Primordial Non-Gaussianity (f_NL): The Tiny Skew in Inflation's Density Field

What f_NL Is: A Skew in the Primordial Field

Inflation seeded the cosmos with primordial curvature perturbations — the density contrasts that later grew into galaxies and the cold spots of the cosmic microwave background. To leading order these fluctuations are Gaussian: a field whose statistics are fully captured by its two-point function (the power spectrum), with a symmetric bell-curve distribution of values at every point. A purely Gaussian field has zero three-point correlation.

Primordial non-Gaussianity is any deviation from that ideal. The simplest parametrization writes the gravitational potential as a Gaussian piece plus a small quadratic correction:

• Φ(x) = φ_G(x) + f_NL · (φ_G(x)² − ⟨φ_G²⟩)

Here φ_G is the underlying Gaussian field and f_NL sets the amplitude of the quadratic term. Squaring a symmetric field introduces skewness: the distribution develops a longer tail on one side. Because the correction is quadratic, its signature lives in the bispectrum — the Fourier-space three-point function — which vanishes for Gaussian statistics. A nonzero, well-shaped bispectrum is the smoking gun of primordial non-Gaussianity.

The Mechanism: Why Inflation Predicts (Almost) Zero

The size and shape of f_NL are set by how the inflaton field interacts. In the simplest picture — a single scalar field slowly rolling down a smooth potential with a canonical kinetic term — the interactions are suppressed by the slow-roll conditions themselves. The perturbations are nearly free fields, so the skew is minuscule.

Juan Maldacena showed in 2002 that in any single-field attractor model, the squeezed-limit local non-Gaussianity is fixed by the tilt of the power spectrum through a consistency relation:

• f_NL^local = (5/12)(1 − n_s)

With the measured scalar spectral index n_s ≈ 0.965, this predicts f_NL^local ≈ 0.017 — far below any foreseeable detection. The physical reason: a long-wavelength mode simply rescales the local background, and in single-field inflation that rescaling is unobservable locally. This makes a robust detection of f_NL^local ≳ 1 a clean falsification of all standard single-field inflation. Larger signals require extra ingredients: a second field (curvaton, modulated reheating), a non-canonical kinetic term (DBI/k-inflation → equilateral shape), or a non-attractor phase like ultra-slow-roll, which breaks the consistency relation and can give f_NL = 5/2.

Key Quantities and Bispectrum Shapes

f_NL is dimensionless, and its sign and shape carry as much information as its magnitude. The bispectrum B(k₁,k₂,k₃) depends on triangle configurations of three wavevectors, and different physics peaks in different triangles:

• Local (squeezed): peaks when one side is tiny (k₃ ≪ k₁ ≈ k₂). B_local ∝ f_NL[P(k₁)P(k₂) + 2 perms]. Traces super-horizon / multi-field physics.
• Equilateral: peaks for k₁ ≈ k₂ ≈ k₃; arises from higher-derivative self-interactions active near horizon crossing.
• Orthogonal: constructed to be nearly orthogonal to the other two; a diagnostic of effective-field-theory operators.
• Folded: peaks for degenerate flat triangles (k₁ ≈ 2k₂ ≈ 2k₃); flags non–Bunch-Davies initial states.

Worked scaling: the fractional non-Gaussianity of the field is roughly f_NL × Φ_rms. Since primordial fluctuations have amplitude Φ_rms ~ 10^-5 (the CMB temperature anisotropy level, ΔT/T ~ 10^-5), a signal of f_NL ~ 100 still corresponds to a skew of order 100 × 10^-5 = 10^-3 — a one-part-in-a-thousand distortion, which is why extracting f_NL demands millions of modes.

How It's Measured: CMB and Galaxy Surveys

The premier laboratory has been the cosmic microwave background. Because the CMB is a nearly two-dimensional snapshot of linear primordial physics, its bispectrum maps almost directly onto the primordial one. ESA's Planck satellite estimated the CMB bispectrum from ~10^6 temperature and polarization multipoles using separable template-fitting, binned, and modal estimators. The Planck 2018 results (Akrami et al., arXiv:1905.05697) gave, at 68% CL (temperature + E-mode polarization):

• f_NL^local = −0.9 ± 5.1
• f_NL^equil = −26 ± 47
• f_NL^ortho = −38 ± 24

All are consistent with zero — no non-Gaussianity detected, tightening the noose on exotic inflation.

The future belongs to large-scale structure. Local f_NL imprints a distinctive scale-dependent bias: it modulates how galaxies trace dark matter, adding a Δb ∝ f_NL / k² term that blows up on the largest scales. Surveys like DESI, Euclid, SPHEREx, and radio/21-cm experiments, especially with multi-tracer techniques that beat cosmic variance, aim for σ(f_NL^local) ~ 1 — the threshold where a generic multi-field signal would appear.

How It Differs from Its Cousins

It is easy to conflate primordial non-Gaussianity with other statistical or observational effects, but the distinctions are physical:

• Primordial vs. late-time (gravitational) non-Gaussianity: Gravity is nonlinear, so structure formation generates non-Gaussianity for free — that's why the present-day galaxy field is highly non-Gaussian even if the initial field was pure Gaussian. f_NL specifically targets the initial skew, which must be disentangled from this evolved signal.
• Non-Gaussianity vs. the power spectrum: the power spectrum (two-point) fixes the amplitude and tilt (A_s, n_s) of fluctuations; f_NL (three-point) probes their interactions. They are complementary — inflation can match the spectrum yet differ wildly in f_NL.
• Local vs. equilateral/orthogonal: not just different numbers but different momentum dependence, letting a bispectrum measurement diagnose which physics is responsible.
• vs. tensor-to-scalar ratio r: r probes inflation's energy scale via gravitational waves; f_NL probes the field content and interactions. Together they carve out inflationary model space.

Significance and Open Questions

Primordial non-Gaussianity is arguably the single most powerful discriminator between models of inflation. A confirmed detection of f_NL^local ≳ 1 would rule out all single-field slow-roll inflation in one stroke, forcing a multi-field or non-attractor origin — a genuine revolution. Conversely, ever-tighter null results push the simplest models forward and stress-test alternatives to inflation (ekpyrotic/bouncing cosmologies, which generically predict much larger, distinctively-shaped non-Gaussianity).

Open frontiers include:

• The σ(f_NL) ~ 1 barrier: reaching order-unity sensitivity via galaxy surveys is the field's flagship near-term goal; the challenge is controlling observational systematics that mimic the k^-2 scale-dependent-bias signal.
• Cosmological collider physics: heavy particles present during inflation leave oscillatory, mass-dependent features in the squeezed bispectrum — a potential particle spectrometer at energies of 10^13–10^14 GeV, far beyond any collider.
• Primordial black holes: the same non-attractor dynamics that boost f_NL to O(1) also govern whether inflation produces PBH dark matter.

For now, the universe's density field remains stubbornly, beautifully Gaussian — consistent with the simplest story we can tell about its birth.

The main bispectrum shapes of primordial non-Gaussianity, their triangle configurations, physical origin, and Planck 2018 constraints (68% CL, temperature + polarization)

Shape	Triangle peak	Physical origin	Planck 2018 f_NL
Local	Squeezed (k₃ ≪ k₁ ≈ k₂)	Multi-field inflation, curvaton, super-horizon nonlinearity	-0.9 ± 5.1
Equilateral	k₁ ≈ k₂ ≈ k₃	Non-canonical kinetic terms, DBI / k-inflation, higher-derivative	-26 ± 47
Orthogonal	Between equilateral and folded	Galileon / effective field theory of inflation	-38 ± 24
Single-field slow-roll	Squeezed, but tiny	Maldacena consistency relation	≈ 0.017 (predicted)
Ultra-slow-roll (non-attractor)	Squeezed	Violates consistency relation; PBH scenarios	= 5/2 (predicted)