CMB B-Modes

The signal worth a Nobel Prize

The cosmic microwave background is not just a map of temperature; it is also faintly polarized, at the level of a few microkelvin. That polarization carries a pattern, and the pattern splits cleanly into two flavours that mathematicians have known about for centuries in the study of vector fields: a curl-free part and a divergence-free part. Borrowing the language of electromagnetism, cosmologists call them E-modes (curl-free, like the electric field around a charge) and B-modes (divergence-free, like the magnetic field around a current). The names are an analogy to the decomposition, not a claim about real electric or magnetic fields.

This decomposition would be a dry formality except for one extraordinary fact: the simplest things that wrinkled the early universe — ordinary density ripples — are physically incapable of producing B-modes. A clean B-mode pattern on large patches of the sky cannot come from density fluctuations at all. It can only come from a source with intrinsic handedness, and the leading candidate is a sea of primordial gravitational waves generated during cosmic inflation in the first 10⁻³⁵ seconds. Detecting that signal would be direct evidence that inflation happened and would pin down its energy scale — somewhere near the grand-unification scale of 10¹⁶ GeV, a regime forever beyond any particle accelerator. That is why the B-mode is among the most coveted measurements in all of physics.

How the CMB gets polarized at all

Polarization is born at the surface of last scattering, about 380,000 years after the Big Bang, when the universe cooled enough for electrons and protons to combine into neutral hydrogen and the fog of free electrons cleared. Just before that moment, photons were still scattering off free electrons by Thomson scattering. Thomson scattering has a crucial property: it polarizes light, but only if the radiation hitting the electron is anisotropic in a specific way — it must have a quadrupole, meaning the intensity along one axis differs from the perpendicular axis.

A monopole (uniform glow) or a dipole (one side hotter than the other) produces no net polarization after averaging over scattering directions. Only the quadrupole survives. So the recipe for CMB polarization is:

free electron  +  local quadrupole in the radiation  →  linear polarization
(Thomson)         (intensity hot on one axis,
                   cold on the perpendicular axis)

The quadrupole itself is set up by the flows of the photon-baryon fluid as it sloshes in the gravitational potential wells. The key question becomes: what kind of quadrupole does each type of perturbation make, and does that quadrupole have a handedness?

Why scalars make E-modes and tensors make B-modes

Cosmological perturbations come in three types classified by how they transform under rotations: scalar (density and potential fluctuations), vector (vorticity, which decays away and is negligible), and tensor (gravitational waves). Their symmetry decides everything.

Scalar perturbations are defined by a single function — the density contrast — and have no preferred sense of rotation. The quadrupole they produce is aligned with the gradient of the velocity field, so the resulting polarization rods point radially in and tangentially around hot and cold spots. That is a pure gradient pattern: curl-free, parity-even, an E-mode. By symmetry a scalar simply cannot produce a curl. This is a theorem, not an approximation.

Tensor perturbations — gravitational waves — are different. A gravitational wave stretches space along one transverse axis while squeezing it along the perpendicular axis, and the pattern rotates as the wave propagates (the two polarizations are conventionally labelled "plus" and "cross"). This injects a handed, rotating quadrupole at last scattering. A handed quadrupole produces both a gradient (E) and a curl (B) component in roughly equal measure. So:

Perturbation	Symmetry	Quadrupole handedness	Polarization produced
Scalar (density)	No preferred rotation	None	E-modes only
Vector (vorticity)	Decays away	—	Negligible
Tensor (grav. waves)	Handed, rotating	Yes	E-modes and B-modes

This is the entire logic of the B-mode hunt in one table. Density ripples flood the sky with E-modes (detected at high significance since DASI in 2002). They cannot make B-modes. So a clean primordial B-mode is a unique fingerprint of tensor perturbations — gravitational waves — and inflation is the only standard mechanism that generates them with a near-scale-invariant spectrum at an observable amplitude.

The tensor-to-scalar ratio and the energy of inflation

The amplitude of the primordial B-mode is quantified by the tensor-to-scalar ratio r — the ratio of gravitational-wave power to density-fluctuation power at a pivot scale, conventionally k = 0.05 Mpc⁻¹:

r = A_t / A_s          (tensor power / scalar power)

A_s ≈ 2.1 × 10⁻⁹       (measured from CMB temperature, Planck)

In single-field slow-roll inflation, r is directly tied to the energy density driving inflation through the consistency relation. The Hubble rate during inflation, and hence the inflationary potential V, scales as:

V^(1/4) ≈ 1.06 × 10¹⁶ GeV × (r / 0.01)^(1/4)

r = 0.01  →  V^(1/4) ≈ 1.1 × 10¹⁶ GeV   (near GUT scale)
r = 0.001 →  V^(1/4) ≈ 6   × 10¹⁵ GeV

So measuring r is equivalent to measuring the energy scale of inflation. A detection near r ≈ 0.01 would place inflation at the grand-unification scale — a once-in-a-century link between the largest structures in the cosmos and the deepest theory of particle physics. The single-field consistency relation also predicts the tensor spectral tilt, n_t = −r/8, which a future detection could test independently to confirm an inflationary origin.

Two bumps: recombination and reionization

The primordial B-mode power spectrum has two distinctive peaks, set by the two epochs when CMB photons last scattered off free electrons:

Feature	Multipole ℓ	Angular scale	Physical origin
Reionization bump	ℓ ≈ 5	tens of degrees	Rescattering at z ≈ 8 when first stars reionized the gas
Recombination bump	ℓ ≈ 80	≈ 2°	Gravitational waves crossing the horizon at z ≈ 1090
Lensing peak	ℓ ≈ 1000	≈ 0.2°	E→B conversion by lensing of intervening structure

Ground-based experiments like BICEP/Keck at the South Pole target the recombination bump at ℓ ≈ 80, where gravitational waves entering the horizon at last scattering leave their largest imprint. The reionization bump at ℓ ≈ 5 lives on the very largest angular scales and is accessible mainly to satellites with full-sky coverage; it is the target of the upcoming LiteBIRD mission. Note that primordial B-modes cut off at small scales (high ℓ) because gravitational waves redshift away once they enter the horizon — which is precisely why the lensing B-mode, rising toward ℓ ≈ 1000, is separable in principle.

The BICEP2 false alarm of 2014

On 17 March 2014, the BICEP2 collaboration announced a detection of primordial B-modes corresponding to r ≈ 0.20, with headlines proclaiming the first direct evidence for inflation and gravitational waves. The excitement was immense — a Nobel Prize seemed in the offing. Within weeks, doubts mounted. BICEP2 had observed at a single frequency, 150 GHz, and had used a model of Galactic dust foregrounds that turned out to underestimate the polarized dust emission in their patch of sky.

Polarized thermal emission from spinning, magnetically aligned interstellar dust grains also produces B-modes, and its spectrum is different from the CMB's: dust brightens toward higher frequencies, the CMB does not. When ESA's Planck satellite released its 353 GHz polarization maps — where dust dominates — a joint BICEP2/Keck/Planck analysis in early 2015 showed that the entire BICEP2 signal was consistent with dust. The "detection" evaporated. The episode is now the canonical lesson in foreground contamination: any credible B-mode experiment must observe at several frequencies to separate the frequency-independent CMB from the frequency-dependent dust and synchrotron foregrounds.

Lensing B-modes: the guaranteed signal

There is a second, completely separate way to make B-modes that has nothing to do with inflation — and it is guaranteed to exist. As the pristine, curl-free E-mode pattern from last scattering travels for 13.8 billion years, its photons are gravitationally deflected by the lumpy intervening matter (galaxy clusters, the cosmic web, dark matter). This shearing remaps the polarization on the sky and converts a small fraction of E-modes into B-modes. The conversion is geometric: lensing does not care about parity, so it leaks power from the abundant E-modes into B.

This lensing B-mode peaks at small angular scales (ℓ ≈ 1000) and was first detected by the South Pole Telescope's SPTpol camera in 2013, then confirmed by POLARBEAR and others. It is a real, now-routine measurement — which is why "B-modes have been detected" is true even though the primordial signal has not. The catch is that on the large scales where the primordial signal lives, the lensing B-mode is a contaminating noise floor. Subtracting it — delensing, using a reconstructed map of the lensing potential — is one of the central technical challenges for next-generation experiments.

Numbers: what experiments have achieved

Experiment	Era	Result on r / B-modes
DASI	2002	First detection of CMB E-mode polarization
BICEP2	2014	Claimed r ≈ 0.20 — later shown to be dust
SPTpol	2013	First detection of lensing B-modes
BICEP2/Keck + Planck	2015	r < 0.12 after dust subtraction
BICEP/Keck (BK18)	2021	r < 0.036 (95%), the current best limit
LiteBIRD (planned)	~2032	Target δr ≈ 0.001, full-sky reionization bump
CMB-S4 (planned)	~2030s	Target σ(r) ≈ 0.001 from deep ground surveys

The march from r < 0.20 to r < 0.036 in seven years already excludes the simplest large-field inflation models, such as a quadratic V(φ) ∝ φ² potential, which predicts r ≈ 0.15. Models like Starobinsky R² inflation predict r ≈ 0.003 — below current sensitivity but squarely within reach of CMB-S4 and LiteBIRD. The next decade is where the question gets answered or the simplest single-field picture gets seriously squeezed.

Common misconceptions and edge cases

• E and B are not real electromagnetic fields. The names are an analogy to the curl-free / divergence-free Helmholtz decomposition of any vector (technically, spin-2 tensor) field on the sphere. There is no actual electric or magnetic field involved in the polarization pattern.
• B-modes have been detected — just not the primordial ones. Lensing B-modes are firmly measured. The unconfirmed, prize-worthy signal is the large-scale primordial B-mode from inflationary gravitational waves.
• r is an upper limit, not a measurement of zero. Current data are consistent with r = 0. That does not rule out inflation; it rules out high-energy-scale inflation models and is consistent with many low-r scenarios. Inflation does not predict a specific r — different potentials predict different values, some far below 0.001.
• Foregrounds are the enemy, not instrument noise. The fundamental obstacle is astrophysical: polarized Galactic dust (dominant above ~100 GHz) and synchrotron emission (dominant below ~70 GHz). Beating them requires many frequency bands, not just more integration time at one frequency. BICEP2 learned this the hard way.
• Cosmic birefringence can mimic part of the signal. A hypothetical parity-violating rotation of the polarization plane as light crosses the universe would also convert E into B. Searches for this rotation angle are an active, separate program and a possible systematic for the inflationary search.
• Gravitational waves redshift away inside the horizon. This is why the primordial B-mode is confined to large angular scales (low ℓ) — once a tensor mode enters the horizon it oscillates and decays, so it cannot imprint polarization on small scales the way density modes do.