B-Mode Polarization

A handedness the universe wasn't supposed to have

Hold a mirror up to most patterns in the cosmic microwave background and nothing changes — they are symmetric. The temperature map looks the same reflected; so does the smooth, "combed" part of its polarization. But there is a second polarization pattern that does not survive the mirror: a set of tiny swirls, each with a definite clockwise-or-counter-clockwise twist, that flips handedness when reflected. That parity-odd, curl-shaped pattern is a B-mode. The reason cosmologists care so much is brutally simple: the ordinary density ripples that make galaxies and the temperature spots in the CMB are physically incapable of producing it. So if a B-mode is there on the largest scales, something exotic wrote it — and the leading candidate is a sea of gravitational waves shaken out of spacetime itself during inflation, in the first 10⁻³⁴ of a second.

"B-mode" is a name borrowed from vector calculus, not from any magnetic field in the sky. The CMB's polarization, like any pattern of headless rods on a sphere, splits cleanly into a gradient piece and a curl piece — exactly the way any vector field splits into the gradient of a potential (curl-free, like an electrostatic E field) plus the curl of a vector potential (divergence-free, like a magnetostatic B field). The gradient piece is the E-mode; the curl piece is the B-mode. The analogy is mathematical, but it captures the essential geometry: E-modes radiate or wrap symmetrically; B-modes swirl with a handedness.

How the CMB gets polarized at all

The CMB is faintly polarized — only a few percent of the level of its temperature fluctuations — and that polarization is generated by a single, beautiful mechanism: Thomson scattering of light off free electrons. Thomson scattering is the elastic scattering of a photon by a free electron, with the famous angle-dependent cross section

dσ/dΩ = (3 σ_T / 16π) (1 + cos²θ),   σ_T = 6.652 × 10⁻²⁵ cm²

The key fact is that a scattered photon comes out polarized perpendicular to the plane containing the incoming and outgoing rays. If an electron is bathed in perfectly isotropic light, all those polarizations average to zero and the scattered light is unpolarized. The same is true if the surrounding light has only a dipole (hotter on one side, colder on the other). Net polarization appears only if the incoming radiation has a quadrupole anisotropy — hotter from two opposite directions and colder from the perpendicular pair. Then the electron re-radiates more of the field along one axis than the other, and the outgoing light is linearly polarized.

This is why CMB polarization is a probe of the universe at exactly one instant. Quadrupoles only develop in the photon-baryon fluid as it begins to free-stream, which happens during recombination at redshift z ≈ 1090 (about 380,000 years after the Big Bang), when the universe cooled to T ≈ 3000 K and protons and electrons combined into neutral hydrogen. Before that, scattering was so frequent that any quadrupole was instantly washed out; after it, there were too few free electrons to scatter. Polarization is thus generated in a thin shell, the "last-scattering surface," with a generating mechanism that is sensitive to the velocity and metric perturbations there.

Why scalars make only E-modes, and tensors make both

Linear perturbations of the universe sort into three types by how they transform under rotations: scalar (density), vector (vorticity), and tensor (gravitational waves). What B-modes reveal is which of these was present at last scattering.

A scalar density perturbation has only one special direction at any point — the gradient of the density, "downhill." A polarization pattern built from a single axis can only be aligned with that axis, so it is purely a gradient: a curl-free E-mode. Around a hot spot the rods point radially or tangentially; the pattern is mirror-symmetric. There is no way to extract a handedness from a single preferred direction, so scalars are mathematically forbidden from making B-modes (to linear order — second-order lensing is a separate story below).

A tensor perturbation — a gravitational wave — is different. It is a transverse, traceless stretching of space: as the wave passes, it squeezes space along one axis and stretches it along the perpendicular axis, then swaps, and the two polarization states (the "+" and "×" modes) are rotated 45° from each other. That intrinsic shear and handedness imprints a quadrupole on the scattering electrons that has a curl component. The result: gravitational waves generate both E- and B-modes, in a characteristic ratio. A detected large-scale B-mode therefore points almost uniquely at a primordial tensor field.

Perturbation type	Symmetry	Makes E-modes?	Makes B-modes?	Cosmological source
Scalar (density)	parity-even, 1 axis	Yes (dominant)	No (linear order)	Inflaton density fluctuations
Vector (vorticity)	parity-odd	Yes	Yes	Decays in standard cosmology; cosmic strings
Tensor (grav. waves)	parity-odd, handed	Yes	Yes	Inflationary gravitational waves
Lensing (2nd order)	shears E into B	—	Yes (guaranteed)	Large-scale structure deflection

The mathematics: Stokes parameters and a spin-2 field

Linear polarization at a point on the sky is described by the Stokes parameters Q and U, measured relative to a chosen pair of axes. (The circular-polarization parameter V is essentially zero for the CMB, since Thomson scattering does not produce it.) Under a rotation of the measurement axes by an angle ψ, Q and U mix as

(Q ± iU)′ = e∓²ⁱᵠ (Q ± iU)

The factor of 2 in the exponent is the signature of a spin-2 field: rotate your ruler by 180° and the pattern is unchanged, just as a headless rod looks the same flipped end-for-end. Because Q and U depend on an arbitrary choice of axes, you cannot make a coordinate-independent map directly from them. The fix, introduced by Marc Kamionkowski, Arthur Kosowsky and Albert Stebbins, and independently by Matias Zaldarriaga and Uroš Seljak in 1997, is to expand Q ± iU in spin-weighted spherical harmonics and recombine them into two scalar fields that are rotation-invariant:

E(n̂) ∝ ∂² (gradient part)     →  parity-even, curl-free
B(n̂) ∝ ∇×  (curl part)        →  parity-odd, divergence-free

From these one builds angular power spectra — C_ℓ^EE, C_ℓ^BB, and the cross-spectrum C_ℓ^TE between temperature and E-modes. Parity forbids the cross-spectra C_ℓ^TB and C_ℓ^EB in the standard model, so a nonzero TB or EB would itself be a smoking gun for parity-violating ("cosmic birefringence") physics — a hint of which Planck data tentatively showed in 2020 at about 2–3σ, still unconfirmed.

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

The amplitude of the primordial gravitational-wave background is parameterised by the tensor-to-scalar ratio

r = A_t / A_s   (evaluated at pivot scale k = 0.05 Mpc⁻¹)

where A_s ≈ 2.1 × 10⁻⁹ is the measured scalar amplitude. The power of r is that it converts almost directly into the energy scale of inflation, because in single-field slow-roll inflation the tensor amplitude is set by the Hubble rate during inflation, and hence by the potential energy V of the inflaton:

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

So r = 0.01 corresponds to inflation occurring at a Grand-Unified-Theory energy scale of about 10¹⁶ GeV — a regime utterly inaccessible to any particle accelerator (the LHC reaches ~10⁴ GeV). A B-mode detection would be, in effect, a probe of physics a trillion times beyond collider energies. The current 95%-confidence bound, combining BICEP/Keck 2018 with Planck and WMAP, is r < 0.032 (the BICEP/Keck headline number is r < 0.036). That already excludes the simplest "large-field" models such as V ∝ φ², while leaving plateau models like Starobinsky R² inflation (which predicts r ≈ 0.003) comfortably alive — and below the reach of current instruments.

Two bumps: recombination and reionization

A primordial B-mode signal, if it exists, would not appear at a single angular scale. Gravitational waves leave their polarization imprint at two epochs, producing two bumps in the C_ℓ^BB spectrum:

• The recombination bump at ℓ ≈ 80 (angular scales of a couple of degrees). This comes from the last-scattering surface itself. It is where degree-scale experiments such as BICEP point their telescopes.
• The reionization bump at ℓ < 10 (tens of degrees, nearly the whole sky). When the first stars reionized the universe around z ≈ 7–8, a fresh population of free electrons re-scattered CMB photons and re-imprinted the gravitational-wave quadrupole at the largest scales. Measuring this requires near-full-sky coverage, which is why satellites (and the proposed LiteBIRD mission) target it.

Sitting on top of, and ultimately dominating, the small-scale end is the lensing B-mode, which peaks at ℓ ≈ 1000. Disentangling these by angular scale is the core observational strategy of the entire field.

Worked example: how faint is the signal?

Suppose the true tensor-to-scalar ratio is r = 0.01, near the Starobinsky-model value and a typical target for the next generation of experiments. How large is the B-mode that we would have to measure?

The temperature anisotropy of the CMB is about ΔT ≈ 100 μK (root-mean-square, at degree scales) on a mean of T₀ = 2.7255 K. The E-mode polarization is roughly an order of magnitude smaller, a few μK. The primordial B-mode at the recombination bump for r = 0.01 has an amplitude of only

[ℓ(ℓ+1) C_ℓ^BB / 2π]^(1/2) ≈ 0.03 μK   at ℓ ≈ 80, for r = 0.01

That is about 30 nanokelvin — roughly one part in 10⁸ of the CMB's mean temperature, and about one ten-thousandth of the temperature anisotropy. Detecting it means controlling instrumental systematics and Galactic foregrounds to a level far below that. The polarized dust emission that doomed BICEP2 is itself of order 0.1 μK at 150 GHz in even the cleanest patches of sky — larger than the signal — which is why modern experiments observe at many frequencies (BICEP/Keck spans 95, 150 and 220 GHz; Planck mapped 30–353 GHz) to model and subtract the dust and synchrotron foregrounds. The measurement is one of the most demanding in all of observational physics.

Lensing B-modes: the guaranteed signal and the delensing problem

There is one source of B-modes we are certain exists, regardless of inflation: gravitational lensing. As CMB photons travel 13.8 billion light-years to us, the gravity of intervening galaxy clusters and dark-matter filaments deflects them by a few arcminutes. That deflection field is not uniform — it shears the polarization pattern — and shearing a pure E-mode mixes in a little B-mode. This lensing B-mode is guaranteed by the mere existence of cosmic structure.

It was first detected by the South Pole Telescope's SPTpol camera in 2013 (in cross-correlation with a cosmic-infrared-background tracer of the lensing mass), and measured in pure CMB autocorrelation by the POLARBEAR experiment in 2014. The lensing B-mode peaks at ℓ ≈ 1000, well separated from the recombination bump at ℓ ≈ 80, but its broad tail is the dominant contaminant when chasing very small r. Future experiments therefore plan to "delens": use a high-resolution map of the lensing potential (from the CMB itself or from galaxy surveys) to model and subtract the lensing-induced B-modes, cleaning the field to expose any primordial signal beneath. Delensing is expected to be essential for reaching r ~ 10⁻³.

The chase: missions, numbers, and people

The hunt for B-modes is a 30-year arc of steadily improving limits and one famous false start.

Experiment	Type / site	Era	Milestone
DASI	Interferometer, South Pole	2002	First detection of CMB E-mode polarization
WMAP	Satellite	2003–2010	TE cross-spectrum; reionization optical depth
QUaD / BICEP1	Ground, South Pole	2006–2010	First degree-scale B-mode upper limits
SPTpol	Ground, South Pole	2013	First detection of lensing B-modes
POLARBEAR	Ground, Atacama	2014	First B-mode autocorrelation (lensing)
BICEP2	Ground, South Pole	2014	Claimed r ≈ 0.2 — later attributed to dust
Planck	Satellite	2015–2018	All-sky dust maps; foreground separation
BICEP/Keck + Planck	Joint	2021	r < 0.036 (own), r < 0.032 with WMAP
Simons Observatory	Ground, Atacama	2024–	Targeting σ(r) ≈ 0.003
LiteBIRD	Satellite (JAXA)	~2032 (planned)	Full-sky reionization bump, σ(r) < 0.001
CMB-S4	Ground (US)	~2030s (planned)	Detect or exclude r ≳ 0.003

The theoretical foundation was laid in 1997 by two groups working independently: Marc Kamionkowski, Arthur Kosowsky and Albert Stebbins, and Matias Zaldarriaga and Uroš Seljak, who first showed that the polarization splits cleanly into E and B and that B isolates the tensor signal. The connection of inflation to a gravitational-wave background goes back to Alexei Starobinsky (1979) and the inflationary models of Alan Guth, Andrei Linde, Andreas Albrecht and Paul Steinhardt in the early 1980s.

The BICEP2 episode of 2014

On 17 March 2014, the BICEP2 collaboration announced from Harvard a detection of degree-scale B-modes at r ≈ 0.2 — heralded worldwide as direct evidence of inflationary gravitational waves and a likely Nobel Prize. Within months the result unravelled. BICEP2 observed at a single frequency, 150 GHz, in a patch of sky that turned out to contain more polarized Galactic dust than assumed. Polarized thermal emission from elongated, spinning dust grains aligned by the Galactic magnetic field produces its own B-mode pattern, and at 150 GHz it can fully mimic a primordial signal. A 2015 joint analysis of BICEP2/Keck data with Planck's 353 GHz dust maps showed that the observed signal was consistent with dust alone; the primordial detection evaporated. The lasting lesson — now built into every experiment — is that multi-frequency observation and rigorous foreground modelling are not optional. The story is the canonical case study in how a real, well-built instrument can still be fooled by the Galaxy in the foreground.

Common misconceptions and subtleties

• B-modes are not magnetic fields. The "B" is a vector-calculus analogy (curl part, like a magnetostatic field), not a measurement of any magnetic field in the CMB. The polarization is the orientation of an electric field vector, with no preferred sign.
• A B-mode detection is not automatically inflation. Lensing produces B-modes guaranteed; cosmic strings and primordial magnetic fields can produce them too. Only the large-scale (ℓ ≈ 80 and ℓ < 10) primordial bump, after foreground and lensing removal and ideally with the right spectral and scale dependence, points specifically at inflationary gravitational waves.
• E and B are non-local on a cut sky. The clean E/B split is exact only on the full sphere. On a small observed patch, the decomposition leaks ("E-to-B leakage"), and dedicated "pure-B" estimators are needed — a major practical headache for ground-based experiments.
• r is not a probability or an upper limit by nature. r is a physical amplitude that could be anything from ~0 up. We have only ever measured upper bounds because the signal, if present, is below current sensitivity; r < 0.032 does not mean inflation is disfavoured, only that high-energy large-field models are.
• Polarization is generated, not just rotated, at last scattering. The CMB is not polarized at emission and then twisted en route (except by lensing and possible birefringence). The primordial polarization is created at recombination by the local quadrupole, which is why it is such a sharp snapshot of that epoch.