves in the simplest possible way, and the goal here is to show why the entire amplitude on every comoving scale moves together by a single factor. The key input, from aq-astro-linear-growth-of-structure, is that sub-horizon pressureless matter obeys a growth equation with no explicit $k$ -dependence, so all Fourier modes share one time-dependence $D(t)$ . Carrying that through the power spectrum and the variance integral yields $\sigma_8(z)=\sigma_8\,D(z)/D(0)$ , turning the amplitude into a clean probe of the growth history and hence of gravity and dark energy.

Step 1. Start from the linear growth equation, $\ddot{\delta} + 2H\dot{\delta} - 4\pi G\bar{\rho}_m\,\delta = 0$ , which for non-relativistic sub-horizon matter contains no pressure term and no $k$ . Why: the absence of spatial-scale dependence is the whole engine — every mode feels the same Hubble drag and self-gravity, so none can evolve differently.

Step 2. Because the equation is scale-independent, the solution separates, $\delta(\mathbf{x},t) = D(t)\,\delta_0(\mathbf{x})$ , with the spatial pattern frozen and only $D(t)$ evolving. Why: this factorization lets a single growth factor stand in for the time evolution of the entire field, whatever its spatial structure.

Step 3. Since $\tilde{\delta}\propto D$ , the power spectrum and variance scale as its square, $P(k,z) = D^2(z)\,P(k,0)$ and $\sigma_R^2(z) = D^2(z)\,\sigma_R^2(0)$ . Why: the window integral touches only the shape of $P(k)$ , unchanged, so the overall $D^2$ pulls straight through.

Step 4. Take the square root at $R = 8\,h^{-1}\,\mathrm{Mpc}$ to obtain $\sigma_8(z) = \sigma_8(0)\,D(z)/D(0)$ . Why: the payoff — a measured amplitude at any epoch, compared against the $\Lambda$ CDM prediction for $D(z)$ , tests whether structure grew at the demanded rate. *Note: the $D^2$ scaling of $P(k)$ is the standard linear-theory normalization, a direct consequence of the scale-independent growth derived above.*

Limiting cases

Watching how $\sigma_R$ behaves as the smoothing scale, epoch, or parameters are pushed to extremes clarifies exactly what the single number $\sigma_8$ is encoding.

Case	Result	Meaning
Large smoothing radius $R\to\infty$	$\sigma_R^2 \to 0$	The window keeps only the smallest $k$ , and since $P(k)\to0$ as $k\to0$ for $n_s\approx1$ , the RMS vanishes — the Universe is homogeneous on the largest scales.
Small smoothing radius $R\to 0$	$\sigma_R^2 \propto \int k^2 P(k)\,\mathrm{d}k$	The window admits ever-higher $k$ ; $\sigma_R$ grows and formally diverges for a pure power law — cut off in practice by the transfer-function turnover and nonlinearity.
Redshift evolution (linear theory)	$\sigma_8(z) = \sigma_8\,D(z)/D(0)$	Scale-independent growth scales the whole amplitude by $D(z)$ , so $\sigma_8(z)$ falls toward higher $z$ : structure was smoother in the past.
Matter-density degeneracy (lensing)	$S_8 \equiv \sigma_8\sqrt{\Omega_m/0.3}$	Lensing and cluster observables constrain this combination far better than $\sigma_8$ alone, because more amplitude and more matter deepen the same wells.

Common trapsigma-eight is the true, present-day, nonlinear clumpiness on 8 Mpc

It is none of those three things, and each is a genuine trap. First, $\sigma_8$ is built from the linear power spectrum extrapolated to $z=0$ , not the real field — on $8\,h^{-1}\,\mathrm{Mpc}$ the true nonlinear RMS exceeds $0.81$ , so $\sigma_8$ is a bookkeeping amplitude, not a direct measure of today's lumpiness. Second, it is not a fluctuation "at exactly 8 Mpc": it is a window-weighted integral whose weight peaks near $k\approx0.15$ – $0.25\,h\,\mathrm{Mpc}^{-1}$ . Third, a low-redshift survey cannot measure it alone — it is degenerate with $\Omega_m$ , and lensing constrains $S_8$ . Finally, the $h^{-1}$ in the radius means the physical smoothing scale shifts with $H_0$ , one more entanglement with the rest of the parameters.

Worked examples

1 · From sigma-eight to S-eight and the tension (fully worked)

Planck 2018 (TT,TE,EE+lowE+lensing) reports $\sigma_8 = 0.811$ and $\Omega_m = 0.315$ . Convert to the lensing parameter $S_8$ , then quantify how far it sits from the weak-lensing determinations KiDS-1000 ( $S_8 = 0.759 \pm 0.024$ ) and DES Y3 ( $S_8 = 0.776 \pm 0.017$ ).

Step 1. Apply $S_8 = \sigma_8(\Omega_m/0.3)^{0.5}$ . The density ratio is $0.315/0.3 = 1.050$ , with square root $1.0247$ , so $S_8 = 0.811 \times 1.0247 = \mathbf{0.831}$ — matching the Planck-quoted $S_8 = 0.832 \pm 0.013$ . Because $\Omega_m$ is slightly above the reference $0.3$ , $S_8$ lands just above $\sigma_8$ .

Step 2. Form the differences. Against KiDS-1000, $\Delta S_8 = 0.832 - 0.759 = 0.073$ ; against DES Y3, $\Delta S_8 = 0.832 - 0.776 = 0.056$ . The CMB value is the higher one — the early-Universe extrapolation expects a slightly clumpier late Universe than the lensing surveys see.

Step 3. Combine the uncertainties in quadrature. For KiDS, $\sigma_\Delta = \sqrt{0.013^2 + 0.024^2} = 0.0273$ , so the offset is $0.073/0.0273 = \mathbf{2.7\sigma}$ ; for DES, $\sigma_\Delta = \sqrt{0.013^2 + 0.017^2} = 0.0214$ , giving $0.056/0.0214 = \mathbf{2.6\sigma}$ . Each lensing survey sits about $2.6$ – $2.7\sigma$ below Planck — a mild but persistent tension, not a decisive break. This is the $S_8$ tension: too large to ignore, too small to claim new physics.

2 · Rewinding the clock with the growth factor

Take $\sigma_8(z=0) = 0.811$ in a flat $\Lambda$ CDM universe with $\Omega_m = 0.3$ . Using the standard growth-factor fit, the linear growth factor at $z=1$ is $D(z=1)/D(0) = 0.61$ . What was the clustering amplitude on $8\,h^{-1}\,\mathrm{Mpc}$ when the Universe was half its present size?

Step 1. The amplitude scales linearly with the growth factor (equation 6): $\sigma_8(z) = \sigma_8\,D(z)/D(0)$ . There is no scale-dependent correction because linear growth is scale-independent — the same $0.61$ multiplies the amplitude on every comoving scale.

Step 2. Substitute: $\sigma_8(z=1) = 0.811 \times 0.61 = \mathbf{0.49}$ . At $z=1$ the linear clustering amplitude was about $60\%$ of today's, so a typical $8\,h^{-1}\,\mathrm{Mpc}$ sphere departed from the mean by roughly half rather than four-fifths. Structure was demonstrably less assembled.

Step 3. Interpret it as a test, not just a rescaling. Cluster abundances at $z\approx1$ , redshift-space distortions (measuring the related $f\sigma_8$ ), and CMB lensing all probe $\sigma_8(z)$ at different epochs. If the amplitudes trace the $\Lambda$ CDM $D(z)$ curve, gravity plus dark energy pass; a systematic offset would point to modified gravity or evolving dark energy. The scaling law $\sigma_8(z)\propto D(z)$ is what makes $\sigma_8$ a growth-of-structure probe rather than a static number.

3 · Which scales does sigma-eight actually weigh?

The name " $\sigma_8$ " suggests a fluctuation on the $8\,h^{-1}\,\mathrm{Mpc}$ scale. Use the top-hat window to find where the variance integral really gets its contributions.

Step 1. Evaluate the squared window $\tilde{W}^2(kR)$ at a few values of $x=kR$ : at $x=1$ , $\tilde{W}^2 = 0.82$ ; at $x=2$ , $\tilde{W}^2 = 0.43$ ; at $x=\pi$ , $\tilde{W}^2 = 0.092$ . The window falls to its first zero at $x \approx 4.49$ , i.e. $k \approx 4.49/(8\,h^{-1}\,\mathrm{Mpc}) = \mathbf{0.56\ h\,Mpc^{-1}}$ . Modes with $k$ above this are almost entirely averaged away by the sphere.

Step 2. Weight against the power spectrum. The integrand of equation (1) is $\propto k^3 P(k)\,\tilde{W}^2(kR)$ per logarithmic interval; the dimensionless power $\Delta^2(k) = k^3 P(k)/2\pi^2$ rises toward higher $k$ while $\tilde{W}^2$ cuts it off above $k\approx0.5\,h\,\mathrm{Mpc^{-1}}$ . The product peaks near $\mathbf{k \approx 0.15}$ – $\mathbf{0.25\ h\,Mpc^{-1}}$ , i.e. wavelengths $\lambda = 2\pi/k \approx 25$ – $40\,h^{-1}\,\mathrm{Mpc}$ .

Step 3. Draw the lesson. $\sigma_8$ is not the fluctuation at $8\,h^{-1}\,\mathrm{Mpc}$ but a broad, window-weighted average dominated by somewhat larger scales, $k\sim0.15$ – $0.25\,h\,\mathrm{Mpc^{-1}}$ — in the quasi-linear regime where linear theory stays defensible. This is why the parameter is robust enough to normalize $P(k)$ : it never leans on the deeply nonlinear small scales.

Physics of the portrait

The amplitude of clustering is inseparable from the person who built the statistical language for it. Before the 1970s "how clumpy is the Universe" was a qualitative question; it became a precise, measurable one — a variance, a correlation function, a power spectrum — largely through one theorist, and the scale $8\,h^{-1}\,\mathrm{Mpc}$ is a fossil of his galaxy-survey analyses.

Origins · 1 minP. J. E. Peebles and how this was discovered

P. J. E. Peebles

1935– · structure formation and the statistics of clustering

Phillip James Edwin Peebles, at Princeton from the 1960s onward, turned cosmology from a subject of order-of-magnitude arguments into a quantitative statistical science. With collaborators he introduced the galaxy two-point correlation function $\xi(r)$ and its Fourier partner $P(k)$ as the fundamental descriptors of large-scale structure, and his 1980 monograph The Large-Scale Structure of the Universe codified the machinery — smoothing windows, variances, correlation hierarchies — on which $\sigma_8$ rests. The characteristic scale is itself a Peebles-era artefact: analyses of the early galaxy surveys, culminating in Davis and Peebles' 1983 study of the CfA redshift catalogue, found that bright galaxies in spheres of radius $8\,h^{-1}\,\mathrm{Mpc}$ fluctuate by roughly order unity, making that radius the natural anchor for the clustering amplitude. Peebles also helped predict the cosmic microwave background and pioneered the cold-dark-matter picture in which these fluctuations grow, and was awarded the 2019 Nobel Prize in Physics for "theoretical discoveries in physical cosmology." The modern $\sigma_8$ is the precision descendant of the statistical framework he built.

History

1970

— Peebles and Yu develop the statistical description of galaxy clustering via the correlation function

1980

— Peebles publishes The Large-Scale Structure of the Universe, codifying ξ(r) and P(k)

1983

— Davis and Peebles anchor the 8 h⁻¹ Mpc scale where bright galaxies fluctuate by order unity

1992

— COBE detects CMB anisotropies, fixing the primordial amplitude that σ8 normalizes

2018

— Planck reports σ8 = 0.811, and KiDS and DES weak lensing expose the S8 tension

Going further

Texts. Dodelson, Modern Cosmology, 2nd ed., Ch. 7–9 (power spectrum, growth, and normalization) · Ryden, Introduction to Cosmology, 2nd ed., §11–12 (structure formation) · Mo, van den Bosch and White, Galaxy Formation and Evolution, Ch. 6–7 (the density field, windows, and $\sigma_8$ ) · Peebles, The Large-Scale Structure of the Universe (the founding statistical framework) · Carroll and Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Ch. 30 · Binney and Tremaine, Galactic Dynamics, 2nd ed., §9 · Planck Collaboration 2020 (Planck 2018 results VI), A&A 641, A6. The top-hat variance integral uses window transform $\tilde{W}(x)=3(\sin x - x\cos x)/x^3$ and growth-factor scaling $\sigma_8(z)=\sigma_8\,D(z)/D(0)$ . Numbers used here: $\sigma_8 = 0.811$ and $\Omega_m = 0.315$ (Planck 2018 results VI), $S_8 = 0.832 \pm 0.013$ (Planck), $S_8 = 0.759 \pm 0.024$ (KiDS-1000, Asgari et al. 2021), $S_8 = 0.776 \pm 0.017$ (DES Y3, 2022); the growth factor $D(z=1)/D(0)=0.61$ is the Carroll–Press–Turner fit for $\Omega_m=0.3$ ; the window first-zero $kR\approx4.49$ ; the $S_8$ arithmetic gives $0.811\times1.0247=0.831$ ; the tension significances are $\approx2.6$ – $2.7\sigma$ ; and $\sigma_8(z=1)=0.811\times0.61\approx0.49$ .

Prerequisites. Density Perturbations and the Density Contrast · The Primordial Power Spectrum · Transfer Function and the Matter Power Spectrum · Linear Growth of Structure. Leads to. The ΛCDM Concordance Model · Cold Dark Matter · The CMB Anisotropy Power Spectrum.

Ask the tutor: ask me to walk through why the $S_8$ tension survives in the newest lensing data, or to work the variance integral for a power-law $P(k)$ step by step.

Is sigma-eight the actual clumpiness of the Universe on 8 Mpc scales today?

Not quite — and the distinction matters. $\sigma_8$ is computed from the linear power spectrum extrapolated to $z=0$ : it answers "how clumpy would $8\,h^{-1}\,\mathrm{Mpc}$ spheres be if linear theory held all the way to today." The real Universe on that scale is mildly nonlinear, so the true RMS fluctuation is somewhat larger than $0.81$ . Cosmologists use the linear version deliberately — it is calculable, maps directly onto the primordial amplitude $A_s$ and the growth factor, and avoids the model-dependent physics of nonlinear collapse. Think of $\sigma_8$ as a clean amplitude label for $P(k)$ , not a census of today's lumps.

Why the peculiar scale of 8 h⁻¹ Mpc — is there something fundamental about it?

Nothing fundamental; it is a well-chosen historical convention. In the early galaxy redshift surveys, the counts of bright galaxies in spheres of radius $8\,h^{-1}\,\mathrm{Mpc}$ fluctuated by roughly order unity, so $\sigma\approx1$ for galaxies there — a convenient, memorable anchor. The scale also sits near the boundary between the linear regime ( $\delta\ll1$ ) and the nonlinear regime ( $\delta\gtrsim1$ ), a sensible place to normalize. The $h^{-1}$ appears because galaxy distances are inferred from redshifts divided by $H_0$ , so comoving separations carry factors of $h^{-1}$ . A different radius would give a different number with the same information.

What exactly is the S-eight tension, and should we be worried?

It is a mild, persistent disagreement about how clumpy the low-redshift Universe is. Planck, extrapolating the CMB forward through $\Lambda$ CDM linear growth, predicts $S_8=\sigma_8\sqrt{\Omega_m/0.3}\approx0.83$ ; weak-lensing surveys (KiDS-1000, DES Y3, HSC) measure the late Universe directly and find $S_8\approx0.76$ – $0.78$ , a few percent lower, at roughly $2$ – $3\sigma$ . That is intriguing but not decisive. It could signal new physics (modified gravity, evolving dark energy, or neutrino properties that suppress late-time growth), or residual systematics — intrinsic galaxy alignments, photometric-redshift errors, or baryonic feedback on small-scale power. Euclid, LSST, and Roman are designed to settle it.

How is sigma-eight related to the amplitude A_s that inflation sets?

They are two ways of quoting the same amplitude, connected by an integral. Inflation predicts $A_s$ , and $P(k)\propto A_s\,k^{n_s}T^2(k)D^2(z)$ , so $\sigma_8^2\propto A_s\int k^{n_s+2}T^2(k)\tilde{W}^2(kR)\,\mathrm{d}k$ . Thus $\sigma_8\propto\sqrt{A_s}$ at fixed shape, but the proportionality constant folds in the transfer function and hence $\Omega_m h^2$ , the baryon fraction, and $n_s$ . Cosmologists often report $\sigma_8$ rather than $A_s$ because it is a dimensionless, low-redshift quantity observers can compare across surveys without fixing a pivot scale or $A_s$ convention. The two are interchangeable once the full parameter set is specified.

Does increasing the Hubble constant change sigma-eight?

Yes, indirectly, through the $h^{-1}$ in the smoothing radius. The sphere is $8\,h^{-1}\,\mathrm{Mpc}$ , so raising $h$ shrinks its physical (Mpc) size; a smaller sphere averages out fewer small-scale modes and admits more high- $k$ power, raising $\sigma_8$ for a fixed $P(k)$ . Changing $h$ at fixed physical densities also alters $\Omega_m$ and the transfer-function shape, moving $\sigma_8$ again. This tangle is one reason $\sigma_8$ is never quoted in isolation, and why the well-measured lensing direction is $S_8$ rather than $\sigma_8$ alone.

How do galaxy clusters and cosmic shear each get at sigma-eight?

They exploit two different consequences of the amplitude. Cluster abundances rely on the number of massive halos being exponentially sensitive to the fluctuation amplitude on cluster mass scales: a slightly higher $\sigma_8$ yields dramatically more clusters, so counting them versus mass and redshift constrains $\sigma_8$ and its growth history $\sigma_8(z)$ . Cosmic shear instead measures the coherent distortion of background galaxy shapes, with a shear power scaling roughly as $\sigma_8^2\Omega_m^{1.5}$ — hence $S_8$ . Both are late-Universe probes, which is why their consistency with the Planck extrapolation is such a stringent test of $\Lambda$ CDM.