Galaxy Bias

Every galaxy survey faces the same awkward fact: it maps the light, but structure-formation theory predicts the mass — and the mass is overwhelmingly invisible dark matter. Galaxies are not painted onto the density field uniformly; they ignite only in the hearts of dark matter halos, which condense at the rare, high-density peaks of the primordial Gaussian field. The consequence is unavoidable: the galaxy map looks more clustered than the mass map. Galaxy bias is the bridge that quantifies the mismatch — the statistical relation between the galaxy overdensity $\delta_g(\mathbf{x})$ and the total matter overdensity $\delta_m(\mathbf{x})$ . In its simplest and most-used form the relation is a single number: linear, scale-independent, deterministic bias $\delta_g = b\,\delta_m$ , so the two-point statistics inherit a factor $b^2$ : $P_g(k) = b^2 P_m(k)$ and $\xi_g(r) = b^2 \xi_m(r)$ . This underwrites nearly every measurement drawn from galaxy clustering — the matter power spectrum, the BAO standard ruler, the growth of structure — but because $P_g$ scales as $b^2$ , clustering alone measures the product $b\sigma_8$ , never either piece by itself. The physical origin is the peak-background split: a long-wavelength perturbation modulates the local collapse threshold, and rare high peaks respond far more sharply than the bulk field, giving the Kaiser result $b \approx 1 + (\nu^2-1)/\delta_c$ . Managing the breakdown of this picture on smaller scales — where bias turns nonlinear, scale-dependent, and stochastic — while breaking the $b$ – $\sigma_8$ – $f$ degeneracy with redshift-space distortions, lensing, and the bispectrum, is a central art of modern large-scale-structure cosmology.

Key takeaways

Galaxies are biased tracers: their overdensity $\delta_g$ differs from the matter overdensity $\delta_m$ , so on large scales the two power spectra differ by a factor $b^2$ .
The workhorse model is linear, scale-independent, deterministic bias $\delta_g = b\,\delta_m$ , valid where $|\delta| \ll 1$ ; more massive, luminous, or higher-redshift populations carry larger $b$ .
Bias arises because galaxies form in the highest peaks — the peak-background split shows rare, high-threshold halos cluster more strongly than the mass, giving $b > 1$ .
Bias drifts toward unity as structure grows: for conserved tracers $b(z) - 1 \propto 1/D(z)$ , so a highly biased high- $z$ population becomes nearly unbiased today.
On small scales bias becomes nonlinear, scale-dependent, and stochastic, needing higher-order terms $b_2$ , $b_{s^2}$ and a stochasticity contribution.
Bias is the central nuisance parameter of survey cosmology: it degenerates with $\sigma_8$ and the growth rate $f$ , broken with redshift-space distortions, lensing, or the bispectrum.
Key idea: galaxies trace the mass with an amplitude that differs from unity — on large scales $\delta_g = b\,\delta_m$ and hence $P_g(k) = b^2 P_m(k)$ — because galaxies form in rare high-density peaks whose clustering is enhanced relative to the underlying dark matter.

Where this lives today

Bias is the working currency of every modern redshift survey. DESI, mapping tens of millions of galaxies and quasars, and Euclid, surveying the near-infrared sky, both anchor their cosmology on the linear-bias picture: they extract the baryon-acoustic scale as a standard ruler — safe because scale-independent bias leaves the BAO position fixed — and fit the redshift-space anisotropy to measure $f\sigma_8(z)$ across cosmic time, testing whether gravity behaves as general relativity predicts. The frontier pushes the linear description to smaller scales with an effective-field-theory bias expansion ( $b_2$ , tidal, stochastic terms). On the largest scales, surveys hunt the $k^{-2}$ scale-dependent bias from local non-Gaussianity to constrain $f_{\rm NL}$ to order unity — a probe of inflation where highly biased quasars are sharpest. And cross-correlations against CMB lensing or weak-lensing shear isolate $b$ from $\sigma_8$ directly, turning the old degeneracy into a consistency test of the $\Lambda$ CDM concordance model.

The physical picture

Picture the dark matter field as a rolling landscape whose height at each point is the local overdensity $\delta_m$ . Galaxies light up only where gravity has piled up enough matter to cross a collapse threshold — the critical linear overdensity $\delta_c \approx 1.686$ from spherical-collapse theory. Now superimpose a very long-wavelength ripple: a broad region where the average height is already lifted. Small-scale peaks that would have fallen just short are nudged over the barrier, so galaxy formation is boosted wherever the background is high; the same bumps stay below the barrier wherever it is low. The galaxies over-represent the overdense regions, and their map looks more clustered than the mass. That amplification is the bias, and the first figure makes it concrete.

Figure

The peak-background split. Small-scale density peaks of equal intrinsic height cross the fixed collapse barrier δ_c ≈ 1.686 only where the long-wavelength background δ_l is high (left), so galaxies over-populate overdense regions and cluster more strongly than the mass — the origin of bias b > 1

Two levers set the strength of the bias. The first is rarity: the higher the required peak, the more violently its abundance reacts to a shift in the barrier, because the Gaussian tail is exponentially steep — so massive clusters, luminous red galaxies, and high-redshift quasars are all strongly biased, while faint dwarfs can be anti-biased with $b < 1$ . The second lever is time: as structure grows and $\delta_m$ swells, an initially rare population becomes ordinary and drifts back toward tracing the mass, with $b - 1$ shrinking as $1/D(z)$ . The second figure turns the first lever into a curve.

Figure

Halo bias climbs steeply with peak height, b(ν) = 1 + (ν² − 1)/δ_c with δ_c = 1.686. A typical peak ν = 1 is unbiased (b = 1); an L-star-like ν = 1.69 gives b = 2.09; a rare cluster peak ν = 3.37 gives b = 7.15 — rare objects cluster far more strongly than the mass

The core equations

Everything begins with the overdensity — the dimensionless contrast measuring how far a local density sits above or below the cosmic mean. The same construction applies to the counted galaxies and to the total matter, both with zero mean by construction.

\delta_g(\mathbf{x}) = \frac{n_g(\mathbf{x}) - \bar n_g}{\bar n_g}, \qquad \delta_m(\mathbf{x}) = \frac{\rho_m(\mathbf{x}) - \bar\rho_m}{\bar\rho_m} \qquad (1)

where: $n_g$ — local galaxy number density; $\bar n_g$ — mean galaxy number density; $\rho_m$ — local matter density; $\bar\rho_m$ — mean matter density. Reading: $\delta > 0$ is overdense, $\delta = -1$ a void. The central ansatz asserts that on large scales the galaxy field responds linearly to the matter field it lives in, through a single constant. This is the leading term of a general local expansion, kept alone because $|\delta_m| \ll 1$ there.

\delta_g = b_1\,\delta_m \qquad (2)

where: $b_1$ — linear bias parameter, dimensionless; $\delta_g, \delta_m$ — galaxy and matter overdensities. Reading: $b_1 > 1$ amplifies contrast, $b_1 < 1$ damps it, $b_1 = 1$ unbiased. Two-point statistics square the ansatz. Because the power spectrum and the correlation function are each a product of two field values, the constant $b_1$ pulls out as $b_1^2$ — and the shape of $P_m(k)$ is untouched.

P_g(k) = b_1^2\,P_m(k), \qquad \xi_g(r) = b_1^2\,\xi_m(r) \qquad (3)

where: $P_g, P_m$ — galaxy and matter power spectra; $\xi_g, \xi_m$ — the corresponding correlation functions; $k$ — comoving wavenumber; $r$ — comoving separation. Reading: bias rescales the amplitude but leaves every feature-position (BAO wiggles included) in place. In practice bias is often quoted through the clustering amplitude in $8\,h^{-1}$ Mpc spheres — the same fiducial scale that defines $\sigma_8$ for the mass. That ratio is what a survey directly measures.

b_1 = \frac{\sigma_{8,g}}{\sigma_8}, \qquad \sigma_{8,g}^2 = \langle \delta_g^2 \rangle_{R = 8\,h^{-1}\mathrm{Mpc}} \qquad (4)

where: $\sigma_8$ — rms matter contrast in $8\,h^{-1}$ Mpc spheres; $\sigma_{8,g}$ — rms galaxy contrast in the same spheres. Reading: only the product $b_1\sigma_8$ is observable — the seed of the degeneracy. The physical value of $b_1$ for halos follows from the peak-background split. For rare high peaks the Kaiser formula, sharpened by Press–Schechter statistics, ties the bias to the peak height $\nu$ .

b(\nu) \simeq 1 + \frac{\nu^2 - 1}{\delta_c}, \qquad \nu \equiv \frac{\delta_c}{\sigma(M)} \qquad (5)

where: $\nu$ — peak height, the collapse threshold in units of the rms; $\delta_c$ — critical linear collapse overdensity, $\approx 1.686$ ; $\sigma(M)$ — rms contrast on mass scale $M$ . Reading: $\nu > 1$ gives $b > 1$ ; $\nu = 1$ gives $b = 1$ ; halos with $\nu < 1$ can be anti-biased. Bias also evolves. If a galaxy population is neither created nor destroyed after formation, comoving conservation fixes how its bias slides toward unity as the mass field grows.

b(z) - 1 = \frac{b(z_0) - 1}{D(z)/D(z_0)} \qquad (6)

where: $D(z)$ — linear growth factor; $z_0$ — reference redshift; $b(z)$ — linear bias at $z$ . Reading: since $D$ grows, $b - 1$ shrinks — a strongly biased high- $z$ population relaxes toward $b \to 1$ . Peculiar velocities distort the measured clustering along the line of sight. The linear (Kaiser) result couples the bias to the growth rate through the angle $\mu$ — the anisotropy that lets surveys separate $b$ from the growth of structure.

P_g^{s}(k,\mu) = \left(b_1 + f\mu^2\right)^2 P_m(k) \qquad (7)

where: $\mu$ — cosine of the angle between $\mathbf{k}$ and the line of sight; $f = \mathrm{d}\ln D/\mathrm{d}\ln a$ — linear growth rate; $P_g^{s}$ — redshift-space galaxy power spectrum. Reading: the $\mu$ -dependence breaks the amplitude degeneracy, delivering $f\sigma_8$ . Finally, the primordial fluctuations may be slightly non-Gaussian. The coupling of long and short wavelengths then imprints a scale dependence on the bias that blows up on the largest scales.

\Delta b(k,z) = 2 f_{\rm NL}\,(b_1 - 1)\,\delta_c\,\frac{3\Omega_m H_0^2}{2\,k^2\,T(k)\,D(z)\,c^2} \qquad (8)

where: $f_{\rm NL}$ — local non-Gaussianity amplitude; $T(k)$ — transfer function; $\Omega_m, H_0$ — matter density parameter and Hubble constant; $c$ — speed of light. Reading: the $k^{-2}$ divergence makes highly biased tracers on huge volumes a competitive probe of inflation.

Derivation AWhy the galaxy power spectrum is b squared times the matter power spectrum

The goal is to show that the single most-used relation in survey cosmology — the galaxy power spectrum equals the matter power spectrum times the square of one bias number — follows immediately from the real-space ansatz $\delta_g = b_1\delta_m$ and nothing more. We Fourier-transform the ansatz, insert it into the definition of the power spectrum, read off the factor $b_1^2$ , and check that the shape of $P_m$ — including the baryon-acoustic wiggles — is left completely untouched.

Step 1. Fourier-transform the ansatz. Because $b_1$ is constant in real space, $\delta_g(\mathbf{x}) = b_1\delta_m(\mathbf{x})$ transforms to $\tilde\delta_g(\mathbf{k}) = b_1\,\tilde\delta_m(\mathbf{k})$ . Why: multiplication by a constant commutes with the Fourier integral — a scale-independent bias stays scale-independent in $k$ -space.

Step 2. Recall the definition of the power spectrum for a homogeneous field $X$ : $\langle \tilde\delta_X(\mathbf{k})\,\tilde\delta_X^*(\mathbf{k}')\rangle = (2\pi)^3 \delta_D(\mathbf{k} - \mathbf{k}')\,P_X(k)$ . Why: this is the variance of each Fourier mode, and it holds for the galaxy and matter fields alike.

Step 3. Substitute into the galaxy two-point function: $\langle \tilde\delta_g\,\tilde\delta_g^*\rangle = b_1^2\,\langle \tilde\delta_m\,\tilde\delta_m^*\rangle$ . Why: the two factors of $\delta_g$ each carry a real $b_1$ , which pulls out as $b_1^2$ .

Step 4. Read off $P_g(k) = b_1^2 P_m(k)$ , and by the duality of $\xi$ and $P$ , $\xi_g(r) = b_1^2\xi_m(r)$ . Why: the constant rescales only the amplitude, so every feature of $P_m(k)$ — the turnover, the BAO peaks — sits at the same $k$ , keeping biased galaxies a clean standard ruler.

Go deeper · Derivation B · 1 minHalo bias from the peak-background split

Derivation BHalo bias from the peak-background split

The aim is to derive the Kaiser high-peak formula $b = 1 + (\nu^2 - 1)/\delta_c$ from first principles, showing how a long-wavelength background perturbation modulates the abundance of collapsed halos. We split the field into a halo-forming short-wavelength part and a nearly constant background, lower the collapse barrier by the background, differentiate the Press–Schechter abundance, and convert from the Lagrangian to the Eulerian frame.

Step 1. Split $\delta = \delta_s + \delta_l$ , with $\delta_l$ constant across a halo-forming region. A background overdensity raises the local mean, so a halo forms wherever the small-scale field exceeds the reduced barrier $\delta_{c,\rm eff} = \delta_c - \delta_l$ . Why: collapse cares only about the total reaching $\delta_c$ , so a positive background does part of the job.

Step 2. Take the exponential core of the Press–Schechter abundance, $n(M) \propto \exp(-\delta_c^2/2\sigma^2)$ . Why: the number of rare peaks is set by the Gaussian tail, whose exponential is the part that responds to any shift in the barrier.

Step 3. Differentiate with respect to the threshold: $\partial\ln n/\partial\delta_c = -\delta_c/\sigma^2 = -\nu^2/\delta_c$ . Since $\delta_l$ shifts the threshold by $-\delta_l$ , the abundance responds as $\delta_h = (\nu^2/\delta_c)\,\delta_l$ . Why: the halo overdensity responds to the background in proportion to $\nu^2$ , so rare high- $\nu$ peaks react far more strongly than typical ones.

Step 4. The mass-function prefactor contributes a $-1$ , giving the Lagrangian bias $b_L = (\nu^2 - 1)/\delta_c$ . Why: differentiating the full mass function, not just its exponential, supplies the offset that makes $b_L$ vanish at $\nu = 1$ .

Step 5. Convert to the Eulerian frame by adding 1: $b = 1 + (\nu^2 - 1)/\delta_c$ . Why: the observed bias must also account for the bulk motion of the background matter, which contributes exactly $+1$ .

Go deeper · Derivation C · 1 minThe Kaiser boost of the redshift-space power spectrum

Derivation CThe Kaiser boost of the redshift-space power spectrum

The purpose is to derive the anisotropic factor $(b_1 + f\mu^2)^2$ that couples the bias to the growth rate whenever clustering is measured in redshift space rather than real space. We start from the real-space bias, invoke the continuity relation between velocity divergence and density, apply the redshift-space mapping along the line of sight, and Fourier-transform to expose the angular dependence before squaring.

Step 1. In real space the galaxy field is $\delta_g = b_1\delta_m$ . Why: this is the starting linear bias; the redshift-space distortion is a kinematic correction layered on top of it.

Step 2. Linear continuity ties the velocity divergence to the growth rate, $\theta \equiv \nabla\!\cdot\!\mathbf{v} = -f\,\delta_m$ . Why: in the growing mode the peculiar velocity is sourced by the same density field, with $f$ setting how fast matter streams into overdensities.

Step 3. The redshift-space map displaces each galaxy along the line of sight by its radial velocity: $\delta_g^s = \delta_g - (aH)^{-1}\,\partial v_\parallel/\partial x_\parallel$ . Why: distance is inferred from redshift, so a line-of-sight velocity gradient distorts the apparent density along that direction only.

Step 4. Fourier-transform: the gradient brings down $\mu^2 = (k_\parallel/k)^2$ , and using $\theta = -f\delta_m$ the field becomes $\delta_g^s(\mathbf{k}) = (b_1 + f\mu^2)\,\delta_m(\mathbf{k})$ . Why: only the radial velocity distorts the map, so the growth term enters weighted by $\mu^2$ , making the clustering anisotropic.

Step 5. Square to get the power spectrum, $P_g^s(k,\mu) = (b_1 + f\mu^2)^2 P_m(k)$ . Why: the explicit $\mu$ -dependence is the lever that breaks the $b$ – $\sigma_8$ degeneracy — fitting the anisotropy delivers $f\sigma_8$ , an independent handle on the growth of structure. *Angle-averaging $(b_1 + f\mu^2)^2$ over all directions reproduces the monopole boost $1 + \tfrac{2}{3}\beta + \tfrac{1}{5}\beta^2$ with $\beta = f/b_1$ .*

Limiting cases

The bias description simplifies or ramifies depending on the scale, the tracer, and the primordial statistics:

Case	Result	Meaning
Large-scale linear regime	$\delta_g = b_1\delta_m,\quad P_g = b_1^2 P_m$	For $k \lesssim 0.1\,h\,\mathrm{Mpc}^{-1}$ and $\|\delta\| \ll 1$ , one scale-independent $b_1$ suffices.
Weakly nonlinear regime	$\delta_g = b_1\delta + \tfrac{1}{2}b_2\delta^2 + b_{s^2}s^2 + \varepsilon$	Quadratic density and tidal terms plus stochastic $\varepsilon$ make bias scale-dependent and non-deterministic.
High-mass / high-peak halos	$b(\nu) \approx 1 + \dfrac{\nu^2-1}{\delta_c}$	Rare halos ( $\nu \gg 1$ ) are strongly biased; low-mass halos ( $\nu < 1$ ) can be anti-biased.
Primordial non-Gaussianity	$\Delta b(k) \propto \dfrac{f_{\rm NL}(b_1-1)\delta_c}{k^2 T(k) D(z)}$	Local non-Gaussianity adds a scale-dependent bias diverging as $k^{-2}$ on the largest scales.
Late-time evolution	$b(z) - 1 \propto 1/D(z)$	For conserved tracers bias relaxes toward unity as structure grows.

Common trapBias is not a single universal number, and clustering alone cannot give you σ_8

Two traps hide in the innocent equation $P_g = b^2 P_m$ . First, $b$ is not a property of the universe — it is a property of the sample: linear bias holds only on large scales and only for a specified population, and changing the luminosity cut, colour, or redshift changes $b$ . On small scales the same population is scale-dependent, nonlinear, and stochastic, so one number is simply wrong. Second, because the amplitude scales as $b^2\sigma_8^2$ , clustering measures only the product $b\sigma_8$ — never $b$ and $\sigma_8$ separately. Any headline reporting " $\sigma_8$ from galaxy clustering" has secretly assumed a bias, or broken the degeneracy with extra information (RSD giving $f\sigma_8$ , or lensing giving $b\sigma_8$ with a different power of $b$ ). Treating $b$ as known when it is not is the fastest route to a spurious tension.

Worked examples

1 · From clustering amplitude to bias (confidence builder)

A galaxy sample has a correlation function four times the amplitude of the dark matter on large scales, with $\sigma_8 = 0.81$ . Find the linear bias, and say what the sample can and cannot tell you about $\sigma_8$ .

Step 1. On large scales the deterministic relation $\xi_g = b_1^2\xi_m$ holds, so a factor-of-four amplitude means $b_1^2 = 4$ , hence $b_1 = 2$ . The galaxies are twice as clustered in contrast as the mass. b₁ = 2.

Step 2. The measured clustering amplitude is proportional to $b_1\sigma_8 = 2 \times 0.81 = 1.62$ . This is the only combination the two-point clustering pins down. b₁σ₈ = 1.62.

Step 3. So the sample cannot report $\sigma_8$ alone: had the true bias been $b_1 = 1.8$ , a fit would prefer $\sigma_8 = 0.90$ and give an identical $\xi_g$ . Separating them needs an extra observable — redshift-space distortions (Example 3) or lensing. Clustering fixes b₁σ₈ = 1.62, not σ₈ alone.

2 · Peak-background bias of a galaxy and a cluster (fully worked)

Estimate the linear bias at $z = 0$ (with $\delta_c = 1.686$ ) of an $L^*$ -like host with $\sigma(M) = 1.0$ and a rare cluster with $\sigma(M) = 0.5$ , and compare their clustering strengths.

Step 1. Compute the peak heights $\nu = \delta_c/\sigma(M)$ . For the galaxy host, $\nu = 1.686/1.0 = 1.686$ ; for the cluster, $\nu = 1.686/0.5 = 3.372$ . The cluster sits far out in the Gaussian tail. ν_gal = 1.69, ν_clus = 3.37.

Step 2. Apply the Kaiser formula $b = 1 + (\nu^2 - 1)/\delta_c$ . Galaxy host: $b = 1 + (1.686^2 - 1)/1.686 = 1 + (2.843 - 1)/1.686 = 1 + 1.093 = 2.09$ . Cluster: $b = 1 + (3.372^2 - 1)/1.686 = 1 + (11.37 - 1)/1.686 = 1 + 6.15 = 7.15$ . b_gal = 2.09, b_clus = 7.15.

Step 3. Translate into clustering: since $P_g = b^2 P_m$ , the hosts show $b^2 = 4.4$ times the matter power, the clusters $b^2 = 51$ times. Clusters are thus $51/4.4 \approx 12$ times more strongly clustered in power than the $L^*$ hosts, despite living in the same mass field — a direct demonstration that rarity drives bias. Clusters cluster ≈ 12× more strongly than the galaxy hosts.

3 · Measuring the growth rate with redshift-space distortions (fully worked)

A galaxy population has $b_1 = 2.0$ at $z = 0.5$ , in a flat $\Omega_m = 0.3$ cosmology with $\sigma_8(z{=}0) = 0.81$ . Predict the growth rate $f$ , the distortion parameter $\beta$ , the quadrupole-to-monopole ratio, and the observable $f\sigma_8(z{=}0.5)$ .

Step 1. Growth rate. At $z = 0.5$ the matter fraction is $\Omega_m(z) = \Omega_m(1+z)^3/E^2 = 0.591$ , so $f \approx \Omega_m(z)^{0.55} = 0.591^{0.55} = 0.749$ . f = 0.75.

Step 2. Distortion parameter $\beta = f/b_1 = 0.749/2.0 = 0.374$ . The Kaiser multipole prefactors are $P_0 \propto 1 + \tfrac{2}{3}\beta + \tfrac{1}{5}\beta^2 = 1.278$ and $P_2 \propto \tfrac{4}{3}\beta + \tfrac{4}{7}\beta^2 = 0.580$ . β = 0.37.

Step 3. Quadrupole-to-monopole ratio: $P_2/P_0 = 0.580/1.278 = 0.454$ . A nonzero quadrupole is the fingerprint of coherent infall — the anisotropy that carries the growth signal. P₂/P₀ = 0.45.

Step 4. Observable. Report $f\sigma_8(z)$ with $\sigma_8(z) = \sigma_8(0)\,D(z)/D(0) = 0.81 \times 0.773 = 0.626$ , so $f\sigma_8 = 0.749 \times 0.626 = 0.47$ — exactly what BOSS and DESI report per redshift bin to test gravity against $\Lambda$ CDM. f σ₈(z = 0.5) = 0.47.

Physics of the portrait

Galaxy bias is where six decades of large-scale-structure theory converge: a puzzle sharpened by cluster over-clustering, explained by the statistics of rare peaks, and elevated into the nuisance parameter every survey must control.

Origins · 1 mingalaxy bias and how this was discovered

galaxy bias

the enhanced clustering of rare tracers, δ_g = b δ_m · large-scale structure and halo statistics

The idea that light does not faithfully trace mass crystallized in the early 1980s, when Nick Kaiser confronted a stubborn observation: rich Abell clusters are far more strongly correlated than galaxies, which are in turn more correlated than the mass. In a decisive 1984 paper he showed this was not a paradox but a selection effect — if clusters mark the rare high peaks of a Gaussian field, their correlation function is automatically amplified relative to the field, by a factor growing with the peak threshold. The high-peak formula $b \approx 1 + (\nu^2 - 1)/\delta_c$ was born, and with it the word "bias." Bardeen, Bond, Kaiser, and Szalay (1986) laid down the full statistics of Gaussian-field peaks, and the mechanism was recast as the peak-background split: Mo & White (1996) derived the halo bias from extended Press–Schechter, and Sheth & Tormen (1999) refined it with ellipsoidal collapse, matching simulations to a few percent. Kaiser's own 1987 analysis of redshift-space distortions turned bias from a nuisance into a tool, coupling it to the growth rate $f$ . The modern era treats bias as an effective field theory (Desjacques, Jeong & Schmidt 2018), cataloguing every allowed operator — density, tidal, higher-derivative, stochastic — while Dalal and collaborators (2008) showed that primordial non-Gaussianity leaves a spectacular $k^{-2}$ signature. What began as an anomaly in a cluster catalogue is now the connective tissue of survey cosmology.

History

1984

— Kaiser explains the enhanced clustering of rich clusters by high-peak biasing

1986

— Bardeen, Bond, Kaiser and Szalay formalize the statistics of Gaussian-field peaks

1987

— Kaiser derives linear redshift-space distortions, coupling bias to the growth rate

1996

— Mo and White derive halo bias from the peak-background split

2008

— Dalal and collaborators predict scale-dependent bias from primordial non-Gaussianity

Going further

Texts. Dodelson, Modern Cosmology, 2nd ed. — large-scale structure and redshift-space distortions · Mo, van den Bosch & White, Galaxy Formation and Evolution — halo model, mass functions, peak-background split · Peacock, Cosmological Physics — biasing and density-field statistics · Binney & Tremaine, Galactic Dynamics, 2nd ed. · Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed.

Papers. Kaiser (1984), ApJ 284, L9 — high-peak biasing · Bardeen, Bond, Kaiser & Szalay (1986), ApJ 304, 15 — peak statistics · Kaiser (1987), MNRAS 227, 1 — RSD · Mo & White (1996), MNRAS 282, 347 — halo bias · Sheth & Tormen (1999), MNRAS 308, 119 — ellipsoidal collapse · Dalal et al. (2008), PRD 77, 123514 — $f_{\rm NL}$ bias · Desjacques, Jeong & Schmidt (2018), Phys. Rep. 733, 1 — bias review. Key numerical checks: $P_g = b^2 P_m$ follows from the Fourier definition of the power spectrum; $\delta_c \approx 1.686$ from spherical collapse; $b = 1 + (\nu^2-1)/\delta_c$ gives $b = 2.09$ at $\nu = 1.69$ and $b = 7.15$ at $\nu = 3.37$ ; the Kaiser factor with $f \approx \Omega_m(z)^{0.55}$ yields $f = 0.75$ , $\beta = 0.37$ , $P_2/P_0 = 0.45$ , and $f\sigma_8 = 0.47$ at $z = 0.5$ for $\Omega_m = 0.3$ , $\sigma_8 = 0.81$ ; bias evolution $b(z) - 1 \propto 1/D(z)$ with $D(z{=}3)/D(0) = 0.32$ ; and the $k^{-2}$ scale-dependent bias from local non-Gaussianity.

Prerequisites. Density Perturbations and the Density Contrast · The Transfer Function and the Matter Power Spectrum · Linear Growth of Structure · Cold Dark Matter. Leads to. Baryon Acoustic Oscillations · The $\Lambda$ CDM Concordance Model · The Primordial Power Spectrum.

Ask the tutor: Ask me to walk through the peak-background split for a specific halo mass, or to show how redshift-space distortions turn galaxy bias from a nuisance into a measurement of how fast cosmic structure grows.

Is bias a property of the universe or of my galaxy sample?

Of your sample, emphatically. Bias measures how a particular population of tracers relates to the underlying mass, and it changes when you change the population. Brighter, redder, more massive, and higher-redshift objects all carry larger bias because they occupy rarer, higher peaks. Two surveys of the same volume can report different $b$ simply because they select different galaxies. When someone quotes "the bias," they mean that of a defined sample on large scales — never a universal constant, which is why every analysis must state its selection cuts alongside its bias value.

Why does bias leave the BAO scale untouched but rescale everything else?

Scale-independent bias multiplies the entire matter power spectrum by the same constant $b^2$ at every wavenumber. Multiplying a curve by a constant changes its height but not the location of any feature — so the turnover, the broadband slope, and crucially the baryon-acoustic wiggles all sit at exactly the same $k$ . Since BAO cosmology reads off the position of the acoustic peak, not its amplitude, an unknown constant bias is harmless — which is why galaxy surveys use biased tracers as a clean standard ruler. The moment bias becomes scale-dependent this protection weakens, so analyses check for scale-dependent bias carefully near the BAO scale.

How can a galaxy population have bias less than 1?

Anti-bias ( $b < 1$ ) happens for tracers living in low peaks — low-mass halos with $\nu < 1$ , or blue star-forming galaxies that avoid the densest, most-quenched environments. In $b = 1 + (\nu^2 - 1)/\delta_c$ , any peak with $\nu < 1$ gives a negative Lagrangian term and pulls $b$ below unity. Physically these objects are under-represented in the richest regions, so their contrast is damped relative to the mass. It is real and measured, though most flux-limited samples are dominated by massive, positively biased hosts, so anti-bias is easy to overlook.

What exactly is the peak-background split doing?

It is a divide-and-conquer trick for a field spanning a huge range of scales. You split the density into a short-wavelength part, responsible for halo collapse, and a long-wavelength "background" nearly constant across any one halo-forming region. The background does not form halos itself; it gently raises or lowers the effective collapse threshold the small-scale field must cross. High-background regions overproduce halos, low-background regions underproduce them, and the ratio of that response to the background is the bias. It converts an intractable nonlinear problem into a first-order derivative of the halo abundance with respect to the barrier — which is why the formula comes out so clean.

Why do redshift-space distortions help if they are a distortion, not a signal?

Because the distortion carries information you cannot get any other way. Peculiar velocities — galaxies streaming into overdense regions — squash the apparent clustering along the line of sight, and the strength of that squashing is set by the growth rate $f$ , so the anisotropy $(b + f\mu^2)^2$ encodes $f$ directly. Because $b$ multiplies isotropically while $f$ enters only through the line-of-sight angle $\mu$ , fitting the anisotropy separates the two, delivering $f\sigma_8(z)$ — a sensitive test of gravity, since modified-gravity theories predict a different growth history than GR.

What breaks the b–σ_8 degeneracy in practice?

Anything that measures a different combination of $b$ and $\sigma_8$ . RSD adds sensitivity to $f\sigma_8$ (independent of $b$ ), so combined with the amplitude $b\sigma_8$ they solve for both. Galaxy–galaxy lensing measures the galaxy–matter cross power $\propto b\sigma_8^2$ — a different power of $\sigma_8$ than clustering's $b^2\sigma_8^2$ — again breaking the tie and constraining the stochasticity coefficient $r$ . The bispectrum scales differently again, carrying $b^2$ and $b_2$ distinctly. Modern surveys combine two or more of these to quote $\sigma_8$ and $b$ separately rather than only their product.