Stochastic Gravitational-Wave Background

A stochastic gravitational-wave background is the gravitational analogue of the cosmic microwave background: radiation from so many overlapping sources, or from processes so spread across the sky and across cosmic time, that no single event can be resolved. Once the metric perturbation $h_{ab}(t,\vec x)$ is a superposition of countless uncorrelated wavelets, its instantaneous waveform is meaningless — what survives, and what carries all the physics, are its statistical moments. Under the standard idealizations of stationarity, isotropy, unpolarization, and Gaussianity, the entire signal collapses onto one function of frequency: the power spectrum, most usefully packaged as the dimensionless energy density $\Omega_{\mathrm{gw}}(f)$ . Two families of source feed the hum. Astrophysical backgrounds are the incoherent pile-up of weak or distant events — compact-binary inspirals, spinning neutron stars, supernovae — encoding the integrated star-formation and merger history of the Universe. Cosmological backgrounds come from the first instants after the Big Bang: vacuum tensor fluctuations amplified by inflation, turbulence in a first-order phase transition, oscillating cosmic-string loops. Because gravitons decouple essentially at the Planck time, such a background streams to us unscattered — an undistorted photograph of physics at energies no accelerator will ever reach. The background is detected differently from any chirp. A lone detector sees it as excess noise identical to its own hiss, so one correlates two detectors: the signal is common while the noises are independent, and the cross-correlation grows with patience. The geometric efficiency of that correlation is the overlap reduction function, and for pulsar pairs it takes a universal quadrupolar shape — the Hellings-Downs curve whose detection in 2023 opened the nanohertz sky.

Key takeaways

A stochastic gravitational-wave background is a random, statistically isotropic superposition of waves from unresolved sources — early-universe processes and the summed population of astrophysical binaries — described by a power spectrum, not a waveform.
The standard measure is the dimensionless energy density $\Omega_{\mathrm{gw}}(f)=\rho_c^{-1}\,\mathrm{d}\rho_{\mathrm{gw}}/\mathrm{d}\ln f$ , built from the Isaacson stress-energy of the metric perturbation and normalized by the critical density.
A single detector cannot separate the background from its own noise, so it is measured by cross-correlating two or more detectors, with the geometric efficiency set by the overlap reduction function $\gamma(f)$ .
For pulsar timing that correlation has a universal quadrupolar shape, the Hellings-Downs curve $\zeta(\theta)$ ; matching it across many pulsar pairs is the fingerprint of a genuine spin-2 background.
Cosmological backgrounds — inflation, phase transitions, cosmic strings — probe energies far beyond colliders and propagate freely from the Planck era, a window earlier than the cosmic microwave background.
Key idea: a stochastic background is not a waveform but a power spectrum — model the strain as a Gaussian random field, characterize it by $\Omega_{\mathrm{gw}}(f)$ from the Isaacson energy density, and detect it by cross-correlating independent detectors weighted by the overlap reduction function.

The core insight

A stochastic background is not a waveform but a power spectrum: model the strain as a Gaussian random field, characterize it by the dimensionless energy density $\Omega_{\mathrm{gw}}(f)$ derived from the Isaacson stress-energy, and detect it by cross-correlating independent detectors weighted by the overlap reduction function.

Where this lives today

The stochastic background has just crossed from prediction into observation, and it now anchors a fast-moving program. In the nanohertz band the pulsar timing arrays — NANOGrav and the European, Parkes, and Chinese arrays, now merging through the International Pulsar Timing Array — are sharpening the 2023 evidence toward a confident detection and asking whether the source is purely supermassive black-hole binaries or carries a cosmological admixture. In the audio band, LIGO-Virgo-KAGRA push their upper limit toward the $\Omega_{\mathrm{gw}}\sim10^{-9}$ level where unresolved stellar-mass mergers should appear, a signal Einstein Telescope and Cosmic Explorer should resolve. In the millihertz band, LISA will confront a guaranteed Galactic white-dwarf foreground while reaching for primordial backgrounds down to $\Omega_{\mathrm{gw}}\sim10^{-12}$ , a direct probe of electroweak-scale physics. At the femtohertz extreme, CMB experiments tighten the tensor-to-scalar ratio $r$ , bounding the inflationary energy scale below $\sim10^{16}\ \mathrm{GeV}$ . Across every band, the common language is the single spectrum $\Omega_{\mathrm{gw}}(f)$ .

The physical picture

Imagine a packed stadium where every person hums at a random pitch and phase. You cannot pick out any one voice, yet the total acoustic energy at each frequency is well defined, and two microphones set side by side register correlated rumble while their own electronic hiss stays independent. A stochastic background is exactly this for spacetime ripples: a statistical hum whose energy spectrum — not whose waveform — is the observable. Because gravitons interact almost not at all, a hum from the first instants after the Big Bang reaches us unscattered; the microwave background shows the Universe at 380 000 years, but a cosmological background would image it at $10^{-30}$ seconds, a deeper if fainter photograph. That hum spreads across twenty decades of frequency, and different instruments listen in different bands. The first figure is the working map of the whole field: the predicted and constrained level of $\Omega_{\mathrm{gw}}(f)$ from the femtohertz of the microwave background to the kilohertz of ground-based interferometers, each observatory pinned to a real number.

Figure

The stochastic background across twenty decades of frequency: inflation and the cosmic microwave background at femtohertz probe Ω ≲ 10⁻¹⁵, pulsar timing arrays report Ω_gw h² ≈ 2×10⁻⁹ near a nanohertz, LISA will reach Ω ≈ 10⁻¹² at a millihertz, and LIGO-Virgo-KAGRA bound Ω ≲ 10⁻⁹ near 25 Hz

The trick that turns an invisible hum into a measurement is correlation. In a single detector the background is buried in noise, but the cross-correlation of two detectors accumulates only the common gravitational part, so its signal-to-noise grows as $\sqrt T$ with observation time — patience and bandwidth substitute for raw strength, and a background far below any single instrument's noise floor can still be dug out. For a pulsar timing array the correlation carries a decisive angular signature: it depends only on the angular separation of two pulsars, and for an isotropic spin-2 field it traces one specific quadrupolar curve. The second figure is that curve — the fingerprint the arrays hunt for.

Figure

The Hellings-Downs correlation ζ(θ) for an isotropic spin-2 background: correlation falls from ½ at zero separation, crosses zero near 49°, reaches a minimum of about −0.15 near 82° (an anticorrelation), then rises to +¼ at 180° — the quadrupolar shape that distinguishes a gravitational background from monopole clock errors or dipole ephemeris errors

The core equations

The starting point is a mode expansion: the background is a sum of plane waves of both polarizations, arriving from every direction, with random amplitudes. Writing $\hat n$ for the propagation direction and $A\in\{+,\times\}$ for the polarization, the transverse-traceless metric perturbation is a Fourier integral over frequency and sky.

h_{ab}(t,\vec x)=\sum_{A}\int_{-\infty}^{\infty}\!\mathrm{d}f\!\int_{S^2}\!\mathrm{d}^2\hat n\;\tilde h_A(f,\hat n)\,e^A_{ab}(\hat n)\,e^{2\pi i f(t-\hat n\cdot\vec x/c)} \qquad (1)

where: $\tilde h_A(f,\hat n)$ — random Fourier amplitude of polarization $A$ from direction $\hat n$ ; $e^A_{ab}(\hat n)$ — transverse-traceless polarization tensors; $\hat n\cdot\vec x/c$ — retardation along the line of sight; the integral over $\pm f$ enforces reality. The field is deterministic in form but random in its coefficients — everything statistical lives in $\tilde h_A$ . Because the amplitudes are random, the physics is in their two-point function. Stationarity forces a $\delta(f-f')$ , isotropy a $\delta^2(\hat n,\hat n')$ , unpolarization a $\delta_{AA'}$ , and the remaining function of frequency is the strain power spectral density $S_h(f)$ .

\langle\tilde h_A(f,\hat n)\,\tilde h^*_{A'}(f',\hat n')\rangle=\delta_{AA'}\,\delta^2(\hat n,\hat n')\,\delta(f-f')\,\tfrac{1}{8\pi}S_h(f) \qquad (2)

where: $\langle\cdot\rangle$ — ensemble average over realizations; $S_h(f)$ — one-sided strain PSD, units $\mathrm{Hz}^{-1}$ ; the three Kronecker or Dirac deltas encode the four standard idealizations. A single function $S_h(f)$ specifies the whole background, and the factor $\tfrac{1}{8\pi}=\tfrac{1}{4\pi}\cdot\tfrac12$ spreads the one-sided power isotropically over the sphere and splits it between the $\pm f$ halves. To convert strain into energy we need the energy density of the wave field. Gravitational-wave energy is not local, but the Isaacson short-wave average over several wavelengths gives an effective density set by the mean-square time derivative of the strain.

\rho_{\mathrm{gw}}=\frac{c^2}{32\pi G}\,\langle\dot h_{ab}\,\dot h^{ab}\rangle \qquad (3)

where: $\dot h_{ab}$ — time derivative of the TT perturbation; $\langle\cdot\rangle$ — average over many wavelengths; $c,G$ — light speed and Newton's constant. This is the object that makes "how much energy is in the ripples" well-posed, and the only place general relativity (through Isaacson) enters the spectrum. The standard measure normalizes that energy density per logarithmic frequency by the critical density, giving a dimensionless spectrum comparable across every band at once.

\Omega_{\mathrm{gw}}(f)\equiv\frac{1}{\rho_c}\frac{\mathrm{d}\rho_{\mathrm{gw}}}{\mathrm{d}\ln f},\qquad \rho_c=\frac{3H_0^2 c^2}{8\pi G} \qquad (4)

where: $\rho_c$ — critical density; $H_0$ — Hubble constant; $\mathrm{d}\rho_{\mathrm{gw}}/\mathrm{d}\ln f$ — energy per logarithmic frequency interval. Because $\rho_c\propto H_0^2$ carries the Hubble uncertainty, limits are quoted as $\Omega_{\mathrm{gw}}h^2$ with $h=H_0/(100\,\mathrm{km\,s^{-1}\,Mpc^{-1}})$ . Combining (2)–(4) gives the workhorse relation between the spectrum and the strain spectral density, derived below. Its steep $f^3$ weighting is the single most important fact about the two descriptions.

\Omega_{\mathrm{gw}}(f)=\frac{4\pi^2}{3H_0^2}\,f^3\,S_h(f)=\frac{2\pi^2}{3H_0^2}\,f^2\,h_c^2(f),\qquad h_c^2(f)\equiv 2f\,S_h(f) \qquad (5)

where: $h_c(f)$ — characteristic strain, a dimensionless amplitude per logarithmic frequency; the two forms are identical given $h_c^2=2fS_h$ . A flat strain spectrum is a steeply rising energy spectrum; pulsar timing quotes $h_c$ while interferometers quote $\Omega_{\mathrm{gw}}$ . Detection geometry is captured by the overlap reduction function: the sky- and polarization-averaged product of the two detectors' antenna patterns, phased by the light-travel delay across their separation.

\gamma(f)=\frac{5}{8\pi}\int_{S^2}\!\mathrm{d}^2\hat n\sum_A F_1^A(\hat n)\,F_2^A(\hat n)\,e^{2\pi i f\,\hat n\cdot\Delta\vec x/c} \qquad (6)

where: $F_I^A(\hat n)$ — antenna response of detector $I$ to polarization $A$ ; $\Delta\vec x$ — separation vector; the normalization sets $\gamma(0)=1$ for coincident, coaligned interferometers. It decays and oscillates once the wavelength is shorter than the separation, defining the usable correlation band. For pulsar pairs the same geometric average, carried out analytically over an isotropic background, gives a closed-form function of angular separation alone — the Hellings-Downs curve.

\zeta(\theta)=\frac{1}{2}+\frac{3}{2}\,x\ln x-\frac{x}{4},\qquad x=\frac{1-\cos\theta}{2} \qquad (7)

where: $\theta$ — angular separation of two pulsars; $x$ — the "chord" variable; $\zeta$ is normalized to $\tfrac12$ at zero separation (excluding the pulsar-term self-correlation). The shape — down through zero near $49°$ , a shallow anticorrelation minimum near $82°$ , up to $+\tfrac14$ at $180°$ — is fixed by the spin-2 nature of gravity and nothing else.

Derivation AFrom the Isaacson energy density to the dimensionless spectrum

The goal is to turn the abstract statement "the background has strain power $S_h(f)$ " into the number an observer reports, $\Omega_{\mathrm{gw}}(f)$ , and to see precisely where the steep $f^3$ frequency weighting comes from. We start from the Isaacson short-wave-averaged energy density, insert the mode expansion, use the statistical specification to collapse the sky-and-polarization integral onto $S_h(f)$ , differentiate per logarithmic frequency, and divide by the critical density.

Step 1. Write the energy density from Isaacson, $\rho_{\mathrm{gw}}=(c^2/32\pi G)\langle\dot h_{ab}\dot h^{ab}\rangle$ — the effective stress-energy of the ripples averaged over several wavelengths. Why: without a gravitational-wave energy density there is no $\rho_{\mathrm{gw}}$ to normalize, and $\Omega_{\mathrm{gw}}$ would be empty.

Step 2. Insert the mode expansion (1) and apply the two-point function (2). Each time derivative brings a factor $(2\pi f)$ , the sky integral of the deltas removes one direction integral, and $\sum_A e^A_{ab}e^{A\,ab}=4$ , giving $\langle\dot h_{ab}\dot h^{ab}\rangle=4\int_0^\infty\!\mathrm{d}f\,(2\pi f)^2 S_h(f)$ . Why: this converts a random field into a deterministic integral over its spectrum — the ensemble average has erased the random phases.

Step 3. Differentiate the integrand and multiply by $f$ to pass from $\mathrm{d}f$ to $\mathrm{d}\ln f$ : $\mathrm{d}\rho_{\mathrm{gw}}/\mathrm{d}\ln f=(c^2/32\pi G)(2\pi f)^2\,4f\,S_h(f)=(\pi c^2/2G)\,f^3 S_h(f)$ . Why: the extra $f$ from the logarithmic measure promotes the two derivative powers into the cubic weighting that dominates the physics.

Step 4. Divide by $\rho_c=3H_0^2c^2/8\pi G$ . The prefactor becomes $4\pi^2/3H_0^2$ , with $c^2$ and $G$ cancelling, leaving $\Omega_{\mathrm{gw}}(f)=(4\pi^2/3H_0^2)\,f^3 S_h(f)$ . Why: the cancellation is why the result depends only on $H_0$ — the gravitational coupling drops out, making $\Omega_{\mathrm{gw}}h^2$ a convention-free number. *The equivalent form $\Omega_{\mathrm{gw}}=(2\pi^2/3H_0^2)f^2 h_c^2$ with $h_c^2=2fS_h$ follows immediately, and a flat $S_h$ gives $\Omega_{\mathrm{gw}}\propto f^3$ .*

Go deeper · Derivation B · 1 minWhy cross-correlating two detectors beats their noise

Derivation BWhy cross-correlating two detectors beats their noise

The aim is to show why a background statistically identical to noise in one detector becomes detectable in two, and to extract the scaling law that governs every stochastic search: signal-to-noise grows as the square root of observation time, not linearly. We model each output as common signal plus independent noise, form a filtered cross-correlation, and choose the filter maximizing the ratio of its mean to its variance.

Step 1. Write each output as $s_I(t)=h_I(t)+n_I(t)$ , $I=1,2$ , with $h_I$ the common strain and $n_I$ the instrument noise. The defining assumption is $\langle n_1 n_2\rangle=0$ : the noises are physically independent while $h_1$ and $h_2$ respond to the same passing spacetime. Why: this asymmetry — signal common, noise independent — is the entire lever, and why two detectors do what one never can.

Step 2. Form $\hat C=\int\!\mathrm{d}t\,\mathrm{d}t'\,s_1(t)s_2(t')\,Q(t-t')$ . Taking the expectation, the noise cross-terms vanish and only the signal survives: $\langle\hat C\rangle=(3H_0^2/20\pi^2)\,T\!\int_0^\infty\!\mathrm{d}f\,\gamma(f)\Omega_{\mathrm{gw}}(f)f^{-3}\tilde Q(f)$ , growing linearly in $T$ . Why: the mean signal accumulates coherently for as long as you watch, while the noise accumulates only as its square root.

Step 3. The variance is dominated by noise-times-noise, $\sigma^2\propto T\!\int\!\mathrm{d}f\,P_1(f)P_2(f)|\tilde Q(f)|^2$ , with $P_I$ the noise spectra. Since both mean and variance scale as $T$ , the ratio $\langle\hat C\rangle/\sigma$ scales as $\sqrt T$ . Why: doubling the run improves signal-to-noise by only $\sqrt2$ — stochastic searches are marathons, with amplitude sensitivity improving as $(T\Delta f)^{1/4}$ .

Step 4. Maximize by varying the filter, giving $\tilde Q(f)\propto\gamma(f)\Omega_{\mathrm{gw}}(f)\big/\big[f^3 P_1 P_2\big]$ and optimal $\mathrm{SNR}=(3H_0^2/10\pi^2)\sqrt{2T}\big[\int_0^\infty\!\mathrm{d}f\,\gamma^2\Omega_{\mathrm{gw}}^2 f^{-6}P_1^{-1}P_2^{-1}\big]^{1/2}$ . Why: the filter weights each frequency by its signal and the detectors' quiet there, and the $\gamma^2$ shows a small overlap reduction function costs sensitivity quadratically. *The $\mathrm{SNR}\propto\sqrt T$ scaling is the defining property of stochastic detection.*

Go deeper · Derivation C · 1 minThe Hellings-Downs angular correlation

Derivation CThe Hellings-Downs angular correlation

The goal is to derive the specific quadrupolar shape of the expected correlation between the timing residuals of two pulsars illuminated by an isotropic, unpolarized background, and to see why that shape is a clean discriminator against clock and ephemeris errors. We write each pulsar's timing response as a projection of the strain onto its line of sight, correlate two such responses over the background, and average over the sky to a closed form in $x=(1-\cos\theta)/2$ .

Step 1. A gravitational wave shifts a pulsar's apparent pulse arrival by a redshift $z_I(t)=\sum_A\int\!\mathrm{d}^2\hat n\,F_I^A(\hat n)\,h_A(t,\hat n)$ , where $F_I^A$ projects the polarization tensor onto the Earth-pulsar direction $\hat p_I$ . Why: the pulsar is a freely falling clock, and a passing wave modulates the light-travel time along the line of sight — this projection is where the geometry enters.

Step 2. Correlate two pulsars. Using the isotropic, unpolarized specification (2), the random phases average out and $\langle z_1 z_2\rangle\propto\int\!\mathrm{d}^2\hat n\sum_A F_1^A F_2^A$ , a pure geometric overlap of the two antenna patterns. Why: only the common background survives the average, so the correlation is a fixed function of geometry, independent of the unknown spectrum amplitude.

Step 3. Perform the solid-angle integral over the spin-2 antenna patterns, yielding $\zeta(\theta)=\tfrac12+\tfrac32 x\ln x-\tfrac{x}{4}$ with $x=(1-\cos\theta)/2$ . Why: the logarithm is the signature of spin-2 structure — a monopole gives a constant, a dipole a $\cos\theta$ , and only the quadrupole this curve.

Step 4. Evaluate the landmarks: $\zeta(0)=\tfrac12$ , a zero near $49°$ , a minimum $\zeta\approx-0.15$ near $82°$ (a genuine anticorrelation), and $\zeta(180°)=+\tfrac14$ . Why: positive at small angles, negative in the middle, positive again at antipodes is the observable fingerprint, and reproducing it across many pulsar pairs separates a gravitational background from clock or ephemeris systematics. *Numerical evaluation gives the zero at $49.3°$ , minimum $-0.152$ at $82.5°$ , and $+0.25$ at $180°$ .*

Limiting cases

The same spectrum is probed across twenty decades, each band with its own source physics and detection method — the observational landscape of the first figure.

Case	Result	Meaning
Nanohertz (pulsar timing)	$f\sim10^{-9}\!-\!10^{-7}\,\mathrm{Hz}$ , $\Omega_{\mathrm{gw}}h^2\sim2\times10^{-9}$	supermassive black-hole binaries; measured via Hellings-Downs correlations. NANOGrav 2023 evidence
Millihertz (LISA)	$f\sim10^{-4}\!-\!10^{-1}\,\mathrm{Hz}$ , reach $\Omega_{\mathrm{gw}}\sim10^{-12}$	Galactic white-dwarf confusion foreground plus possible phase-transition or cosmic-string signals
Deca- to kilohertz (ground)	$f\sim10\!-\!10^3\,\mathrm{Hz}$ , $\Omega_{\mathrm{gw}}\lesssim10^{-9}$ near 25 Hz	integrated stellar-mass compact-binary mergers; LIGO-Virgo-KAGRA cross-correlation
Femtohertz (CMB B-modes)	$f\sim10^{-18}\!-\!10^{-16}\,\mathrm{Hz}$ , $\Omega_{\mathrm{gw}}\lesssim10^{-15}$	inflationary tensor modes as primordial B-mode polarization; bounded by tensor-to-scalar ratio $r\lesssim0.03$
Flat strain spectrum	$S_h=\text{const}\Rightarrow\Omega_{\mathrm{gw}}\propto f^3$	energy density weights high frequencies; strain-flat is not energy-flat

Common trapThe dimensionless spectrum is not the strain amplitude

It is tempting to treat $\Omega_{\mathrm{gw}}(f)$ and the strain as interchangeable measures of "how big" the background is. They are not. $\Omega_{\mathrm{gw}}$ is an energy density per logarithmic frequency relative to the critical density, and equation (5) shows it scales as $f^3 S_h(f)$ — so a flat strain spectral density is a steeply rising energy spectrum $\propto f^3$ , and the same field looks utterly different depending on which axis you plot. Two traps travel with this. First, $\Omega_{\mathrm{gw}}$ inherits the Hubble uncertainty through $\rho_c\propto H_0^2$ , so a bare $\Omega_{\mathrm{gw}}$ without a stated $h$ is ambiguous at the tens-of-percent level — hence $\Omega_{\mathrm{gw}}h^2$ . Second, a random field has no waveform to match-filter; the detection statistic is a cross-correlation whose significance grows only as $\sqrt T$ . Confusing one realization's strain amplitude with the ensemble spectrum is the commonest error in reading a sensitivity plot.

Worked examples

1 · From the pulsar-timing characteristic strain to Ω_gw (fully worked)

The NANOGrav 15-year dataset is consistent with a background of characteristic strain $h_c\approx2\times10^{-15}$ at the reference frequency $f=1\ \mathrm{yr}^{-1}$ . Convert this to a dimensionless energy density, using $H_0=67.9\ \mathrm{km\,s^{-1}\,Mpc^{-1}}$ (so $h=0.679$ ).

Step 1. Fix the constants in SI. The reference frequency is $f=1/(1\ \mathrm{yr})=1/(3.156\times10^7\ \mathrm{s})=\mathbf{3.17\times10^{-8}\ Hz}$ . The Hubble rate is $H_0=67.9\times10^3\ \mathrm{m\,s^{-1}}/(3.086\times10^{22}\ \mathrm{m})=\mathbf{2.20\times10^{-18}\ s^{-1}}$ , so $H_0^2=4.84\times10^{-36}\ \mathrm{s^{-2}}$ .

Step 2. Use the characteristic-strain form of equation (5), $\Omega_{\mathrm{gw}}=\dfrac{2\pi^2}{3H_0^2}f^2 h_c^2$ . Assemble the numerator: $f^2=(3.17\times10^{-8})^2=1.00\times10^{-15}\ \mathrm{Hz^2}$ , and $h_c^2=(2\times10^{-15})^2=4.0\times10^{-30}$ , so $f^2 h_c^2=4.02\times10^{-45}$ . The prefactor is $2\pi^2/3=6.58$ .

Step 3. Divide by $H_0^2$ : $\Omega_{\mathrm{gw}}=6.58\times(4.02\times10^{-45})/(4.84\times10^{-36})=\mathbf{5.5\times10^{-9}}$ . Multiplying by $h^2=0.461$ to remove the Hubble dependence gives $\Omega_{\mathrm{gw}}h^2=\mathbf{2.5\times10^{-9}}$ .

Step 4. Sanity-check: this sits at the nanohertz marker of the first figure and matches what the arrays report — near $\Omega_{\mathrm{gw}}h^2\sim$ few $\times10^{-9}$ , far above the $\sim10^{-15}$ inflationary level but plausibly sourced by supermassive black-hole binaries. A characteristic strain $h_c\approx2\times10^{-15}$ at $1\ \mathrm{yr}^{-1}$ corresponds to $\Omega_{\mathrm{gw}}h^2\approx2.5\times10^{-9}$ .

2 · The strain a ground-based network must resolve

LIGO-Virgo-KAGRA aim to detect or bound an astrophysical background at $\Omega_{\mathrm{gw}}(25\ \mathrm{Hz})=10^{-9}$ . What characteristic strain and strain amplitude spectral density does this correspond to, and why is it hard?

Step 1. Invert equation (5) for the characteristic strain: $h_c^2=\dfrac{3H_0^2\,\Omega_{\mathrm{gw}}}{2\pi^2 f^2}$ . With $H_0^2=4.84\times10^{-36}\ \mathrm{s^{-2}}$ , $\Omega_{\mathrm{gw}}=10^{-9}$ , and $f=25\ \mathrm{Hz}$ : numerator $3H_0^2\Omega_{\mathrm{gw}}=1.45\times10^{-44}$ , denominator $2\pi^2 f^2=19.74\times625=1.23\times10^{4}$ .

Step 2. Then $h_c^2=1.45\times10^{-44}/1.23\times10^{4}=1.18\times10^{-48}$ , giving a characteristic strain $h_c=\mathbf{1.1\times10^{-24}}$ . The corresponding amplitude spectral density is $\sqrt{S_h}=\sqrt{h_c^2/2f}=\sqrt{1.18\times10^{-48}/50}=\mathbf{1.5\times10^{-25}\ Hz^{-1/2}}$ .

Step 3. Compare with the instruments. Advanced detectors reach $\sqrt{S_h}\sim10^{-23}\ \mathrm{Hz^{-1/2}}$ near 25 Hz, so the target sits roughly two orders of magnitude below the single-detector noise — which is why the search is a cross-correlation over years of coincident data. A background of $\Omega_{\mathrm{gw}}=10^{-9}$ at 25 Hz means a characteristic strain of only $\sim10^{-24}$ , recoverable only by long cross-correlation.

Physics of the portrait

The defining object of this whole subject is not a person but a curve — the specific angular correlation an isotropic spin-2 background imprints on a pulsar timing array. It is the pattern that, once seen across dozens of pulsar pairs, turns "common red noise" into "gravitational-wave background," and it descends from a lineage of ideas: Isaacson's energy density that lets strain become $\Omega_{\mathrm{gw}}$ , the 1970s realization that pulsars are gravitational-wave antennas, and the cross-correlation detection theory that made a sub-noise signal a measurable statistic.

Origins · 1 minHellings-Downs curve and how this was discovered

Hellings-Downs curve

the quadrupolar inter-pulsar correlation ζ(θ) of an isotropic spin-2 background · pulsar-timing detection of the stochastic gravitational-wave background

The curve is the observational heart of nanohertz gravitational-wave astronomy, and its story is a chain of borrowed ideas made rigorous. The prerequisite was Isaacson's 1968 effective stress-energy, which first gave gravitational waves a well-defined energy density and let a random strain field be recast as $\Omega_{\mathrm{gw}}(f)$ . In 1978 Sazhin, and in 1979 Detweiler, noted that the ultra-regular ticking of pulsars would be modulated by a passing wave, turning each millisecond pulsar into a galaxy-scale clock antenna. In 1983 Ronald Hellings and George Downs asked the decisive question — if many pulsars are bathed in one isotropic background, how are their timing residuals correlated? Averaging the spin-2 antenna response over the sky, they found the correlation depends only on angular separation and traces a quadrupolar curve. For two decades it was a formula with no signal to fit, while the detection machinery matured separately: Allen and Romano's 1999 optimal cross-correlation statistic and overlap reduction function formalized how to dig a background out of independent noise. The two threads met in 2023, when NANOGrav and the European, Parkes, and Chinese arrays independently reported inter-pulsar correlations following the Hellings-Downs shape — the fingerprint that a common pulsar signal was genuinely a spin-2 background, most likely the summed hum of supermassive black-hole binaries across cosmic history.

History

1968

— Isaacson defines the effective stress-energy of high-frequency gravitational waves

1979

— Detweiler shows pulsar timing residuals respond to a gravitational-wave background

1983

— Hellings and Downs derive the quadrupolar inter-pulsar correlation curve

1999

— Allen and Romano formalize optimal cross-correlation and the overlap reduction function

2023

— NANOGrav and partner pulsar timing arrays report evidence for a nanohertz background

Going further

Texts. Maggiore, Gravitational Waves Vol. 1 (2007), Ch. 7 (stochastic backgrounds, $\Omega_{\mathrm{gw}}$ , overlap reduction function) · Misner, Thorne & Wheeler, Gravitation (1973), §35–37 (wave energy and stress-energy) · Wald, General Relativity (1984), §4.4 · Carroll, Spacetime and Geometry (2004), Ch. 7 · Allen & Romano, Phys. Rev. D 59, 102001 (1999) (detection statistic and $\gamma(f)$ ) · Hellings & Downs, ApJ 265, L39 (1983) · NANOGrav Collaboration, ApJ Lett. 951, L8 (2023). The relation $\Omega_{\mathrm{gw}}=(4\pi^2/3H_0^2)f^3 S_h$ is derived from the Isaacson stress-energy; the $\mathrm{SNR}\propto\sqrt T$ result and the closed-form $\zeta(\theta)$ are confirmed with the zero at $49.3°$ , minimum $-0.152$ at $82.5°$ , and $+0.25$ at $180°$ . Quoted numbers — $h_c\sim2\times10^{-15}$ giving $\Omega_{\mathrm{gw}}h^2\approx2.5\times10^{-9}$ , the LVK bound $\Omega_{\mathrm{gw}}\lesssim10^{-9}$ near 25 Hz, and $r\lesssim0.03$ — are standard published values.

Prerequisites. Isaacson Stress-Energy of Gravitational Waves · Transverse-Traceless Gauge · Plus and Cross Polarizations · Gravitational Wave Equation · Quadrupole Formula. Leads to. Multipole Expansion of Gravitational Radiation · Helicity and Spin-2 Nature of Gravitons.

Ask the tutor: Want to see how the Hellings-Downs curve emerges from averaging the spin-2 antenna pattern over the sky, or how the $\sqrt T$ growth sets the years-long integration a real search needs?

Is the stochastic background actually random, or does it just look random?

For all practical purposes it is genuinely random. A cosmological background superposes modes with independent quantum phases set in the early Universe; an astrophysical background is the incoherent sum of enormous numbers of unresolved sources from every direction. Either way the phases are uncorrelated, so a single realization looks like noise and no template can match it. What is not random is the statistical description — the power spectrum $\Omega_{\mathrm{gw}}(f)$ and the inter-detector correlation are fixed, reproducible quantities, exactly as you predict the microwave background's spectrum but not its particular hot and cold spots.

What is the difference between an astrophysical and a cosmological background?

An astrophysical background is the pile-up of unresolved late-Universe events — chiefly compact binaries — and its spectrum tracks the history of star formation and mergers; near a nanohertz it follows the binary-inspiral power law $h_c\propto f^{-2/3}$ . A cosmological background comes from the very early Universe — amplified inflationary modes, phase transitions, cosmic strings — and encodes physics at energies far above any collider. The observational challenge is separation: from the cosmological viewpoint the astrophysical background is a foreground to be modeled and subtracted to expose the primordial signal beneath.

Why can a single detector never see the background on its own?

Because it enters a single detector exactly as excess broadband noise, with no feature to distinguish it from the instrument's own noise, which no real detector knows to the needed precision. Cross-correlation sidesteps this: two detectors share the gravitational signal but have independent noise, so correlating their outputs isolates the common part. The mean of the correlation is pure signal and grows with time while the noise contribution averages toward zero — the price being patience, since significance climbs only as the square root of observation time.

What exactly does the overlap reduction function do?

The overlap reduction function $\gamma(f)$ measures how efficiently two detectors respond to the same background. Coincident and coaligned, they see identical strain and $\gamma=1$ ; but real detectors are separated and misaligned, so once the wavelength is shorter than the separation the responses decorrelate and $\gamma$ falls and oscillates. It sets the usable band of a pair and enters the optimal signal-to-noise squared, so a small $\gamma$ costs sensitivity twice over. For pulsar timing the analogous quantity over an isotropic background is precisely the Hellings-Downs curve.

Why is a cosmological background such a deep probe of the early Universe?

Because gravitons interact only gravitationally — almost not at all. Photons scattered constantly until recombination, so the microwave background shows us the Universe no earlier than about 380 000 years. Gravitational waves decouple essentially at the Planck time and then propagate freely, carrying an undistorted imprint of whatever produced them — inflation at $10^{-36}$ seconds, a phase transition, a cosmic-string network. A cosmological background is thus the earliest possible direct messenger, reaching us from energies inaccessible to laboratories, at the cost of being extraordinarily faint.

Does the 2023 pulsar-timing evidence already prove a gravitational-wave background?

It is strong evidence rather than a five-sigma discovery. Multiple arrays independently found a common red-noise signal whose inter-pulsar correlations follow the Hellings-Downs shape at roughly three to four sigma — "evidence for," not yet "definitive detection of." That shape is the discriminator: a common clock error would appear as a monopole, a solar-system ephemeris error as a dipole, and only a spin-2 background produces the quadrupolar curve. Confirming it at higher significance, then decoding whether the source is black-hole binaries or something cosmological, is the work of the next several years.