Silk Damping

A blur with a sharp edge

The cosmic microwave background is a snapshot of the universe at the moment it became transparent, about 380,000 years after the Big Bang. Across the sky it carries hot and cold spots — temperature anisotropies at the level of a few parts in 100,000 — that map the density ripples of the primordial plasma. If you zoom in on that map, something curious happens: the texture gets sharp and detailed down to about a tenth of a degree, and then it goes smooth. The smallest features are simply not there. They were not lost to instrumental blur or to foregrounds; they were erased in the plasma itself, before the light ever set out. The agent of that erasure is Silk damping.

The mechanism is a random walk. In the hot, ionised plasma before recombination, photons could not travel freely — they were constantly Thomson-scattered off free electrons, which in turn were electromagnetically locked to the protons. Photons and baryons behaved as a single tightly-coupled fluid. But the coupling was not perfect. Each photon, between scatterings, drifted a small distance; over the age of the universe up to recombination it executed an enormous number of these small steps, diffusing a net comoving distance of a few megaparsecs. Wherever a density perturbation was smaller than that diffusion length, photons leaking out of the hot crests and into the cold troughs averaged the two together, smearing the perturbation into nothing. Joseph Silk worked this out in 1968, and the effect has carried his name ever since.

How photon diffusion erases structure

Picture the photon–baryon fluid before recombination as a thick fog of light that pushes back against gravity. When dark matter pulls baryons into an overdense region, the trapped photons heat up and their radiation pressure resists further compression. This tug-of-war drives the acoustic oscillations that produce the famous peaks in the CMB power spectrum. Silk damping is what happens when you account for the fact that the fog is not perfectly opaque.

A photon's mean free path is ℓ_mfp = 1/(n_e σ_T), where n_e is the free-electron number density and σ_T is the Thomson cross-section. Early on, n_e is huge and ℓ_mfp is tiny; photons can barely move. As the universe expands and cools, n_e drops, ℓ_mfp grows, and the photons range farther between scatterings. A random walk of N steps each of length ℓ_mfp covers a net distance of roughly √N × ℓ_mfp — far less than the total path length N × ℓ_mfp, but steadily increasing. This net comoving distance is the diffusion length λ_D. By the time recombination arrives and the fog clears entirely, λ_D has grown to a few comoving megaparsecs.

Any perturbation with a comoving wavelength shorter than λ_D is bridged by the diffusing photons and washed out; longer-wavelength perturbations survive. The smoothing is not a step function but a steep exponential. In Fourier space, a perturbation of wavenumber k is suppressed by a factor exp[-(k/k_D)²], where k_D ≈ 2π/λ_D is the damping wavenumber. Because the suppression is Gaussian in k, the high-wavenumber (small-scale) power does not gently roll over — it plunges. Projected onto the sky, this is what turns the acoustic peaks into a steeply falling damping tail above multipole ℓ ≈ 1000.

Worked example: from a few Mpc to ℓ ≈ 1000

Let us translate the Silk scale into an angular scale on the sky, which is what an experiment actually measures. Two numbers do the job: the comoving diffusion length at last scattering and the comoving distance to the last-scattering surface.

Silk scale (comoving)        λ_D  ≈ 6  Mpc       (a few Mpc; depends on Ω_b h²)
Comoving distance to LSS      D_A  ≈ 14,000 Mpc   (standard ΛCDM)
Angular scale                 θ_D  ≈ λ_D / D_A
                                   ≈ 6 / 14,000  rad
                                   ≈ 4.3 × 10⁻⁴ rad
                                   ≈ 0.025°  ≈ 1.5 arcmin (full diffusion length)

A single diffusion length subtends a very small angle, but the damping does not switch on all at once at that scale — it sets in gradually as wavelengths approach λ_D. The relationship between a comoving wavenumber k and an observed multipole is roughly ℓ ≈ k × D_A. Taking the damping wavenumber k_D ≈ 2π/λ_D ≈ 2π/6 Mpc⁻¹ ≈ 1.0 Mpc⁻¹ (in the conventions used for the diffusion envelope, k_D works out near 0.14 Mpc⁻¹ at recombination once the proper diffusion integral is done, giving):

ℓ_D ≈ k_D × D_A ≈ 0.14 Mpc⁻¹ × 14,000 Mpc ≈ 1960

Damping envelope:   D(ℓ) = exp[ -(ℓ / ℓ_D)² ]
   at ℓ = 1000:   D ≈ exp[-(0.51)²] ≈ exp(-0.26) ≈ 0.77   → ~23% suppression
   at ℓ = 1500:   D ≈ exp[-(0.77)²] ≈ exp(-0.59) ≈ 0.55   → ~45% suppression
   at ℓ = 2000:   D ≈ exp[-(1.02)²] ≈ exp(-1.04) ≈ 0.35   → ~65% suppression
   at ℓ = 3000:   D ≈ exp[-(1.53)²] ≈ exp(-2.34) ≈ 0.10   → ~90% suppression

So the effect becomes appreciable around ℓ ≈ 1000 — the third and fourth acoustic peaks are already noticeably suppressed — and by ℓ ≈ 3000 it has knocked out about 90% of the primary signal. This is exactly the behaviour seen in the data: the acoustic peaks shrink steadily and then vanish into the damping tail. The first peak near ℓ ≈ 220 is untouched; the seventh and eighth peaks, which would otherwise sit beyond ℓ ≈ 2000, are damped almost out of existence.

Quantitative anatomy of the diffusion length

The full damping calculation tracks the imperfect coupling of photons and baryons in the tight-coupling expansion of the Boltzmann equations. The comoving diffusion length is an integral over conformal time η of the photon mean free path weighted by the expansion:

k_D⁻²(η) = ∫₀^η dη'  [ 1 / (6 (1+R) n_e σ_T a) ] × [ R²/(1+R) + 16/15 ]

   where  R = 3ρ_b / 4ρ_γ   (baryon-to-photon momentum ratio)
          a = scale factor,  σ_T = Thomson cross-section
          n_e = free-electron number density

Three features of this expression matter physically. First, the diffusion length depends inversely on n_e, so it is exquisitely sensitive to the ionisation history — anything that changes how many free electrons there are at a given time (the baryon density, the helium fraction) shifts the damping scale. Second, the factor (R²/(1+R) + 16/15) splits the damping into a part driven by the baryons' shear viscosity and a part from heat conduction; both transport energy out of perturbations. Third, because recombination is gradual, the visibility function — the probability that a photon last scattered at a given time — has a finite width corresponding to roughly Δz ≈ 80 in redshift, or about 80,000 years. The diffusion integral runs across this entire interval, so the effective damping is even stronger than a naive instantaneous estimate, and the finite thickness adds its own line-of-sight (Landau) smearing on top.

The headline numbers fall out: at recombination (z ≈ 1090), the comoving Silk scale λ_D = 2π/k_D ≈ 5–7 Mpc, and the corresponding damping multipole ℓ_D ≈ 1800–2000. Raising the baryon density increases n_e, shortens the mean free path, and pushes the damping to smaller scales (higher ℓ_D); a higher helium fraction locks electrons into neutral helium earlier, lowering n_e at fixed baryon density and moving the damping to larger scales (lower ℓ_D). This is precisely the lever that lets the damping tail measure both quantities.

Variants and related damping mechanisms

Mechanism	What diffuses / smears	Coupled or free?	Scale affected	Where it shows up
Silk (diffusion) damping	Photons random-walk out of plasma overdensities	Tightly coupled, leaky	λ_D ≈ few comoving Mpc	CMB temperature damping tail, ℓ > 1000
Landau (LSS-thickness) damping	Line-of-sight averaging over finite last-scattering width	At decoupling	Δz ≈ 80 thickness	Extra small-scale smearing of all anisotropies
Neutrino free-streaming	Relativistic neutrinos stream across perturbations	Free (decoupled at z ≈ 10⁹)	Free-streaming length	Suppresses small-scale matter power; shifts ℓ_D via N_eff
Hot dark matter free-streaming	Light, fast dark matter particles erase structure	Free	~ tens of Mpc for eV-mass	Why hot DM fails to form galaxies
Baryon (acoustic) drag	Baryons released from photon pressure at drag epoch	Transition	Sound horizon ~150 Mpc	Sets the BAO scale, not a damping
Polarization (E-mode) damping	Same diffusion acting on scattering-generated polarization	Tightly coupled, leaky	Slightly smaller than λ_D	Damping tail in the EE power spectrum

The key distinction is between diffusion while particles are still coupled (Silk damping, and its E-mode counterpart) and free-streaming after particles have decoupled (neutrinos, hot dark matter). Silk damping is a true random walk inside the fog; free-streaming is a single ballistic flight across the perturbation. Landau damping is neither — it is a projection effect from the finite duration of last scattering, which is why it cannot be removed by improving angular resolution.

Observational status

The damping tail is one of the best-measured features of modern cosmology. The story progressed in resolution: COBE (1992) mapped only the largest scales; WMAP (2003–2012) resolved the first three acoustic peaks and the onset of damping; Planck (2013–2018) measured the temperature spectrum out to ℓ ≈ 2500 with cosmic-variance-limited precision through the heart of the damping tail; and the ground-based, large-aperture instruments — the Atacama Cosmology Telescope (ACT) and the South Pole Telescope (SPT) — extend the temperature and polarization spectra to ℓ ≈ 4000 and beyond, deep into the regime where Silk damping has all but extinguished the primary signal.

What the tail buys cosmology is independent leverage on parameters that the peak heights alone constrain weakly:

• Baryon density. The damping scale depends on n_e ∝ Ω_b h², so the tail's position pins the baryon density independently of the odd/even peak-height ratio that also measures it — a powerful internal cross-check. Planck finds Ω_b h² ≈ 0.0224.
• Primordial helium. Because helium locks up electrons, the damping tail is sensitive to the helium mass fraction Y_p. CMB-only measurements give Y_p ≈ 0.24, in agreement with Big Bang nucleosynthesis predictions from the same Ω_b h² — a striking consistency between physics 3 minutes and 380,000 years after the Big Bang.
• Effective number of relativistic species. Extra radiation (parametrised by N_eff) speeds the early expansion. It shrinks the sound horizon and the diffusion length by different amounts, so it shifts the damping tail relative to the peaks. The measured N_eff ≈ 3.0 is consistent with the three Standard-Model neutrino species and tightly constrains any extra "dark radiation."

The smallest scales that Silk damping removed from the primary temperature map are not entirely lost to cosmology. CMB lensing, the Sunyaev–Zel'dovich effects, and cross-correlation with galaxy surveys reach back into that regime through secondary signals — so the damped scales still inform structure-formation studies even though their primordial imprint on the temperature spectrum is gone.

Common pitfalls and misconceptions

• "Silk damping is instrumental blur." No — it is a physical erasure that happened in the plasma before any light reached us. Even a perfect, infinite-resolution telescope would see the damping tail, because the small-scale perturbations were genuinely smoothed away at recombination.
• "It only affects the highest multipoles." The suppression sets in gradually. By ℓ ≈ 1000 the third and fourth peaks are already measurably damped (~20%); it is not a sharp cutoff but a steep exponential that grows through the whole high-ℓ range.
• "Photons stream freely out of the spots." They diffuse, not stream. The plasma is still optically thick; photons execute a random walk of countless tiny steps, covering a net distance √N times smaller than their total path. Confusing diffusion with free-streaming gives the wrong scale by orders of magnitude.
• "Silk damping and the sound horizon are the same scale." They are different. The sound horizon (≈150 Mpc comoving) sets the acoustic peak spacing and the BAO ruler; the Silk scale (≈ a few Mpc) sets where the peaks fade out. They differ by roughly an order of magnitude and probe different physics.
• "Damping erases the acoustic peaks entirely." Only the high-order peaks. The first and second peaks sit well below ℓ ≈ 1000 and are essentially undamped; the tail removes the small-scale peaks while leaving the large-scale structure intact.
• "It only matters for temperature." Polarization (E-mode) anisotropy is generated by the same Thomson scattering and is damped on a closely related scale. The EE damping tail provides an independent measurement, and because polarization is sourced only at scattering, its damping is in some ways a cleaner probe of the diffusion physics.

Why it matters

Silk damping is a beautiful example of how an apparent loss of information becomes a precision tool. The small-scale temperature fluctuations really are gone — diffused away in the fog of the early universe — but the exact way they faded out is dictated by the microphysics of the plasma: how many electrons there were, how much helium had formed, how fast the universe was expanding. By measuring the shape and position of the damping tail, cosmologists read those conditions directly off the sky and check them against Big Bang nucleosynthesis and the Standard Model of particle physics. A 1968 calculation about photons taking a random walk has become one of the sharpest tests of fundamental physics at the dawn of time.