Pulsar Timing Arrays and Nanohertz Waves
A pulsar timing array turns the Galaxy itself into a gravitational-wave detector. Its clocks are millisecond pulsars — neutron stars spun up to hundreds of rotations per second, whose radio pulses arrive with a regularity that rivals atomic time. After fitting a deterministic timing model — spin and spin-down, astrometry, binary orbit, interstellar dispersion — a residual remains: the observed-minus-predicted arrival time of each pulse, measured to tens of nanoseconds over decades. A gravitational wave sweeping through the Galaxy stretches and squeezes the proper distance along each Earth–pulsar line of sight, modulating the photon travel time and depositing a slow, correlated wandering into these residuals. The physics is linearized gravity. In transverse-traceless gauge a plane wave imprints a fractional frequency shift on each pulsar set by an antenna pattern, the projection ½ n̂ⁱn̂ʲh_ij onto the line of sight. The response splits into an Earth term, sampled now and shared by every pulsar, and a pulsar term, sampled thousands of years ago at the source. The Earth terms correlate; the far-flung pulsar terms decohere into self-noise — and that correlation is the whole game. The flagship target is the nanohertz stochastic background from the cosmological population of inspiralling supermassive black hole binaries, whose superposed radiation produces a near-isotropic hum with characteristic strain h_c ∝ f^(−2/3). What proves the hum is gravitational and not a clock error is its angular fingerprint: Hellings and Downs showed in 1983 that an isotropic spin-2 background correlates pulsar pairs along a universal quadrupolar curve, and in 2023 four collaborations reported evidence for exactly that curve — the opening of the nanohertz window.
- A pulsar timing array (PTA) uses an ensemble of millisecond pulsars as a galaxy-scale detector, watching timing residuals — deviations of pulse arrival times from a deterministic model.
- A passing wave perturbs the photon travel time on each line of sight, producing a redshift z(t) that integrates into a residual R(t) = ∫z dt of tens to hundreds of nanoseconds for h ~ 10⁻¹⁵.
- PTAs work the nanohertz band, f ~ 1–100 nHz (periods of years to decades), set below by the span T and above by the cadence.
- The dominant target is the stochastic background from inspiralling supermassive black hole binaries, with a robust spectrum h_c(f) ∝ f^(−2/3).
- The unambiguous signature is the Hellings–Downs curve: the pair correlation depends only on angular separation γ — a quadrupolar fingerprint of a spin-2 field, with a minimum near 82.5°; in 2023 NANOGrav, EPTA, PPTA and CPTA reported ~3–4σ evidence for it.
- Key idea: a gravitational wave changes the photon travel time along each pulsar line of sight; the resulting timing-residual cross-correlation between pulsar pairs depends only on their angular separation — the Hellings–Downs curve — a model-independent, quadrupolar (spin-2) signature of the nanohertz gravitational-wave background.
A gravitational wave changes the photon travel time along each pulsar line of sight; the resulting timing-residual cross-correlation between pulsar pairs depends only on their angular separation — the Hellings–Downs curve — providing a model-independent, quadrupolar (spin-2) signature of the nanohertz gravitational-wave background.
Where this lives today
The nanohertz window is now an operating branch of GW astronomy, centred on the International Pulsar Timing Array (IPTA), which pools NANOGrav, the EPTA, the PPTA, the InPTA, and the CPTA into a dataset of roughly a hundred millisecond pulsars timed for up to two decades. The 2023 evidence for Hellings–Downs correlations, with A_yr ~ 2×10⁻¹⁵ and a spectrum broadly consistent with the SMBHB prediction, turned the field from a search into a measurement: the amplitude constrains the cosmic merger rate and binary masses, and probes the "final-parsec problem" of whether binaries can shed enough angular momentum to reach the GW-driven regime. The next milestones are resolving an individual nearby binary as a continuous wave, detecting anisotropy, and testing gravity's polarization content through departures from the pure quadrupolar curve that would betray scalar or vector modes. The same nanosecond timekeeping doubles as solar-system science, since an ephemeris error masquerades as a dipole, and new facilities like MeerKAT, FAST, and the Square Kilometre Array will move the field toward routine nanohertz spectroscopy.
The physical picture
Picture each pulsar as an exquisitely steady clock ticking out in the Galaxy, and Earth as a second clock reading the ticks as they arrive. A gravitational wave passing through the thousands of light-years between them stretches and compresses the proper distance along the line of sight, as it oscillates the separation of two freely floating masses. When the path lengthens the pulses arrive a hair late; when it shortens, a hair early. Each shift is a fractional frequency change of order the strain h ~ 10⁻¹⁵, but integrated over years it builds into a residual of tens to hundreds of nanoseconds — detectable because the arrival times are predicted so well. The reason a PTA can claim a detection is geometric, not merely sensitive. A wandering of Earth's reference clock would shift every residual identically (a monopole), and a solar-system ephemeris error would shift them with a dipole pattern; a genuine gravitational-wave background imprints a quadrupole, because gravity is a spin-2 field with two transverse polarizations. Recovering that quadrupolar angular pattern is a direct test that the common signal is radiation, not nuisance.
The nanohertz band is set by the experiment. The lowest resolvable frequency is roughly 1/T, the inverse of the multi-decade span (anything slower is absorbed into the spin-down fit); the highest is the Nyquist frequency 1/(2Δt) of the weeks-long cadence. This window lands exactly on the radiation from supermassive black hole binaries during the slow phase of their inspiral — a fortunate resonance between source and detector. The signal is deeply red: a blessing, since R ~ h/f grows toward low f, but also a curse, since a red process is just what spin noise, clock drift, and ephemeris error look like. Only the angular correlation breaks the degeneracy.
The core equations
The starting point is the redshift a single plane wave induces in a pulsar's apparent pulse frequency. Propagating a photon through a TT-gauge plane wave yields a projection of the metric perturbation onto the line of sight, differenced between two epochs — the atom of every PTA observable.
where: z — fractional shift of the apparent pulse frequency; ν₀ — intrinsic pulse frequency; n̂ — unit vector from Earth toward the pulsar; Ω̂ — wave propagation direction; h_ij — TT metric perturbation; t_e, t_p — Earth-term (now) and pulsar-term (emission, t_p = t_e − (L/c)(1+Ω̂·n̂)) epochs. Reading: the antenna factor ½ n̂ⁱn̂ʲ/(1+Ω̂·n̂) times the wave field differenced across the two ends. The observable is not z but its running integral: a fractional frequency error means pulses arrive progressively early or late, and the accumulated slippage is the residual. Integration in time is division by frequency in the Fourier domain, reddening the signal.
where: R — timing residual (observed minus modelled arrival time); the second relation is the frequency-domain scaling. Reading: integration adds f⁻¹ to the amplitude, so the residual is one power of frequency redder than the strain, favouring low frequencies. The spin-2 signature is carried by the correlation between two pulsars' Earth-term responses, averaged over an isotropic, unpolarized background. It depends only on the angle γ between the lines of sight, through x = sin²(γ/2).
where: Γ(γ) — expected normalized correlation for a pulsar pair; γ — angular separation; x = sin²(γ/2); δ_ab — Kronecker delta adding the self (pulsar-term) piece only for a pulsar with itself. Reading: for distinct pulsars (a≠b) drop δ_ab; the curve is ½ at γ→0, dips to −0.15 near 82.5°, returns to +0.25 at 180°. The x ln x term is the quadrupolar fingerprint no scalar or vector field mimics. The binary-population background has a spectrum fixed by kinematics rather than by the uncertain mass function. Circular, GW-driven inspiral forces a universal power law in characteristic strain.
where: h_c(f) — characteristic strain per logarithmic frequency interval; A_yr — amplitude at f_yr = 1 yr⁻¹ ≈ 32 nHz; α = −2/3 — spectral index for GW-driven circular binaries. Reading: only A_yr encodes the astrophysics; the slope is a theorem of energy balance, with recent PTAs finding A_yr ~ 2×10⁻¹⁵. Translating that strain into the residual power spectrum a PTA measures compounds two reddenings. The integration of equation (2) stacks on the strain slope of equation (4), giving a steep spectrum.
where: P_R(f) — one-sided PSD of the residuals; h_c — characteristic strain; f — GW frequency. Reading: converting h_c to residual PSD adds f⁻³, and with h_c² ∝ f⁻⁴ᐟ³ the power falls as f⁻¹³ᐟ³ — index γ = 13/3 ≈ 4.33, the number a PTA fits. The same background can be quoted as a cosmological energy density. This links the strain to the critical density through the Isaacson stress-energy of the wave field.
where: Ω_gw — GW energy density per logarithmic frequency in units of critical density; H₀ — Hubble constant; h_c — characteristic strain. Reading: since ρ_gw = ⟨ḣ_ij ḣ^ij⟩/32πG carries two time derivatives, Ω_gw gains f²; for α = −2/3 it rises as f²ᐟ³. Finally, the population superposes into a stationary background because each source barely chirps on human timescales. That slowness follows from the quadrupole-formula energy loss of a circular binary.
where: f — GW frequency (= 2× orbital frequency); 𝓜 = (m₁m₂)^(3/5)/(m₁+m₂)^(1/5) — chirp mass; G, c — Newton's constant and the speed of light. Reading: because df/dt ∝ f^(11/3) is tiny in the nanohertz band, each binary is effectively monochromatic over a 15-year span, so the ensemble adds into a stationary hum.
The goal is to obtain equation (1) from first principles: propagate a photon from pulsar to Earth through a single TT-gauge plane wave, track how the wave shifts its measured frequency, and show that the shift collapses to a difference of the metric perturbation at the emission and reception epochs, divided by a geometric projection factor. The two distinct pieces — the shared Earth term and the pulsar term sampled thousands of years in the past — then fall out automatically, along with the reason the array behaves as one detector.
Step 1. Work in TT gauge, ds² = −c²dt² + (δ_ij + h_ij^TT)dxⁱdxʲ, with a single plane wave depending on spacetime only through the phase t − Ω̂·x⃗/c. Both pulsar and Earth observers are static to first order, u^μ = (1,0,0,0). Why: in the gauge where free masses stay at fixed coordinates, the photon frequency is read off the field at the endpoints.
Step 2. The measured frequency is ν ∝ p_μu^μ, so the shift between emission and reception is fixed by how the null four-momentum evolves across the wave; impose g_μν p^μ p^ν = 0 with g_ij = δ_ij + h_ij^TT and solve for the first-order change in spatial momentum. Why: the wave changes not the photon's local energy but the phase relationship between the two static clocks, which is what a frequency measurement compares.
Step 3. The path along n̂ is unperturbed at zeroth order, so the shift is the line integral Δν/ν = −½ n̂ⁱn̂ʲ ∫∂_t h_ij dλ. Because h_ij depends only on the phase t − Ω̂·x⃗/c, the integrand is a total derivative and telescopes to h_ij(t_e) − h_ij(t_p). Why: a single-phase wave leaves no distributed imprint along a straight ray — only the boundary values survive, hence a two-term difference.
Step 4. Converting affine parameter to coordinate time along the line of sight contributes dλ ∝ dt/(1 + Ω̂·n̂), giving z = ½ n̂ⁱn̂ʲ/(1 + Ω̂·n̂) [h_ij(t_e) − h_ij(t_p)]. Why: the denominator is the wave-pulsar geometry, singular only for a wave travelling along the line of sight.
Step 5. Identify the terms: h_ij(t_e) is the Earth term, the same event for every pulsar, hence correlated; h_ij(t_p) with t_p = t_e − (L/c)(1 + Ω̂·n̂) is the pulsar term, at each pulsar's own location a light-travel time ago. Why: pulsars kiloparsecs apart have uncorrelated pulsar terms that average away, leaving the Earth term to carry the angular signal.
Go deeper · Derivation B · 2 minThe Hellings–Downs angular correlation curve
The goal is to derive equation (3), the universal correlation between two pulsars' residuals induced by an isotropic, unpolarized stochastic background. We form the expected product of the two Earth-term responses, average over all wave directions and both polarizations with equal weight, and carry out the solid-angle integral to reach the closed form in x = sin²(γ/2) — whose x ln x term and values at 0°, 82.5°, and 180° are the quadrupolar fingerprint that separates gravitational radiation from any monopolar or dipolar contaminant.
Step 1. Define C_ab = ⟨R_a(t)R_b(t)⟩, averaged over realizations and time. For a stationary isotropic background it can depend only on the angle γ between n̂_a and n̂_b. Why: isotropy forbids a preferred direction, collapsing a function of two directions to one of a single angle.
Step 2. Discard the pulsar terms. Every Earth term samples the field at the same event t_e, but each pulsar term samples a scattered location a light-travel time earlier with fL/c ~ 10³, so those phases average to zero in cross-correlation, leaving only the ½δ_ab auto piece. Why: this makes Earth the common node; without decoherence the correlation would depend intractably on distances.
Step 3. Write the survivor as a directional integral, Γ(γ) ∝ ∫ (d²Ω̂/4π) Σ_A (n̂_a n̂_a e^A)(n̂_b n̂_b e^A)/[(1+Ω̂·n̂_a)(1+Ω̂·n̂_b)], with e^A_ij the two TT polarization tensors, A = +, ×. Why: summing over exactly the two spin-2 polarizations — no scalar or vector modes — is where the field content enters; a different spin would change the answer.
Step 4. Align coordinates with the pair and integrate: Γ(γ) = (3/2) x ln x − x/4 + 1/2 with x = sin²(γ/2) for a≠b. Setting dΓ/dx = (3/2)(ln x + 1) − 1/4 = 0 gives x = e^(−5/6) = 0.435, so the minimum is at γ = 82.5°, Γ_min = −0.152. Why: the logarithm, from the (1+Ω̂·n̂) denominators integrated over the sphere, bends the curve into its dip and cannot come from a low-order multipole.
Step 5. Anchor values: x→0 gives Γ(0⁺) = ½; x = 1 gives Γ(180°) = +¼, half the zero-lag value; the curve crosses zero near γ = 50°. Why: the positive lobe, negative dip near 90°, and smaller rebound at 180° form a shape no scalar or vector background reproduces — recovering it is the detection claim. *Note: the anchor values Γ(0⁺)=0.5, minimum −0.152 at 82.5°, and Γ(180°)=0.25 correct an earlier erroneous draft that gave a minimum of −0.018 and Γ(180°)=½.*
Go deeper · Derivation C · 1 minWhy the binary background has characteristic strain f to the minus two-thirds
The goal is to show that h_c(f) ∝ f^(−2/3) for the superposed radiation of a cosmological population of circular, GW-driven supermassive black hole binaries, and that this slope is fixed by energy balance, not by the poorly known mass distribution. We count how long binaries linger near each frequency, multiply by the strain each contributes, and sum over the population; the mass function survives only in the amplitude, leaving a universal power law — the prediction PTAs test.
Step 1. From the quadrupole formula df/dt ∝ f^(11/3), so per logarithmic frequency dt/d ln f ∝ f^(−8/3); since the number radiating near a frequency tracks the dwell time, dN/d ln f ∝ f^(−8/3). Why: slow-chirping low-frequency binaries pile up, populating the low end of the band — the origin of the red spectrum.
Step 2. A single circular binary radiates strain-squared h² ∝ f^(4/3)/d² at distance d, straight from the quadrupole formula. Why: tighter, higher-frequency binaries are louder per source — a blue tendency that partly offsets Step 1.
Step 3. Combine: h_c²(f) ∝ (dN/d ln f) × h² ∝ f^(−8/3) × f^(4/3) = f^(−4/3), hence h_c(f) ∝ f^(−2/3). Why: the two tendencies net to exactly −2/3, independent of the mass function, which enters only A_yr.
Step 4. Propagate to residuals: with R ∝ h/(2πf), P_R ∝ h_c²/f³ ∝ f^(−13/3), index γ = 13/3. Why: residual reddening stacks on strain reddening, giving one of the steepest spectra in astrophysics, with power concentrated in the lowest bins.
Limiting cases
Key regimes and scalings that bracket PTA physics:
| Case | Result | Meaning |
|---|---|---|
| Sensitivity band | 1/T ≲ f ≲ 1/(2Δt) | T ~ 15–20 yr sets the low edge ~1 nHz; a cadence Δt of weeks sets the upper edge at hundreds of nHz. |
| SMBH-binary background | h_c(f) = A_yr (f/f_yr)^(−2/3) | Population strain; f_yr = 1 yr⁻¹ ≈ 32 nHz, with A_yr ~ 2×10⁻¹⁵ from recent PTAs. |
| Residual spectrum | P_R(f) ∝ f^(−13/3) | Timing-residual PSD of the background; spectral index γ = 13/3 ≈ 4.33 — a steeply red process. |
| Hellings–Downs anchors | Γ(0⁺)=½, min −0.15 @ 82.5°, Γ(180°)=¼ | Auto-correlation (with pulsar term) normalizes to 1; the dip near 82.5° is the spin-2 tell. |
| Pulsar-term decoherence | fL/c ≫ 1 | For L ~ 1 kpc and f ~ 10 nHz, fL/c ~ 10³: pulsar terms oscillate fast and decorrelate, leaving Earth-term correlations to carry the signal. |
| Monochromatic sources | df/dt → 0 in band | An SMBHB at 10 nHz chirps by ~10⁻¹² Hz over 15 yr, far below the 1/T bin width — each source is stationary, so the ensemble is a stationary background. |
Worked examples
1 · Residual amplitude of the background over a decade (fully worked)
Take A_yr = 2×10⁻¹⁵ at f_yr = 32 nHz with slope α = −2/3. Estimate the characteristic strain and the residual amplitude at f = 1/(10 yr), the low frequency a decade-long campaign first reaches, and compare with a timing precision of tens of nanoseconds.
Step 1. Frequency: f = 1/(10 yr) = 3.17 nHz, so f/f_yr = 0.10 and (f/f_yr)^(−2/3) = 10^(2/3) = 4.64 — the reddening boost from a decade below the reference.
Step 2. Strain: h_c = A_yr × 4.64 = 2×10⁻¹⁵ × 4.64 = 9.3×10⁻¹⁵.
Step 3. Residual, equation (2): R ~ h_c/(2πf) = 9.3×10⁻¹⁵/(1.99×10⁻⁸) = 4.7×10⁻⁷ s ≈ 470 ns.
Step 4. A residual of order half a microsecond stands well above tens-of-nanosecond precision — but only after a decade of baseline builds up the low frequency where the red signal is loud. The background gives a several-hundred-nanosecond wandering, detectable once T ~ 10 yr and the Hellings–Downs correlation is confirmed.
2 · Hellings–Downs values and the location of the dip (fully worked)
From the closed form Γ(γ) = (3/2) x ln x − x/4 + ½ with x = sin²(γ/2), find the zero-lag value, the angle and depth of the minimum, and the antipodal value.
Step 1. Zero lag: as γ→0, x ln x→0, leaving Γ(0⁺) = ½ — the maximum cross-correlation, set by the shared Earth term.
Step 2. Minimum: dΓ/dx = (3/2)(ln x + 1) − ¼ = 0 gives ln x = −5/6, so x = e^(−5/6) = 0.435; inverting x = sin²(γ/2) gives γ = 82.5°.
Step 3. Depth: Γ_min = (3/2)(0.435)(−0.833) − 0.109 + 0.5 = −0.152. The correlation goes negative near right angles — an anticorrelation no monopole or dipole can produce.
Step 4. Antipodal: at γ = 180°, x = 1 and x ln x = 0, so Γ(180°) = +¼, half the zero-lag value, with a zero-crossing near 50°. These four anchors — ½, zero near 50°, −0.15 at 82.5°, +¼ at 180° — are the template the 2023 detections matched.
3 · Why an individual binary is monochromatic (fully worked)
A binary of chirp mass 𝓜 = 10⁹ M⊙ radiates at f = 10 nHz. Using GM⊙/c³ = 4.925 μs, find its chirp rate, its dwell time near this frequency, and the frequency drift over a 15-year campaign.
Step 1. In time units G𝓜/c³ = 4925 s, so (G𝓜/c³)^(5/3) = 1.43×10⁶ s^(5/3). With π^(8/3) = 21.2 and f^(11/3) = 4.64×10⁻³⁰, equation (7) gives df/dt ≈ 2.7×10⁻²¹ Hz/s.
Step 2. Dwell time: t ~ f/(df/dt) = 10⁻⁸ / 2.7×10⁻²¹ ≈ 1.2×10⁵ yr — the binary sits near 10 nHz for ~100,000 yr.
Step 3. Drift over T = 15 yr = 4.7×10⁸ s: Δf = (df/dt)T ≈ 1.3×10⁻¹² Hz, against a resolution 1/T ≈ 2.1×10⁻⁹ Hz.
Step 4. The drift is ~600 times smaller than one bin — over any human campaign the source does not move in frequency. Each binary is effectively monochromatic, so the population superposes into a stationary background, as equations (3)–(6) assume.
Physics of the portrait
The portrait here is not a person but a prediction: the Hellings–Downs curve, whose shape is dictated by the spin of the field and which turns a red common signal into a detection.
Origins · 1 minHellings–Downs curve and how this was discovered
The idea of using pulsars as detectors is nearly as old as pulsars. Bell Burnell and Hewish found the first in 1967; within a decade Sazhin (1978) and, more fully, Detweiler (1979) derived how a passing wave perturbs pulse arrival times, with its Earth and pulsar terms. But a single pulsar's red wandering could be spin noise, a clock error, or an ephemeris error as easily as a wave — so the breakthrough was to ask not whether one pulsar wanders but how many pulsars correlate. In 1983 Ronald Hellings and George Downs computed that correlation for an isotropic background and found it depends only on the angular separation γ, tracing a universal quadrupolar curve. That shape follows from gravity being a transverse, traceless spin-2 field: a scalar background would give a monopole correlation, a vector a dipole one. The curve sat as a target for forty years while millisecond pulsars (Backer and colleagues, 1982) and dedicated arrays supplied the stable clocks and decades of nanosecond timing required. In June 2023 four collaborations announced ~3–4σ evidence for a common red process carrying the Hellings–Downs correlation — the first sighting of the nanohertz sky, most plausibly the hum of supermassive black hole binaries.
Going further
References. Misner, Thorne & Wheeler, Gravitation (1973), Ch. 35–37 (linearized waves, detection) · Wald, General Relativity (1984), Ch. 4 (linearized gravity, gauge) · Carroll, Spacetime and Geometry (2004), Ch. 7 (gravitational radiation) · Hartle, Gravity (2003), Ch. 16 · Schutz, A First Course in General Relativity, 2nd ed. (2009), Ch. 9 · Hellings & Downs, ApJ Letters 265, L39 (1983) · Detweiler, ApJ 234, 1100 (1979) · Maggiore, Gravitational Waves Vol. 1 (2008), Ch. 9 · Phinney, arXiv:astro-ph/0108028 (2001) (energy-balance background) · NANOGrav Collaboration, ApJL 951, L8 (2023); EPTA, A&A 678, A50 (2023). The Hellings–Downs anchor values are Γ(0⁺)=½, minimum Γ=−0.152 at γ=82.5°, Γ(180°)=+¼, and a zero-crossing near 50°; an earlier draft erroneously gave a minimum of −0.018 and Γ(180°)=½. Key numerical checks: the 96/5 chirp coefficient and df/dt ≈ 2.7×10⁻²¹ Hz/s for 𝓜 = 10⁹ M⊙ at 10 nHz; the −2/3 and −13/3 spectral exponents; Ω_gw = (2π²/3H₀²)f²h_c² via the Isaacson stress-energy tensor; and R ≈ 470 ns at f = 1/(10 yr) for A_yr = 2×10⁻¹⁵. The 2023 amplitude A_yr ~ 2×10⁻¹⁵ and ~3–4σ significance are observational results (NANOGrav 15-yr, EPTA DR2), not textbook values.
Prerequisites. Linearized Einstein Equations · Transverse-Traceless Gauge · Plus and Cross Polarizations · Quadrupole Formula · Geodesic Deviation and the Tidal Action of Waves. Leads to. Helicity and the Spin-2 Nature of Gravitons · Isaacson Stress-Energy of Gravitational Waves · Multipole Expansion of Gravitational Radiation.
Ask the tutor: Want to see how the Hellings–Downs closed form drops out of the solid-angle integral over the two spin-2 polarizations, or how the −2/3 slope would change if gravity carried scalar or vector modes?
Why is a pulsar timing array sensitive to nanohertz waves specifically, and not to the kilohertz waves LIGO hears?
The band is set by the experiment. The lowest resolvable frequency is about 1/T, the inverse of the observing span — with T of 15–20 years, roughly 1 nHz; the highest is the Nyquist frequency 1/(2Δt) of the weeks-long cadence, a few hundred nHz. LIGO, by contrast, is a 4-km interferometer peaking near 100 Hz. The two probe utterly different sources: PTAs see the decades-long inspiral of billion-solar-mass binaries, LIGO the final seconds of stellar-mass mergers, with LISA filling the millihertz gap between.
What exactly is a timing residual, and how is it different from the redshift z?
The residual R(t) is what remains after subtracting a deterministic timing model — spin, astrometry, binary orbit, interstellar dispersion — from the measured arrival times; it is an observed-minus-predicted time in nanoseconds. The redshift z(t) is the instantaneous fractional shift in the pulsar's apparent frequency, of order the strain h. The two connect by integration, R(t) = ∫z(t′)dt′, and that matters: in the frequency domain it divides by 2πf, reddening the residual by one power of frequency. Confusing z with R is the commonest way to get the spectral index wrong.
Why does the supermassive-binary background follow f^(−2/3) if the binary masses are so uncertain?
Because the slope is a theorem, not a fit. Two effects compete: slow-chirping binaries pile up at low frequency (dwell time ∝ f^(−8/3)), while each source is louder at higher frequency (strain-squared ∝ f^(4/3)). Multiplying gives h_c² ∝ f^(−4/3), so h_c ∝ f^(−2/3). The mass function and merger rate only rescale A_yr; they cannot change the slope — which is why the index is a clean prediction and the amplitude carries the astrophysics.
What are the Earth term and the pulsar term, and why does only the Earth term matter for the correlation?
The response is a difference of the wave field at two epochs. The Earth term h(t_e) is the field when the pulse is received — the same spacetime event for every pulsar, so it correlates. The pulsar term h(t_p) is the field at each pulsar's location one light-travel time earlier, thousands of years ago; because the pulsars are scattered across kiloparsecs with fL/c ~ 10³, these terms sample random phases and average to zero in cross-correlation, contributing only self-noise. The Hellings–Downs signal therefore lives entirely in the shared Earth term — which is why Earth is the common node of a galaxy-scale detector.
If the correlation is the whole signal, why not just use two pulsars?
Because the Hellings–Downs curve is a function of angular separation, and two pulsars give only one point on it. A common red process of the right amplitude could arise from many non-gravitational effects; what distinguishes a spin-2 background is the full quadrupolar shape. Sampling it with confidence needs many pairs spanning all separations, which is why arrays aim for tens of pulsars: the pair count grows as N², and it is the pairs, binned by separation, that trace the curve and beat down the variance.
Is the 2023 result a confirmed detection of supermassive black hole binaries?
It is evidence, not yet a gold-standard detection, and its interpretation is open. The four collaborations reported Hellings–Downs correlations at roughly 3–4σ — strong, but below the 5σ discovery threshold, with significance climbing as datasets lengthen. Even granting the correlation is gravitational, the supermassive-binary population is the most natural but not the only source: cosmological backgrounds from inflation, cosmic strings, or phase transitions remain possible contributors that current data cannot cleanly separate. Distinguishing them will come from the spectral index, any anisotropy, and eventually the resolution of individual binaries.