Stochastic Gravitational-Wave Background

The hum, not the chirp

When LIGO detected its first gravitational wave in 2015, it heard a chirp: a single, clean, rising tone from two black holes spiralling together over a fraction of a second. The stochastic gravitational-wave background is the opposite kind of signal. It is not one source but a crowd — the overlapping emission of so many gravitational-wave sources, scattered across the sky and across cosmic history, that no single one can be picked out. Their waves arrive on top of each other with random, uncorrelated phases, and the sum looks like noise: a featureless, hissing, statistically stationary rumble filling the detector at all times.

That statistical character is the whole challenge and the whole point. A chirp announces itself; you can see it cross a time-frequency plot. The background hides inside the instrument's own noise, indistinguishable from it on a single detector. The only thing that betrays it is correlation. The same background strain stretches and squeezes two widely separated detectors in a way that depends on the angle between them, while their internal noise is independent. Dig out that shared, angle-dependent component and you have measured the universe's gravitational hum — even though you can never point to the speaker making it.

What sources superpose

The background is not a single phenomenon but a sum over every gravitational-wave emitter the universe has ever hosted, weighted by how loud and how numerous they are at a given frequency. Two broad families contribute:

• Astrophysical background. The incoherent pile-up of discrete objects too faint or too numerous to resolve individually. In the nanohertz band this means inspiralling supermassive black hole binaries (10⁸–10¹⁰ M☉ pairs left orbiting after galaxy mergers). At LIGO's ~100 Hz band it is the cumulative drone of every stellar-mass black hole and neutron star merger that has ever happened across the observable universe — roughly one such merger somewhere every few minutes.
• Cosmological background. Relic gravitational waves from the early universe: quantum fluctuations stretched by inflation into a primordial spectrum, vibrating cosmic strings radiating from cusps and kinks, or bubble collisions during a first-order phase transition. These would be a genuine fossil from times no light can reach — emitted when the universe was a fraction of a second old, free-streaming undisturbed ever since.

Crucially, the same observed band can mix both families. The 2023 pulsar-timing signal is most naturally explained by supermassive black hole binaries, but its amplitude sits at the high end of those predictions, leaving room for a cosmological contribution. Disentangling them is the central open question of nanohertz astronomy.

How we describe a random signal

Because the background has no waveform to write down, we characterise it statistically — by how its power is distributed across frequency. The standard measure is the dimensionless energy-density spectrum, the fraction of the universe's critical density carried by gravitational waves per logarithmic frequency interval:

Ω_GW(f) = (1 / ρ_c) · dρ_GW / d ln f

ρ_c = 3 H₀² c² / (8πG) ≈ 8.5 × 10⁻¹⁰ J/m³   (critical density today)

An equivalent, more observational quantity is the characteristic strain h_c(f), the typical fractional stretching of space at frequency f. The two are related by

Ω_GW(f) = (2π² / 3 H₀²) · f² · h_c(f)²

For a population of circular, gravitational-wave-driven supermassive black hole binaries, a classic result (Phinney 2001) predicts a power law set purely by orbital evolution:

h_c(f) = A · (f / f_yr)^(−2/3)

f_yr = 1 / (1 yr) ≈ 31.7 nHz       (the conventional reference frequency)

The −2/3 spectral index is not a fit; it falls out of the fact that binaries spend more time (and so emit more cycles) at lower frequencies, and that energy is carried away as the orbit shrinks. Measuring whether the real background follows f⁻²ᐟ³ is one way to test the supermassive-binary hypothesis against more exotic spectra.

The Hellings-Downs fingerprint

A single pulsar's timing residuals can be reddened by many things — intrinsic spin noise, interstellar plasma, a wandering reference clock. None of those tell you the cause is gravitational. The decisive test is to compare pairs of pulsars. An isotropic gravitational-wave background induces a very specific correlation between the timing residuals of two pulsars as a function of their angular separation γ on the sky — the Hellings-Downs curve, derived in 1983:

Γ(γ) = (1/2) − (1/4) x + (3/2) x ln x ,   where x = (1 − cos γ) / 2

  γ = 0°   →  Γ = +0.5     (a pulsar is maximally correlated with itself)
  γ ≈ 82°  →  Γ ≈ −0.15    (anti-correlation minimum)
  γ = 180° →  Γ ≈ +0.25    (antipodal pulsars partly re-correlate)

This quadrupolar pattern is a direct consequence of the spin-2, transverse nature of gravitational radiation in general relativity. Instrumental noise produces no such angular structure; a clock error produces a monopole; an ephemeris error produces a dipole. Only a gravitational-wave background bends the correlation into the Hellings-Downs shape. The 2023 announcements were, at heart, the statement that the measured pulsar-pair correlations trace this curve at the 3–4σ level.

The spectrum across detectors

The background spans more than twenty orders of magnitude in frequency, and different instruments probe different decades. Each is sensitive to a different mix of sources, because the dominant emitters change with frequency.

Detector / probe	Frequency band	Dominant sources	Method	Status
CMB B-modes (BICEP/Keck, LiteBIRD)	~10⁻¹⁸–10⁻¹⁶ Hz	Inflationary relics	Curl polarisation pattern	r < 0.036 (upper limit)
Pulsar-timing arrays (NANOGrav, EPTA, PPTA, CPTA)	~1–100 nHz	SMBH binaries, cosmic strings	Hellings-Downs correlation	Evidence, 2023 (3–4σ)
LISA (space, launch ~2035)	~0.1 mHz – 0.1 Hz	Galactic white-dwarf binaries, phase transitions	Time-delay interferometry	In development
LIGO / Virgo / KAGRA (ground)	~10–1000 Hz	Unresolved stellar-mass mergers	Two-detector cross-correlation	Ω_GW ≲ 6 × 10⁻⁹ at 25 Hz
Einstein Telescope / Cosmic Explorer	~5–2000 Hz	Mergers to z > 10, phase transitions	Cross-correlation, null streams	Proposed

One signal, many windows. The supermassive binaries that pulsar timing hears at nanohertz are the giant cousins of the stellar-mass mergers LIGO chases at hundreds of hertz; the LISA millihertz band catches the tens of millions of close white-dwarf binaries in our own Galaxy, whose combined hum is so strong it forms a confusion-limited foreground that LISA must model and subtract.

The numbers: a hum you measure in nanoseconds

How loud is the nanohertz background? The NANOGrav 15-year dataset (2023) reports a characteristic strain amplitude of

A ≈ 2.4 × 10⁻¹⁵   at f = f_yr = 1/yr   (median, with sizable uncertainty)
h_c(f) ≈ A (f/f_yr)^(−2/3)

A strain of 10⁻¹⁵ means spacetime is being stretched and squeezed by one part in a quadrillion. Over the ~1 kpc to a typical timing pulsar, that periodically shifts the pulse arrival time by only tens to hundreds of nanoseconds — which is why the detection demanded fifteen years of pulse timing across dozens of millisecond pulsars (NANOGrav used 68 pulsars), each a natural clock stable to better than a microsecond per pulse averaged over weeks. Converting to energy density gives roughly

Ω_GW(f_yr) h² ~ 10⁻⁹   (order of magnitude, nanohertz band)

By comparison, LIGO's tightest constraint at 25 Hz is Ω_GW ≲ 6 × 10⁻⁹, and the inflationary background is expected to be many orders of magnitude fainter still (Ω_GW h² ≲ 10⁻¹⁵ for slow-roll inflation), which is why it has not been detected. The largest known background by far is the cosmic microwave background of photons at Ω ≈ 5 × 10⁻⁵ — the gravitational-wave universe is fantastically quiet.

Where it shows up

• The 2023 nanohertz announcement. On 29 June 2023 NANOGrav, the European, Parkes, Chinese, and Indian pulsar-timing collaborations simultaneously released datasets showing evidence for a Hellings-Downs-correlated background — the first detection of the SGWB in any band, and the opening of nanohertz gravitational-wave astronomy.
• Supermassive black hole binaries. If the signal is astrophysical, it is the integrated voice of millions of SMBH pairs across cosmic time. Its amplitude constrains the merger rate of galaxies and the masses of their central black holes — a probe of structure formation gravitational waves alone can deliver.
• LIGO/Virgo cross-correlation searches. Every observing run sets a deeper upper limit on the high-frequency astrophysical background built from unresolved compact-object mergers. The predicted level (Ω_GW ~ 10⁻⁹ at tens of Hz) is now tantalisingly close to detection in future runs.
• The hunt for primordial B-modes. BICEP/Keck and the forthcoming LiteBIRD and CMB-S4 experiments search the microwave sky for the faint curl-pattern polarisation that an inflationary gravitational-wave background would have imprinted — the most sensitive route to relic spacetime ripples from 10⁻³⁶ s after the Big Bang.
• Cosmic-string and phase-transition limits. Non-detection at a given amplitude already rules out the loudest cosmic-string and early-universe phase-transition scenarios, turning the background's quietness into a constraint on physics far beyond the Standard Model.

Common misconceptions and edge cases

• "It's just detector noise." Individually it is statistically identical to noise — that is exactly why a single detector cannot find it. The discriminator is the cross-detector (or cross-pulsar) correlation with the right angular shape, which noise cannot fake.
• "The background is isotropic and featureless, full stop." Only as a leading approximation. A finite number of nearby, especially massive binaries can make the sky slightly anisotropic, and the loudest single binary may eventually poke above the background as a resolvable "continuous wave." Anisotropy searches are a live frontier.
• "Pulsar timing and LIGO measure the same waves." Same physics, vastly different frequencies. PTAs are sensitive to periods of years (nanohertz, from billion-solar-mass binaries); LIGO to milliseconds (hundreds of hertz, from tens-of-solar-mass binaries). They probe entirely different source populations.
• "The primordial background is the CMB." The CMB is light from 380,000 years after the Big Bang. A primordial gravitational-wave background is spacetime ripples from ~10⁻³⁶ s, detectable only indirectly through CMB B-mode polarisation — and still undetected. The tensor-to-scalar ratio bound is r < 0.036.
• "More pulsars just means more sensitivity, linearly." Sensitivity to the Hellings-Downs correlation scales roughly with the number of pulsar pairs, which grows as N(N−1)/2 — quadratically — and timing precision improves with the square root of observing time. This is why decades-long datasets across dozens of pulsars were the threshold, not a few years across a handful.