How Science Works | How Pulsars Can Help Detect Gravitational Waves

A pulsar is a rotating neutron star whose radiation beam crosses our line of sight. Millisecond pulsars rotate particularly rapidly, and selected examples provide stable timing references over long periods.

The rotation is not perfectly immutable. Pulsars can spin down, move across the sky, orbit companions and exhibit intrinsic timing noise. Intervening plasma and the observing instrument can also alter the measured signal. Precision comes from modelling those effects, not from assuming they do not exist.

Individual pulses vary in shape and strength. Observers commonly combine many rotations to obtain a stable average pulse profile, then compare it with a reference template to estimate an arrival time and its uncertainty.

For an original scale example, take a pulsar with a 5-millisecond period. A 100-nanosecond timing difference is only 0.000020 of a rotation, equivalent to 0.0072° of rotational phase. Measuring such a small phase displacement requires good signal quality, calibration and a suitable profile model.

The telescope records electromagnetic radio radiation. It does not directly receive a gravitational wave as an audible sound. The wave is inferred from a tiny influence on the timing pattern.

This method also differs from ordinary antenna interferometry. A pulsar timing array compares long-term timing records from distinct celestial clocks; it is not simply a set of dishes combined to form a sharper radio image. The existing Radio Telescopes guide explains that separate imaging task.

Contents · Next chapter

A measured pulse arrival time must be placed in a consistent time standard and corrected for the observatory’s motion. Astronomers use a Solar System model to refer observations to an appropriate barycentric description, centred on the Solar System’s centre of mass.

The pulsar model includes quantities such as rotational frequency, its change with time, sky position, proper motion and, where necessary, binary-orbit parameters. Propagation through ionized material and instrument-dependent offsets also enter the analysis.

A timing residual is the difference between a measured arrival time and the arrival time predicted by the fitted model. Its sign convention should be stated; here a positive residual means later than predicted.

Residuals are therefore not raw irregular flashes. They are already the result of an inference process. Changing the timing model or its fitted parameters can change the residual record.

For example, an error in rotational frequency can create a residual trend, while an error in spin-down can create curvature. Fitting those terms removes corresponding patterns, including parts of any very slowly varying signal that resembles them over the observed interval.

A careful gravitational-wave analysis accounts for this filtering and uncertainty rather than treating the fitted residuals as untouched data. Simulated signals passed through the same timing fit help establish which signals could be recovered.

The question becomes: after the known deterministic effects and statistical noise processes are handled, is there a shared pattern across pulsars that matches the expected spacetime signal better than the alternatives?

Contents · Previous chapter · Next chapter

A nanohertz is one billionth of a cycle per second. A gravitational wave at 10 nanohertz has a period of 100 million seconds, approximately 3.17 years using a 365.25-day year.

That slow variation is very different from the much higher-frequency signals targeted by ground-based laser interferometers. The Gravitational Waves and Laser Interferometers guide explains that detector family. Neither method replaces the other; their timescales open different observational windows.

For an original 15-year observing span, 1/T is about 2.11 nanohertz. This gives a useful frequency scale for changes extending across the whole record. It is not a perfectly sharp low-frequency detection boundary: sampling, timing-model fitting and the signal shape all affect sensitivity.

Cadence matters at shorter times, while the total baseline matters for the slowest variations. Irregular observations, gaps and changes in telescope sensitivity make the real response more complicated than equally spaced textbook sampling.

Suppose a synthetic residual is R(t) = R₀ sin(2πft), with R₀ = 100 nanoseconds and f = 10 nanohertz. Its time derivative has amplitude 2πfR₀, approximately 6.28 × 10⁻¹⁵. This is the scale of the corresponding fractional timing-rate variation, with sign depending on convention.

That derivative is not automatically the gravitational-wave strain. The conversion includes sky geometry, polarization and the Earth and pulsar contributions. Likewise, cR₀ ≈ 30 metres is an equivalent light-travel length, not proof that a particular telescope arm or pulsar moved by 30 metres.

Contents · Previous chapter · Next chapter

A gravitational wave changes the relationship between emission and reception events in spacetime. In pulsar timing, the response contains a contribution associated with the wave near Earth and another associated with the pulsar at an earlier emission epoch.

The Earth contribution can connect different pulsars’ residuals because their signals are received in the same region of spacetime. The pulsar contributions generally sample different locations and epochs, and are usually treated as uncorrelated between distinct pulsars in the standard background approximation.

That statement has assumptions. Pulsar distances are large compared with the relevant gravitational wavelength, and special close-pair or finite-distance situations can require more careful treatment. The approximation should not be turned into a claim that the pulsar terms never contain useful information.

With N pulsars there are N(N − 1)/2 distinct pairs. For 67 pulsars, that is 2211 pairs. The arithmetic sounds impressive, but it does not mean 2211 independent equally precise experiments.

Pairs share pulsars, some records are longer or quieter than others, and the sky distribution is uneven. A few very well-timed pulsars may contribute disproportionately. Correlations among the estimates must be included in the uncertainty analysis.

A useful diagram shows Earth at the centre with several pulsars in different sky directions. Mark the angle between two sightlines, not merely the distance between the pulsars on the page. The background test concerns how timing correlations vary with that angular separation.

Contents · Previous chapter · Next chapter

For an isotropic, unpolarized background of the usual transverse tensor gravitational waves in general relativity, the standard pulsar-timing approximation predicts the Hellings–Downs correlation curve. It was derived decades before the modern observational evidence. Read the original Hellings and Downs paper.

For two distinct pulsars, one common normalization is Γ(θ) = ½ + (3/2)x ln(x) − x/4, with x = [1 − cos(θ)]/2. Here θ is their sky separation and ln is the natural logarithm.

The curve is positive at small separations, negative over part of the middle range and positive again toward opposite sky directions. This does not mean the waves repel pulsars at one angle and attract them at another. It describes the statistical relation between their timing responses.

In our chosen normalization, Γ approaches 0.5 as the separation of distinct pulsars approaches zero. The same-pulsar autocorrelation has an additional contribution and is normalized differently, reaching 1 in the corresponding ideal expression. Neither value is automatically the measured Pearson correlation of two noisy residual records.

Original evaluations give Γ ≈ 0.212 at 30°, −0.082 at 60°, −0.145 at 90° and 0.250 at 180°. These values are predictions for the specified model, not four published measurements.

The curve is an ensemble expectation. A finite realization of a stochastic background, measurement noise and an uneven array produce scatter. Real points need not lie perfectly on it, and a few visually convenient points cannot establish a detection by themselves.

Contents · Previous chapter · Next chapter

An error in the terrestrial time standard can introduce a shared timing error across pulsars. In an idealized description, that produces a direction-independent, or monopolar, pattern rather than the Hellings–Downs dependence.

An error in the estimated Solar System barycentre produces a timing effect that depends on the projection of the position error onto each pulsar’s sightline. This creates a dipolar-type angular structure. Its detailed temporal behaviour depends on the ephemeris error.

Neither alternative can be rejected simply because several pulsars show a common slow process. The angular information and the full noise model are essential. In a finite, nonuniform array, different spatial patterns can partly overlap in the estimated statistics.

Interstellar plasma supplies another diagnostic because its leading dispersive delay depends on radio observing frequency approximately as ν⁻². A gravitational-wave timing perturbation is achromatic to leading order: it does not follow that radio-frequency scaling.

For an original two-frequency example, write r(ν) = A + B/ν², with ν measured in gigahertz. Suppose residuals are 4.1 microseconds at 1 GHz and 1.1 microseconds at 2 GHz. Subtracting them gives 3.0 = B(1 − 1/4), so B = 4.0 microseconds·GHz² and A = 0.10 microseconds.

The exercise separates an assigned dispersive term from an achromatic remainder. It does not prove the remainder is gravitational: clock, instrument and pulsar effects can also be achromatic. Real data additionally require uncertainty, profile evolution and more complicated propagation effects to be considered.

Contents · Previous chapter · Next chapter

A “red” process has comparatively strong power at low frequencies, corresponding to slowly varying structure in time. Several pulsars can show processes with similar spectra without those processes having the cross-correlations expected from a gravitational-wave background.

The distinction is between common spectral properties and shared fluctuations with the predicted spatial relationship. Similar-looking power spectra do not by themselves establish the second.

The International Pulsar Timing Array’s second-data-release search, reported in 2022, found evidence for a common low-frequency process but did not find significant Hellings–Downs correlations in that analysis. Its result illustrates why the researchers separated those claims rather than treating them as equivalent. Read the original IPTA search.

In 2023, NANOGrav reported multiple lines of evidence for a stochastic signal correlated among 67 pulsars in its background analysis, with correlations following the Hellings–Downs pattern. The team compared different noise and correlation models using Bayesian and frequentist approaches. This is stronger evidence than a common spectrum alone. Read the original NANOGrav background paper.

That specific sample count belongs to the cited analysis. Counts in general data-release summaries need not be identical because selection and data requirements differ. The label “15-year data set” is also a release name, not a promise that every pulsar has exactly the same observing span.

An evidence claim should therefore identify its data release, sample, statistical model and tested alternatives. A headline detached from those choices can accidentally combine numbers and conclusions from different analyses.

Contents · Previous chapter · Next chapter

Take four invented normalized cross-correlation summaries at separations 30°, 60°, 90° and 180°: 0.22, −0.07, −0.15 and 0.24. They were constructed for this worksheet and are not telescope observations.

Compare them with the fixed-normalization Hellings–Downs values from Chapter 5. The differences are small, and the positive–negative–positive angular structure is similar.

For a simple arithmetic diagnostic, add the squared differences between the four invented values and the predicted curve. The sum is approximately 0.000349. No observational error model has been assigned, so this is a descriptive mismatch score rather than a chi-squared significance.

Now compare a direction-independent constant. With equal weights, the best-fitting constant is the mean, 0.06. Its squared-difference sum is 0.119, much larger for these deliberately constructed numbers.

The exercise demonstrates what angular structure contributes beyond a claim that something is common. It does not recreate the NANOGrav analysis, establish a discovery threshold or give a probability that the gravitational explanation is true.

Real comparisons allow uncertain amplitudes and spectra, account for correlated errors and fit pulsar-specific noise. The plotted summaries may themselves depend on model choices. A dipolar alternative, mixtures of processes and possible systematic effects also need explicit testing rather than dismissal from one sketch.

Most importantly, a favourable curve is not independent evidence if the synthetic values were chosen to resemble it. Here the purpose is to learn the prediction’s shape before reading genuinely uncertain data.

Contents · Previous chapter · Next chapter

A background can be produced by the combined contribution of many unresolved sources. A cosmic population of supermassive black-hole binaries is an important explanation for nanohertz gravitational waves, but a correlation pattern alone does not label each source or determine a unique population history.

The frequency spectrum, anisotropy, departures from simple models and eventual individually resolvable signals can provide additional information. Galaxy populations and electromagnetic observations supply complementary constraints.

A 2026 NANOGrav study explored a more flexible, piecewise power-law reconstruction of the 15-year signal’s spectrum. Rather than forcing the whole spectrum into one slope, it combined models with different numbers of spectral segments. Such work addresses how to characterize the signal for source-model comparisons; it is not an independent identification of one particular binary. Read the original spectral-reconstruction study.

Different physical models can produce partly similar spectra over a limited frequency range. Uncertainty in the pulsar noise and the finite duration of the observations also affects how much structure can be inferred. Extra flexibility must earn its support from the data rather than merely improve a plot.

This is why “we found evidence for a background” should not be expanded into “we watched one black-hole pair merge” or “we proved a specific early-universe theory.” Those statements ask different questions.

The astronomy is compelling precisely because new measurements can sharpen the distinction. More pulsars, longer baselines, improved noise characterization and independent arrays can test whether the inferred angular and spectral properties remain consistent as the evidence grows.

Contents · Previous chapter · Next chapter

Begin with the 5-millisecond pulsar. Verify that 100 nanoseconds is 0.000020 of one rotation. Then explain why an arrival-time estimate normally comes from a measured profile and a timing model rather than a perfectly identical isolated pulse.

Convert 10 nanohertz into a period of about 3.17 years. Compare that with the 15-year observing-span scale of 2.11 nanohertz. State why neither cadence nor total duration alone describes the full sensitivity of a real timing array.

Next solve the two-radio-frequency example. Recover the 0.10-microsecond achromatic component and name one non-gravitational explanation that could still produce it. A clock error or an instrument offset is sufficient to show why chromatic correction is necessary but not decisive.

Draw the four Hellings–Downs predictions and the four synthetic summaries. Add the best constant of 0.06. Describe the angular shape before calculating any score, and explicitly label the numbers as invented.

Finally, build a claim ladder. First establish a well-characterized timing residual. Then test for common spectral power. Then test the predicted angular correlations against alternatives. Only after that ask which source populations explain the spectrum and other observed properties.

The central skill is using many imperfect clocks to test a precise shared prediction. Pulsar timing arrays turn patient radio observations into a probe of slowly changing spacetime while preserving the distinction between measured arrival times, inferred correlations and astrophysical explanations.

Continue to Astrophysics or the How Science Works master guide for related examples of cosmic measurement and inference.

Contents · Previous chapter · Continue to Astrophysics