CMB Lensing: How Cosmic Structure Bends the Oldest Light

What CMB Lensing Is: The Oldest Light on a Distorted Screen

The cosmic microwave background is a snapshot of the universe 380,000 years after the Big Bang, when it cooled enough for electrons and protons to combine and photons to stream freely. Those photons have travelled a comoving distance of roughly 14 gigaparsecs to reach us, passing through every galaxy cluster, filament, and void along the way.

Mass curves spacetime, so each clump of matter acts as a weak lens, subtly redirecting the CMB photons. The net effect is a remapping of the observed CMB: the temperature we measure in a given direction n̂ is actually the true, unlensed temperature from a slightly different direction n̂ + d, where d is the deflection field.

• The source is a single, near-uniform screen — statistically the simplest possible source.
• The lens is the entire projected matter distribution, dominated by dark matter.
• The signal is weak: deflections are ~2.5 arcmin, tiny compared to the ~1° scale of the primary anisotropies.

Crucially, lensing probes all gravitating matter, so it maps the dark cosmic web without needing luminous tracers.

The Mechanism: From Gravitational Potential to Deflection Field

The deflection is the angular gradient of a 2D lensing potential φ, which is a weighted line-of-sight integral of the 3D gravitational potential Ψ:

φ(n̂) = -2 ∫₀^χ* dχ [(χ* − χ)/(χ* χ)] Ψ(χ n̂, χ)

Here χ is comoving distance, χ* ≈ 14 Gpc is the distance to last scattering, and the geometric kernel (χ* − χ)/(χ* χ) weights structure most heavily around the midpoint of the path. The deflection is d = ∇φ.

Two closely related fields summarize the effect:

• Convergence κ = -½∇²φ — an (iso)tropic magnification, directly proportional to the projected mass density.
• Shear γ — an anisotropic stretching, analogous to galaxy weak lensing.

Because lensing conserves surface brightness but remaps positions, it smooths the acoustic peaks in the CMB power spectrum by a few percent and, importantly, converts a fraction of primordial E-mode polarization into B-mode polarization — a signal that contaminates searches for primordial gravitational waves but is itself a clean lensing probe.

Characteristic Numbers and a Worked Estimate

The deflection field has an RMS amplitude of about 2.5 arcminutes and is coherent over several degrees (~2–3°), because the potential is dominated by large-scale, mildly nonlinear structure. The lensing kernel is broad in redshift, peaking near z ≈ 1.8–2, so CMB lensing is most sensitive to structure roughly 10 billion years ago — a redshift range poorly probed by galaxy surveys.

The lensing potential power spectrum peaks at multipoles L ≈ 40–60, corresponding to angular scales of a few degrees. A useful scaling: the convergence power roughly tracks the matter power spectrum projected along the line of sight, C_L^κκ ∝ ∫ dχ W(χ)² P_m(k = L/χ, χ) / χ², where W is the lensing kernel.

• Deflection RMS: ~2.5′ (0.0007 rad)
• Coherence scale: ~2–3°
• Peak lensing multipole: L ≈ 50
• Fractional smoothing of acoustic peaks: a few %
• Lensing B-mode amplitude: ~5 µK·arcmin equivalent white noise

The measured lensing amplitude relative to the Planck ΛCDM prediction is A_lens = 1.013 ± 0.023 (ACT DR6) — a 2.3% measurement fully consistent with the standard model.

How It's Measured: Quadratic Estimators and Map-Making

Lensing correlates modes of the CMB that would otherwise be independent, breaking the statistical isotropy of the unlensed sky. The workhorse technique is the quadratic estimator (Hu & Okamoto 2002): one correlates the observed CMB with gradients of the temperature and polarization fields to reconstruct a map of the deflection (or convergence) field. Squaring and averaging that reconstruction yields the lensing power spectrum C_L^φφ.

• Reconstruction: off-diagonal T–T, T–E, E–E, E–B, T–B correlations each give an estimator; they are optimally combined.
• Bias subtraction: the dominant Gaussian 'N0' bias and higher-order 'N1' bias must be modeled and removed with simulations.
• Foregrounds: the thermal Sunyaev–Zel'dovich effect and the cosmic infrared background mimic lensing in temperature; polarization-based estimators are cleaner.

Milestone detections came from ACT (2011) and SPT (2012) via cross-correlation, then Planck's full-sky lensing map (2013, 2018). By 2024, ACT DR6 reached 43σ and Planck PR4 ~42σ; combined, the ACT+Planck lensing spectrum reaches an effective ~58σ. Next-generation estimators (maximum a posteriori, or MAP) improve on the quadratic method at low noise, and CMB-S4 aims to map lensing at very high resolution.

How It Differs from Its Cousins

CMB lensing sits within the broader family of gravitational lensing, but its diffuse, cosmological-distance source makes it distinct.

• vs. strong lensing: Strong lensing (multiple images, Einstein rings) is a highly nonlinear deflection near massive galaxies or clusters. CMB lensing is weak everywhere — no multiple images, just a percent-level statistical remapping.
• vs. galaxy weak lensing (cosmic shear): Both trace the cosmic web, but the CMB source plane is a single, precisely known screen at z≈1100, free of intrinsic-alignment and photometric-redshift systematics that plague galaxy shear. CMB lensing peaks at higher redshift (z≈2 vs z≈0.4).
• vs. the integrated Sachs–Wolfe (ISW) effect: ISW is an energy shift from evolving potentials; lensing is a positional remapping. Both correlate the CMB with large-scale structure but through different physics.

A powerful modern strategy is cross-correlation: multiplying the CMB lensing map against galaxy catalogs (e.g., DESI luminous red galaxies) isolates structure in redshift slices, sharpening constraints on the growth of structure and the sum of neutrino masses.

Why It Matters: Weighing the Dark Universe and Open Questions

CMB lensing has become a cornerstone probe of structure growth, encapsulated in the parameter combination S₈ = σ₈ (Ω_m/0.3)^0.5, which measures how clumpy the universe is. Because it sees all matter at high redshift with a clean source, it sidesteps the systematics of galaxy surveys.

• The S₈ tension: Several low-redshift galaxy weak-lensing surveys hinted that the universe is slightly less clumpy than the CMB predicts. Strikingly, the ACT DR6 and Planck PR4 lensing measurements find no such deficit — they agree with the ΛCDM prediction from the primary CMB (A_lens ≈ 1.01), deepening the puzzle about why some other probes disagree.
• Neutrino mass: Massive neutrinos suppress small-scale structure; lensing constrains Σm_ν, currently pushing the upper bound toward ~0.06–0.12 eV.
• Delensing: Removing the lensing B-mode is essential for detecting primordial gravitational waves (the tensor-to-scalar ratio r); lensing maps are used to 'clean' polarization data.

Open questions include reconciling S₈ across probes, controlling foreground biases at high resolution, and whether next-generation experiments (Simons Observatory, CMB-S4) will reveal cracks in ΛCDM through the growth history that lensing so cleanly records.

CMB lensing versus galaxy weak lensing (cosmic shear): two probes of the same cosmic web, with different source planes.

Property	CMB lensing	Galaxy weak lensing (cosmic shear)
Source	Last scattering surface, single plane at z≈1100	Distribution of galaxies, z≈0.3–1.5
Comoving distance to source	~14 Gpc (fixed, precisely known)	~1–4.5 Gpc (varies, needs photo-z)
Peak sensitivity redshift	z ≈ 1.8–2	z ≈ 0.3–0.7
Primary observable	Deflection field → convergence κ, potential φ	Shear γ of galaxy shapes
Main systematic	Foregrounds (tSZ, CIB), noise bias N0/N1	Intrinsic alignments, shape/photo-z errors
Detection significance (2024)	~43σ (ACT DR6), ~42σ (Planck PR4)	~30σ (DES, KiDS, HSC)