Pulsar Timing Arrays

Scrollable formula. Use Left and Right arrow keys to reveal hidden content.

Scrollable table. Use Left and Right arrow keys to reveal additional columns.

Scrollable figure. Use Left and Right arrow keys to inspect hidden content.

A pulsar timing array turns the Milky Way into a nanohertz gravitational-wave detector. Each millisecond pulsar is a distant clock; the Earth and radio telescopes form the common receiving end. A gravitational wave perturbs pulse arrival times in every line of sight, but the scientific observable is not a single residual curve. It is the frequency-dependent, sky-dependent covariance of many residual data sets after timing models, pulsar noise, interstellar propagation, clock errors, and Solar-System ephemeris errors have all been accounted for.

Helpful background. The Gravitational-Wave Likelihood introduces Gaussian-process inference, and Stochastic Cross-Correlation develops cross-correlation logic in another detector regime.

From pulse arrival times to an array response

For pulsar $a$ , a plane gravitational wave changes the light-travel time through an Earth term and a pulsar term. With positive residual meaning a late arrival, its derivative is the redshift $z_a=-\delta\nu_a/\nu_a$ . Integrating gives

r_a(t)=\int^t z_a(t')\,dt'.

The Earth term samples the metric perturbation at one spacetime location for the whole array. It therefore produces correlated residuals. The pulsar terms sample the wave thousands of light-years away at different retarded times; for a stochastic background they usually contribute mainly to each pulsar’s auto-covariance, while for a resolvable source their phases can carry distance and source-evolution information.

The response is meaningful only after the deterministic timing model is fitted. Spin parameters remove low-order trends, astrometry absorbs annual structure, binary parameters remove orbital signatures, and instrumental offsets absorb constants. The remaining data vector is consequently a filtered view of both signals and noise. Pulsar Timing Residuals builds this interface from pulse profiles and times of arrival.

The angular signature

For an isotropic, stationary, Gaussian background with the tensor polarizations of general relativity, the cross-power between distinct pulsars has the form

C_{ab}(f)=\Gamma_{\rm HD}(\zeta_{ab})P_{\rm gw}(f), \qquad \Gamma_{\rm HD}(\zeta) =\frac12+\frac32x\ln x-\frac14x, \qquad x=\frac{1-\cos\zeta}{2}.

The Hellings–Downs function depends only on angular separation $\zeta$ . It is positive for nearby pairs, negative over an intermediate range, and positive again toward antipodal pairs. Its quadrupole and higher multipoles distinguish a tensor background from a clock error, which is monopolar, and an isotropic ephemeris-displacement model, which is dipolar. A single common red spectrum is therefore not the complete signature.

Earth Terms, Pulsar Terms, and Antenna Response derives the single-source geometry. Hellings–Downs Correlations performs the sky average and states the normalization conventions, including the extra auto-correlation contribution.

Inference with real pulsars

A practical covariance model combines several components,

C_{ab}=\delta_{ab}\!\left(C_a^{\rm white}+C_a^{\rm red}+C_a^{\rm chromatic}\right) +C_{ab}^{\rm clock}+C_{ab}^{\rm ephemeris} +\Gamma_{ab}^{\rm gw}C^{\rm gw}.

The components are not separated by color-coded curves in the data. They overlap in temporal spectrum and can covary with the fitted timing model. Multifrequency observations help distinguish chromatic dispersion and scattering from achromatic processes. Sky geometry helps distinguish monopolar, dipolar, and tensor correlations. Long baselines and stable instrumentation help resolve the lowest Fourier modes.

The central statistical progression is deliberately cautious:

establish red processes in individual pulsars;
test for a spectrum shared across the array;
test whether cross-pulsar correlations are present;
compare tensor, monopole, dipole, and uncorrelated spatial models;
only then infer spectrum, anisotropy, non-Gaussianity, or source population.

Pulsar Noise and Solar-System Ephemerides develops the principal alternatives. Common-Spectrum and Correlated-Process Inference connects likelihoods, Bayesian evidence, optimal statistics, and null tests.

What the observations support

At the evidence cutoff 2026-08-18, the major PTA collaborations have reported common low-frequency processes, and several data sets show evidence for the angular correlations expected from nanohertz gravitational waves. The NANOGrav 15-year analysis gives a particularly explicit comparison of Hellings–Downs, monopole, dipole, and uncorrelated models Agazie et al. 2023, §§3–5. Contemporaneous releases provide independent support with important differences: EPTA’s modern-backend subset has stronger tensor-correlation evidence than its longer heterogeneous span Antoniadis et al. 2023, §§4.2–5 and Table 5; PPTA’s full-likelihood model odds remain insignificant while its cross-correlation-only statistic is approximately $2\sigma$ Reardon et al. 2023, §§3.6–3.7 and 5; and CPTA’s $4.6\sigma$ result compares Hellings–Downs with constant correlation at a single frequency and does not exclude a dipole Xu et al. 2023, §§4–5 and Appendices A and C. Their pulsar sets, noise models, priors, and summary statistics are not interchangeable.

This evidence supports a nanohertz gravitational-wave background as the leading interpretation. It does not uniquely establish the source mixture. A population of supermassive-black-hole binaries is a natural contributor, but environmental coupling, eccentricity, a finite number of bright binaries, cosmic strings, or primordial components can modify the spectrum and sky statistics. Nor has a compelling individual continuous source or burst-with-memory event emerged from the NANOGrav 15-year data. Individual Nanohertz Sources and Memory explains what those searches constrain and why a background can be established first.

Two paths through the chapter

For a first pass, read the pages in order: residual construction → Earth/pulsar response → Hellings–Downs correlations → noise alternatives → array-level inference → individual-source searches. This route separates the observable, the geometric signature, and the statistical claim.

For data analysis, begin with Pulsar Timing Residuals and Pulsar Noise and Solar-System Ephemerides, then move to Common-Spectrum and Correlated-Process Inference. Return to the response pages when translating a covariance implementation or testing nonstandard polarizations.

Common misconceptions

A common spectrum is already a gravitational-wave detection. A shared red spectrum establishes common power, not its spatial origin. The cross-correlations and their competition with clock and ephemeris models carry the specifically gravitational evidence.

The pulsar term is merely noise. It is incoherent across an idealized stochastic-background average, but for a deterministic source it can encode the wave at an earlier epoch and can sharpen parameter inference when pulsar distances are constrained.

The measured spectrum identifies supermassive-black-hole binaries uniquely. A limited frequency lever arm and uncertain binary environments permit several spectral shapes. Population attribution requires additional spectral, angular, demographic, and external evidence.

Review the chapter

1. Spatial alternatives. Suppose every pulsar contains the same red spectrum but cross-covariances vanish. Which level of the evidence progression has been reached, and what is still missing?

Solution

The array has evidence for a common-spectrum process, corresponding to step 2. It has not demonstrated cross-pulsar correlation, much less the Hellings–Downs angular pattern. Clock, ephemeris, gravitational-wave, and other common causes therefore remain unseparated by spatial structure.

2. Earth and pulsar terms. Why does the Earth term generate the Hellings–Downs cross-correlation while the pulsar terms usually do not for an isotropic stochastic background?