Density Perturbations and the Density Contrast

Every map of the cosmos — the CMB's speckled sky, the froth of galaxies in a redshift survey, the filaments and voids of the cosmic web — is at bottom a map of one quantity: how far the local density strays from the cosmic mean. The Cosmological Principle promises a Universe that is homogeneous and isotropic when smeared over large volumes, and the Friedmann equations describe that smooth background with a single mean density $\bar{\rho}(t)$ . Yet the sky we see is lumpy. The reconciliation is not to abandon the smooth model but to write the true density as a background plus a small departure, and give that departure a name: the fractional overdensity, or density contrast, $\delta$ . This one dimensionless field is the master variable of structure formation. In the early Universe it is tiny — of order $10^{-5}$ on the microwave background — and that smallness is a gift, because it lets us linearize the otherwise intractable equations of a self-gravitating fluid. Once linearized, each Fourier mode $\delta_{\mathbf{k}}$ marches to its own drum: the coupled nonlinear problem of cosmic gravity dissolves into independent one-dimensional oscillators, one per wavenumber. And because inflation makes the primordial $\delta$ a Gaussian random field, no single realization is predictable — only its statistics are: the power spectrum $P(k)$ , the correlation function $\xi(r)$ , and the normalization $\sigma_8$ are the observables, while the individual galaxy is a coin toss whose odds the theory sets. This article fixes the definitions, conventions, and statistical machinery of $\delta$ that everything downstream inherits: the linear growth factor, the Jeans analysis, the transfer function, and the reading of the CMB and matter power spectra. It is foundational — nearly every later idea, from baryon acoustic oscillations to Silk damping, is a statement about how the evolution or statistics of this single field are shaped by the contents of the Universe.

Key takeaways

The density contrast $\delta(\mathbf{x},t)=\delta\rho/\bar{\rho}$ is the dimensionless fractional departure of the local density from the cosmic mean — the master variable of structure formation.
By construction $\langle\delta\rangle=0$ and $\delta\geq-1$ : voids saturate at $\delta=-1$ where $\rho\to0$ , while overdense knots grow without bound to $\delta\gg1$ , an asymmetry that breeds non-Gaussianity.
Because $\delta$ is a random field, its statistics carry the physics: the modes $\delta_{\mathbf{k}}$ and their variance $\langle|\delta_{\mathbf{k}}|^2\rangle\propto P(k)$ hold everything a Gaussian field can say.
In the linear regime $|\delta|\ll1$ each mode grows by one scale-independent factor $D(t)$ ; in Einstein–de Sitter $\delta\propto a$ , so gravity fights the Hubble drag and growth is only linear in the scale factor.
Observed structure descends from seeds $\delta\sim10^{-5}$ on the CMB, amplified over $\sim13.8$ Gyr; the standard amplitude is $\sigma_8\approx0.81$ , the rms contrast in $8\,h^{-1}$ Mpc spheres.
Key idea: split the density as $\rho(\mathbf{x},t)=\bar{\rho}(t)[1+\delta(\mathbf{x},t)]$ ; at linear order the dimensionless $\delta$ obeys decoupled evolution equations in Fourier space, so structure formation becomes the growth of the statistics of a Gaussian random field.

The core insight

Split the density into mean plus fluctuation, $\rho(\mathbf{x},t)=\bar{\rho}(t)[1+\delta(\mathbf{x},t)]$ ; the dimensionless $\delta$ obeys, at linear order, decoupled evolution equations in Fourier space, so structure formation becomes the growth of the statistics of a Gaussian random field.

Where this lives today

The statistics of $\delta$ are the central battleground of modern cosmology. Stage-IV surveys map the density field at unprecedented reach: DESI measures tens of millions of redshifts to trace the baryon acoustic peak in $\xi(r)$ across cosmic time, while Euclid and Rubin's LSST use weak gravitational lensing — cosmic shear, which reads the projected matter contrast directly and sidesteps galaxy bias — to pin the amplitude $S_8=\sigma_8\sqrt{\Omega_m/0.3}$ . That amplitude hosts a live tension: lensing surveys (KiDS, DES) have preferred a slightly lower $S_8$ than Planck extrapolated via linear growth, a $2$ – $3\sigma$ discrepancy that, if real, would mean structure grew a touch less than $\Lambda$ CDM predicts — a possible crack hinting at massive neutrinos, evolving dark energy, or new growth physics. On the largest scales, CMB-S4 will hunt the non-Gaussianity that would betray $\delta$ 's inflationary origin. Underpinning it all, $N$ -body simulations evolve $\delta$ from Gaussian initial conditions deep into the nonlinear regime to supply the templates surveys are read against — every programme, at bottom, a measurement of the statistics of a single field.

The physical picture

Colour every point of the Universe by how much denser or emptier it is than average. White regions ( $\delta=0$ ) sit at the mean; red regions ( $\delta>0$ ) are overdense proto-structures; blue regions drain toward $\delta=-1$ as they empty into voids. Cosmic evolution is the deepening of the reds and the bleeding-toward-blue of the blues under gravity — a slow contrast enhancement of a picture that began almost flat. The field is bounded below but not above: you cannot have less than empty space, so $\delta=-1$ is a hard floor, yet you can pile density up without limit — which is why an initially symmetric, near-Gaussian $\delta$ skews as it evolves, voids swelling to fill most of the volume while overdense knots collapse into the sharp peaks of the cosmic web.

Figure

The density split ρ(x) = ρ̄(1 + δ): the density (blue) wanders around the mean ρ̄ (dashed). Overdense regions have δ over 0 and grow without limit; voids drain toward the hard floor δ = −1 where ρ = 0

Because a single realization is unpredictable, cosmology is a statistical science of $\delta$ : we measure the variance on a chosen scale ( $\sigma_8$ ) and the power spectrum $P(k)$ , then confront them with theory, for the individual galaxy is a throw of the dice whose odds — not outcome — theory sets. In the linear regime each Fourier mode is an independent growing mode — the deep reason perturbation theory works at all, for the intractable nonlinear $N$ -body problem becomes a diagonal set of one-dimensional problems, one per wavenumber. And gravity is slow: in matter domination the growing mode climbs only as fast as the scale factor, so it takes the full age of the Universe to turn a $10^{-5}$ ripple into a collapsed galaxy.

Figure

Linear growth of the density contrast in an Einstein–de Sitter universe: the growing mode climbs as δ ∝ a while the decaying mode falls as δ ∝ a^−3/2 and is forgotten. A seed of order 10^−3 at recombination grows by ≈ 10^3 to reach the collapse threshold δ_c = 1.686 today

The core equations

The starting point is a definition, not a law: we posit the split of the true density into a smooth background and a perturbation, and read $\delta$ off as the fractional part. Everything analytic that follows is a consequence of this split plus the fluid and gravity equations.

\rho(\mathbf{x},t)=\bar{\rho}(t)\left[1+\delta(\mathbf{x},t)\right],\qquad \delta(\mathbf{x},t)\equiv\frac{\rho-\bar{\rho}}{\bar{\rho}}=\frac{\delta\rho}{\bar{\rho}} \qquad (1)

where: $\rho$ — local matter density at comoving position $\mathbf{x}$ , time $t$ ; $\bar{\rho}(t)$ — background density, $\propto a^{-3}$ for matter; $\delta$ — dimensionless contrast; $\delta\rho$ — fluctuation $\rho-\bar{\rho}$ ; $a(t)$ — scale factor. Reading: the mean carries the units and expansion history; $\delta$ is a pure number, zero on average and never below $-1$ . The linearized fluid equations have spatially homogeneous coefficients, so plane waves are eigenfunctions. Decomposing $\delta$ into Fourier modes is therefore the natural language of the problem: each mode evolves on its own.

\delta(\mathbf{x},t)=\frac{1}{(2\pi)^3}\int \mathrm{d}^3k\;\delta_{\mathbf{k}}(t)\,e^{i\mathbf{k}\cdot\mathbf{x}},\qquad \delta_{-\mathbf{k}}=\delta_{\mathbf{k}}^{*} \qquad (2)

where: $\mathbf{k}$ — comoving wavevector, physical wavelength $\lambda=2\pi a/k$ ; $\delta_{\mathbf{k}}$ — complex Fourier amplitude. Reading: reality of the field forces $\delta_{-\mathbf{k}}=\delta_{\mathbf{k}}^{*}$ , so independent information lives in half of $k$ -space. Because $\delta$ is random, only the statistics of the modes are predictable. Statistical homogeneity makes distinct modes uncorrelated (a Dirac delta in $\mathbf{k}-\mathbf{k}'$ ); isotropy makes the rest depend only on $k$ — that function is the power spectrum.

\langle\delta_{\mathbf{k}}\,\delta_{\mathbf{k}'}^{*}\rangle=(2\pi)^3\,\delta_{\rm D}^{(3)}(\mathbf{k}-\mathbf{k}')\,P(k) \qquad (3)

where: $P(k)$ — matter power spectrum, dimension (length) $^3$ ; $\delta_{\rm D}^{(3)}$ — Dirac delta enforcing statistical homogeneity. Reading: for a Gaussian field $P(k)$ is the whole story — every higher moment vanishes or is fixed by it, so measuring it measures the cosmology. The real-space counterpart of $P(k)$ is the two-point correlation function $\xi(r)$ , its Fourier transform. It answers a concrete question: given an overdensity here, how overdense is a point a distance $r$ away, on average?

\xi(r)\equiv\langle\delta(\mathbf{x})\,\delta(\mathbf{x}+\mathbf{r})\rangle=\frac{1}{2\pi^2}\int_0^\infty \mathrm{d}k\,k^2\,P(k)\,\frac{\sin kr}{kr} \qquad (4)

where: $\xi(r)$ — dimensionless two-point correlation; $r$ — comoving separation; the $\sin(kr)/kr$ kernel is the angular average of $e^{i\mathbf{k}\cdot\mathbf{r}}$ . Reading: $\xi>0$ means clustering, $\xi<0$ anti-correlation; the baryon acoustic peak is a bump near $r\approx150$ Mpc. To compare theory and data at one number we smooth $\delta$ on a scale $R$ and take its variance. The dimensionless power $\Delta^2(k)$ is the contribution to $\langle\delta^2\rangle$ per $\ln k$ ; integrated against a window it gives the smoothed variance.

\Delta^2(k)\equiv\frac{k^3 P(k)}{2\pi^2},\qquad \sigma^2(R)=\int_0^\infty \frac{\mathrm{d}k}{k}\,\Delta^2(k)\,|W(kR)|^2 \qquad (5)

where: $\Delta^2(k)$ — variance per $\ln k$ ; $\sigma^2(R)$ — variance of $\delta$ smoothed on scale $R$ ; $W(kR)$ — window function, top-hat form $W(x)=3(\sin x-x\cos x)/x^3$ . Reading: at $R=8\,h^{-1}$ Mpc this defines $\sigma_8\approx0.81$ , the standard normalization and the scale where structure crosses from linear to nonlinear. The dynamics that make $\delta$ grow are captured by one second-order equation per mode, derived below. It reads as a damped, driven oscillator, in which Hubble friction, gravitational driving, and a pressure restoring term compete.

\ddot{\delta}_{\mathbf{k}}+2H\dot{\delta}_{\mathbf{k}}+\left(\frac{c_s^2 k^2}{a^2}-4\pi G\bar{\rho}\right)\delta_{\mathbf{k}}=0 \qquad (6)

where: $H=\dot a/a$ — Hubble parameter (friction term $2H\dot\delta$ ); $G$ — Newton's constant; $\bar\rho$ — mean matter density driving collapse; $c_s$ — sound speed setting the pressure term $c_s^2 k^2/a^2$ . Reading: a negative bracket means the mode grows (Jeans instability), a positive one means it oscillates; $c_s=0$ gives the pressureless dark-matter equation.

Derivation ALinear growth equation from the fluid and Poisson system

The goal is to obtain from first principles the second-order differential equation governing the density contrast of a pressureless (cold dark matter) fluid in an expanding universe, and to show it is scale-independent so every Fourier mode grows in lockstep during matter domination. We perturb continuity, Euler, and Poisson to linear order about the Friedmann background in comoving coordinates $\mathbf{x}=\mathbf{r}/a$ , with velocity $\mathbf{v}=H\mathbf{r}+\mathbf{u}$ , $\rho=\bar\rho(1+\delta)$ , and $\Phi=\bar\Phi+\phi$ .

Step 1. Mass conservation $\partial_t\rho+\nabla\cdot(\rho\mathbf{v})=0$ splits into a background piece $\dot{\bar\rho}+3H\bar\rho=0$ (just $\bar\rho\propto a^{-3}$ ) and a linear piece $\dot\delta+\tfrac{1}{a}\nabla_x\cdot\mathbf{u}=0$ . Why: the background part is automatic, so all clustering physics lives in the perturbation equation, which ties the growth of $\delta$ to the divergence of the peculiar velocity.

Step 2. The Euler equation, pressure dropped for cold matter, linearizes to $\dot{\mathbf{u}}+H\mathbf{u}=-\tfrac{1}{a}\nabla_x\phi$ . Why: the $H\mathbf{u}$ term is cosmic drag — expansion dilutes peculiar momentum — and it is this drag that damps growth into a mild power law rather than a runaway exponential.

Step 3. The perturbed Poisson equation is $\nabla_x^2\phi=4\pi G\,a^2\,\bar\rho\,\delta$ , sourced only by the overdensity. Why: the smooth background is absorbed into the Friedmann equations, so only $\delta$ , not $\bar\rho$ , pulls matter together.

Step 4. Differentiate continuity in time, insert the divergence of Euler, and eliminate $\phi$ with Poisson; the velocity divergence cancels, leaving $\ddot\delta+2H\dot\delta=4\pi G\bar\rho\,\delta$ . Why: $k$ never appears — the pressureless equation is identical for every mode, so in matter domination one growth factor $D(t)$ rescales the whole spectrum without changing its shape.

Step 5. A finite sound speed restores a pressure term $-c_s^2\nabla^2\delta/a^2$ , giving $\ddot\delta_{\mathbf{k}}+2H\dot\delta_{\mathbf{k}}+(c_s^2k^2/a^2-4\pi G\bar\rho)\delta_{\mathbf{k}}=0$ . Why: now $k$ re-enters through pressure, and growth requires a negative bracket — modes above the Jeans length collapse while smaller ones oscillate, seeding the baryon acoustic physics. *Setting $c_s=0$ in Einstein–de Sitter recovers $\delta\propto t^{2/3}\propto a$ , as derived in the next section.*

Go deeper · Derivation B · 1 minGrowing and decaying modes in an Einstein–de Sitter universe

Derivation BGrowing and decaying modes in an Einstein–de Sitter universe

The goal is to solve the pressureless growth equation in the matter-dominated Einstein–de Sitter background — the cleanest analytic case — extracting both modes explicitly and showing the growing mode tracks the scale factor while the decaying mode is erased within a few expansion times. In Einstein–de Sitter $a\propto t^{2/3}$ , so $H=2/(3t)$ and $4\pi G\bar\rho=\tfrac{3}{2}H^2=2/(3t^2)$ ; substituting these turns the growth equation into a Cauchy–Euler equation solvable by a power-law ansatz.

Step 1. Insert $H=2/(3t)$ and $4\pi G\bar\rho=2/(3t^2)$ into $\ddot\delta+2H\dot\delta-4\pi G\bar\rho\,\delta=0$ to get $\ddot\delta+\tfrac{4}{3t}\dot\delta-\tfrac{2}{3t^2}\delta=0$ . Why: every coefficient scales as an inverse power of $t$ — a scale-free equation with no characteristic time, whose solutions must be pure powers of $t$ .

Step 2. Try $\delta\propto t^{n}$ ; the equation reduces to $n(n-1)+\tfrac{4}{3}n-\tfrac{2}{3}=0$ , i.e. $3n^2+n-2=0$ . Why: a power law is an eigenfunction of the Cauchy–Euler operator, turning the equation into a quadratic whose roots are the two independent modes.

Step 3. The roots are $n=\tfrac{2}{3}$ and $n=-1$ , so $\delta(t)=A\,t^{2/3}+B\,t^{-1}$ . Since $a\propto t^{2/3}$ , the first is $\delta_+\propto a$ and the second $\delta_-\propto a^{-3/2}$ . Why: $\delta_+\propto a$ builds structure; the decaying mode fades within a few-fold expansion, so initial peculiar velocities are quickly forgotten and only the growing mode survives to be observed.

Step 4. The growing mode climbs only linearly with $a$ , not exponentially, because Hubble friction constantly bleeds energy from the collapse. Why: this slow, power-law growth is why the $\sim10^{-5}$ CMB seeds need the whole age of the Universe to reach $\delta\sim1$ — expansion brakes gravity. *The roots $n=2/3,-1$ confirm that $\delta\propto a$ recovers the growth factor $D(a)=a$ in Einstein–de Sitter.*

Go deeper · Derivation C · 1 minRoot-mean-square contrast and the sigma-eight normalization

Derivation CRoot-mean-square contrast and the sigma-eight normalization

The goal is to build the variance of the density field smoothed on a scale $R$ from the power spectrum, and to show why the top-hat window at $R=8\,h^{-1}$ Mpc became the standard normalization $\sigma_8$ . Smoothing suppresses the small-scale nonlinear modes and leaves a single well-defined number theory and surveys can both quote; it is a convolution in real space, hence a product in Fourier space.

Step 1. Smooth with a normalized window, $\delta_R(\mathbf{x})=\int\delta(\mathbf{x}')\,W(|\mathbf{x}-\mathbf{x}'|/R)\,\mathrm{d}^3x'$ , i.e. $\tilde\delta_R(\mathbf{k})=W(kR)\,\delta_{\mathbf{k}}$ in Fourier space. Why: convolution becomes multiplication, so the window is a low-pass filter keeping modes with $kR\lesssim1$ and discarding small-scale power where linear theory fails — what makes $\sigma(R)$ a linear-theory observable.

Step 2. Form the variance using (3): $\sigma^2(R)=\tfrac{1}{(2\pi)^3}\int\mathrm{d}^3k\,P(k)\,|W(kR)|^2$ , and the angular integral gives $\sigma^2(R)=\tfrac{1}{2\pi^2}\int_0^\infty k^2P(k)|W(kR)|^2\,\mathrm{d}k$ . Why: the Dirac delta collapses the double integral and isotropy reduces it to one dimension, which any tabulated $P(k)$ can be pushed through.

Step 3. For a spherical top-hat of radius $R$ , $W(x)=3(\sin x-x\cos x)/x^3$ with $W\to1$ as $x\to0$ ; evaluating at $R=8\,h^{-1}$ Mpc defines $\sigma_8$ . Why: $W\to1$ means large-scale modes pass unfiltered, so $\sigma_8$ samples the linear amplitude faithfully; the scale was chosen because early galaxy counts gave $\sigma\approx1$ there, modern value near $0.81$ .

Step 4. Interpretation: $\sigma_8\lesssim1$ says the rms linear contrast on $8\,h^{-1}$ Mpc is of order unity today. Why: this marks the boundary between the still-linear large scales and the collapsed small scales — why cluster abundance, sensitive to threshold crossing, is such a sharp probe of $\sigma_8$ . *Numerically $W(10^{-4})=1.0000$ , confirming the large-scale limit; the threshold interpretation uses $\delta_c=1.686$ derived below.*

Limiting cases

The behaviour of $\delta$ separates cleanly by amplitude and by scale relative to the horizon and Jeans length, with standard numbers marking the boundaries.

Case	Result	Meaning
Linear regime	$\vert\delta\vert\ll1$	Modes evolve independently, $\delta_{\mathbf{k}}(t)=D(t)\,\delta_{\mathbf{k}}(t_i)/D(t_i)$ , scale-independent in matter domination. Valid on large scales / early times.
Quasi-linear / turnaround	$\delta\sim1$	Mode coupling switches on and perturbation theory breaks; a spherical overdensity turns around when its linear contrast is $\delta_{\rm lin}\approx1.06$ .
Nonlinear collapse	$\delta\gg1$	Structures virialize; spherical collapse gives $\delta_c\approx1.686$ (linear) and $\Delta_{\rm vir}=18\pi^2\approx178$ (so $\rho\approx178\,\bar\rho$ ) in Einstein–de Sitter.
Void limit	$\delta\to-1$	Density approaches zero, $\rho=\bar\rho(1+\delta)\to0$ ; the floor $\delta=-1$ is exact and drives the non-Gaussian skew.
Sub-Jeans, pressure-supported	$k>k_J$	The bracket in (6) is positive; modes below the Jeans length oscillate as acoustic waves instead of growing — the origin of photon–baryon oscillations.

Common trapThe density contrast is a gauge-independent, unambiguous number on every scale

It is not. Two traps hide here. First, do not confuse the field $\delta=\delta\rho/\bar\rho$ — unbounded above, floored at $-1$ — with its statistical amplitude such as $\sigma_8$ or $P(k)$ : the field is one unpredictable realization, the amplitude a property of the ensemble. Second, and more subtly, $\delta$ is gauge-dependent in general relativity. The split of metric and density into background plus perturbation is not unique; different time slicings (synchronous, comoving, Newtonian) give different $\delta$ . Sub-horizon, all sensible gauges agree with the Newtonian $\delta\rho/\bar\rho$ , but super-horizon the value can differ by order unity and even change sign. Quoting a super-horizon " $\delta$ " without a gauge is therefore meaningless — use the gauge-invariant curvature perturbation on the largest scales instead.

Worked examples

1 · How far must a perturbation grow? (confidence builder)

In an Einstein–de Sitter universe the growing mode obeys $\delta\propto a$ . Recombination is at redshift $z\approx1100$ . What growth factor separates recombination from today, and what seed amplitude does a region need at recombination to be collapsing (reach the linear threshold $\delta_c\approx1.686$ ) now?

Step 1. Redshift and scale factor obey $a=1/(1+z)$ , so from recombination to today $a$ grows by $1+z\approx1101$ . Since $\delta\propto a$ , the contrast grows by the same factor $\approx1.1\times10^3$ .

Step 2. To sit at the collapse threshold today, $\delta_0=\delta_c=1.686$ . Running the growing mode back, the required seed is $\delta_{\rm rec}=\delta_c/(1+z)=1.686/1101=\mathbf{1.5\times10^{-3}}$ .

Step 3. Reading: a seed of a few parts in a thousand reaches collapse today — larger than the $\sim10^{-5}$ CMB temperature ripple, because cold dark matter (unimpeded by radiation pressure) starts growing before recombination and its contrast already exceeds the photon fluctuation. In real $\Lambda$ CDM, dark-energy domination after $z\sim0.3$ suppresses late growth ( $D(a{=}1)/a\approx0.78$ ), nudging the seed slightly up. Growth factor $\approx10^3$ ; required seed $\approx1.5\times10^{-3}$ .

2 · The spherical-collapse thresholds (fully worked)

Derive the two canonical numbers of nonlinear collapse in Einstein–de Sitter — the linear-theory threshold $\delta_c$ and the virial overdensity $\Delta_{\rm vir}$ — from the standard spherical top-hat results.

Step 1. A uniform overdense sphere behaves like a closed sub-universe: it expands, turns around, and collapses. The exact parametric solution gives, at full collapse, a linearly extrapolated contrast $\delta_c=\dfrac{3}{20}(12\pi)^{2/3}$ . With $12\pi=37.699$ and $37.699^{2/3}=11.24$ , $\delta_c=0.15\times11.24=\mathbf{1.686}$ .

Step 2. The true (nonlinear) overdensity when the sphere virializes is $\Delta_{\rm vir}=18\pi^2$ . With $\pi^2=9.8696$ , $\Delta_{\rm vir}=18\times9.8696=\mathbf{177.7}$ , rounded to $178$ — a virialized halo is about $178$ times denser than the mean.

Step 3. Reading: the gulf between $1.686$ and $178$ is the whole point of linear theory. A collapsed cluster has a true contrast of $10^2$ – $10^3$ , yet linear theory tracks it perfectly until its extrapolated contrast reaches the modest $1.686$ , at which instant the real object collapses. This is why threshold-crossing formalisms work: run cheap linear theory, ask where the smoothed field exceeds $1.686$ , and count halos there. $\delta_c=1.686$ (linear), $\Delta_{\rm vir}\approx178$ (true).

3 · What fraction of space is collapsing today? (fully worked)

Model the linear density field smoothed on $8\,h^{-1}$ Mpc as Gaussian with zero mean and rms $\sigma_8=0.81$ . Using the collapse threshold $\delta_c=1.686$ , estimate the fraction of volume that has crossed into collapse.

Step 1. Form the peak height $\nu=\delta_c/\sigma_8=1.686/0.81=\mathbf{2.08}$ : the threshold sits about two standard deviations out in the tail of the smoothed field.

Step 2. For a Gaussian, the fraction above $\nu$ deviations is $\tfrac{1}{2}\,\mathrm{erfc}(\nu/\sqrt2)$ . With $\nu=2.08$ this is $\tfrac12\,\mathrm{erfc}(1.47)=\mathbf{0.019}$ — about $1.9\%$ of the volume on this scale is above threshold.

Step 3. Reading: only a couple of percent of space has collapsed on $8\,h^{-1}$ Mpc scales — consistent with clusters being rare, prominent knots in a mostly under-dense web. The exponential sensitivity of that tail to $\sigma_8$ is why cluster counts are so sharp a probe. (Press–Schechter multiplies this by two for accreting material, giving $\approx3.8\%$ .) $\nu=2.08$ , collapsed fraction $\approx1.9\%$ .

Physics of the portrait

The density contrast is the meeting point of gravitation, statistics, and observational cosmology — and no one did more to weld the three into a quantitative science of the density field than P. J. E. Peebles.

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

P. J. E. Peebles

1935– · statistical theory of cosmic structure and the density field

Phillip James Edwin Peebles turned cosmology from a subject of order-of-magnitude arguments into a precision statistical science, and the density contrast is his central character. Born in Winnipeg in 1935 and at Princeton for his whole career, he arrived just as the 1965 discovery of the microwave background made the hot Big Bang concrete. Over the following decade he built the machinery this article rests on: the correlation function $\xi(r)$ as the working measure of galaxy clustering, its Fourier partner $P(k)$ , and the linear growth theory linking the fluctuations at recombination to the galaxies we count. His 1970 calculation with Jer Yu of the "primeval adiabatic perturbation" predicted acoustic oscillations decades before surveys saw the baryon acoustic peak, and his 1980 monograph The Large-Scale Structure of the Universe codified the statistical formalism of $\delta$ still in use. An early advocate of cold dark matter and a cosmological constant, he assembled much of what became $\Lambda$ CDM. The 2019 Nobel Prize in Physics recognized his "theoretical discoveries in physical cosmology" — which is to say, for teaching us how to read the statistics of the density contrast.

History

1902

— Jeans derives the instability criterion for a self-gravitating gas

1946

— Lifshitz gives the first relativistic theory of cosmological perturbations

1970

— Peebles and Yu compute the matter power spectrum with its acoustic features

1980

— Peebles' The Large-Scale Structure of the Universe codifies ξ(r) and P(k)

1992

— COBE detects the δ ≈ 10⁻⁵ temperature anisotropies of the CMB

Going further

Texts. Ryden, Introduction to Cosmology (2nd ed.), Ch. 11–12 · Dodelson, Modern Cosmology (2nd ed.), Ch. 4, 7, 9 · Mo, van den Bosch & White, Galaxy Formation and Evolution, Ch. 4 · Binney & Tremaine, Galactic Dynamics (2nd ed.), Ch. 5 · Peebles, The Large-Scale Structure of the Universe (1980) · Peacock, Cosmological Physics (1999), Ch. 15–16. Key checkable results: the EdS growing mode follows from $3n^2+n-2=0$ (roots $n=2/3,-1$ ), giving $\delta\propto a$ ; the growth equation is derived from continuity, Euler, and Poisson; the top-hat window satisfies $W(x)=3(\sin x-x\cos x)/x^3\to1$ as $x\to0$ ; the collapse thresholds are $\delta_c=\tfrac{3}{20}(12\pi)^{2/3}=1.686$ and $\Delta_{\rm vir}=18\pi^2=177.7$ ; and the Gaussian collapsed fraction is $\tfrac12\,\mathrm{erfc}(2.08/\sqrt2)=0.019$ . The cited values $\sigma_8\approx0.81$ , $\delta_{\rm CMB}\sim10^{-5}$ , and $z\approx1100$ are standard textbook numbers.

Prerequisites. The ΛCDM Concordance Model · Cold Dark Matter · Dark Energy and the Cosmological Constant. Leads to. Linear Growth of Structure · Jeans Instability (Cosmological) · The Primordial Power Spectrum · Transfer Function and the Matter Power Spectrum.

Ask the tutor: Want to see how this single field $\delta$ grows into the full matter power spectrum $P(k)$ — or how the collapse threshold $\delta_c$ turns into the halo mass function? Ask, and we will trace it step by step.

Why is δ bounded below by −1 but unbounded above, and why does it matter?

Because density cannot be negative: $\rho=\bar\rho(1+\delta)\geq0$ forces $\delta\geq-1$ , reached only in a perfectly empty void. There is no ceiling — inside a collapsed halo $\delta$ climbs to $10^3$ or more. This one-sided bound has a deep consequence: an initially symmetric, Gaussian field cannot stay Gaussian. Underdense regions can only deepen so far before hitting the floor while overdense regions run away, so the distribution grows a long positive tail and positive skewness — voids swell to fill most of the volume while mass piles into rare, sharp peaks, which is why statistics beyond $P(k)$ carry information.

What is the difference between the density contrast and the power spectrum?

The density contrast $\delta(\mathbf{x})$ is a field — one unpredictable realization of where matter sits in our Universe. The power spectrum $P(k)$ is a statistic — a property of the ensemble of all universes the same physics could produce, giving the variance of $\delta$ carried by modes of wavenumber $k$ . You can no more predict $\delta$ at a point than a coin toss; what theory predicts and surveys measure is $P(k)$ (or its partner $\xi(r)$ ). For a Gaussian field $P(k)$ is complete — every higher moment is zero or fixed by it.

Why must a gauge be specified when discussing δ on very large scales?

In general relativity the split into smooth background plus small perturbation is not unique: you must choose how to slice spacetime into moments of constant time, and different slicings (synchronous, comoving, Newtonian) assign different $\delta$ to the same event. Well inside the horizon this is harmless — all reasonable gauges converge on $\delta\rho/\bar\rho$ — but super-horizon they disagree by order unity and $\delta$ can even change sign. So " $\delta\sim10^{-5}$ super-horizon" is meaningless until you name the gauge, which is why cosmologists use the gauge-invariant curvature perturbation there and translate to $\delta$ once modes re-enter the horizon.

If δ grows only as the scale factor, how did tiny seeds become galaxies in time?

It is a close race, resolved in three parts. First, dark matter starts early: being pressureless, it grows before recombination while baryons are still locked to the photons. Second, the relevant seed is larger than the $10^{-5}$ CMB temperature ripple — that number is essentially the potential fluctuation, while the matter contrast on galaxy scales is bigger and grows faster inside the horizon. Third, growth "only as $a$ " still means a factor of about $10^3$ from recombination to now, enough to lift a few $\times10^{-3}$ to order unity — after which nonlinear collapse runs fast. That slow linear phase is the bottleneck that made structure formation take most of cosmic history.

What exactly does σ₈ tell me, and why that particular scale?

$\sigma_8$ is the rms of the linearly evolved density contrast smoothed with a top-hat sphere of radius $8\,h^{-1}$ Mpc, today. A value near one means the typical fluctuation on that scale is of order unity — the boundary where structure crosses from the linear regime (larger scales) into the nonlinear one (smaller). The $8\,h^{-1}$ Mpc choice is historical: early galaxy counts gave $\sigma\approx1$ there, and it stuck — clean enough for linear theory yet accessible to surveys, which is why $\sigma_8$ and its lensing cousin $S_8=\sigma_8\sqrt{\Omega_m/0.3}$ became the standard amplitude probes are compared on.

How do galaxy surveys measure δ if galaxies do not perfectly trace matter?

They measure a biased tracer and correct for it. Galaxies form in the peaks of the dark-matter field, so on large scales $\delta_{\rm gal}\approx b\,\delta_{\rm m}$ with a bias $b$ usually above one, and a survey measures $b^2P(k)$ , not $P(k)$ . Bias is broken several ways: redshift-space distortions (clustering squashed along the line of sight by peculiar velocities) constrain the growth rate independent of $b$ , and combining clustering with weak lensing, which sees the matter directly, breaks the degeneracy. The baryon acoustic peak's position is a bias-independent ruler too, which is why modern analyses jointly fit clustering, lensing, and RSD.