AI-generated illustration for Galileo Whispers. A generic radio-observatory scene represents pulsar timing; no actual observatory, pulsar or measured gravitational-wave pattern is identified.
Pulsar timing can identify a gravitational-wave background because the waves disturb many pulse-arrival records together, with a correlation that depends on the pulsars’ separation on the sky. For the usual isotropic, unpolarized background of general relativity, that expectation is the Hellings–Downs curve. Similar slow timing fluctuations in several stars are only part of the evidence; their spatial relationships supply a different test.
The distinction matters when reading the NANOGrav 15-year analysis published in 2023. Its evidence concerns correlated timing residuals across an array. It does not identify a particular pair of black holes or prove that every contribution to the background has the same origin.
From a pulse train to a timing residual
A rapidly rotating pulsar produces a repeating radio signal. Astronomers fit a timing model that includes its rotation, position and, where relevant, binary orbit. They also correct propagation effects and the observatory’s motion. A residual is the difference between an observed arrival time and that model’s prediction.
A gravitational wave changes the propagation time along the pulsar–Earth path. In the timing response, one contribution depends on the wave near the receiving end, conventionally called the Earth term; another depends on the wave at the pulsar when the pulse was emitted. The pulsars are far apart, so their individual terms generally do not add a common pairwise signal in the standard approximation.
The receiving end is shared, but its response is directional. Two pulsars point along different paths through the same gravitational disturbance. Their timing shifts therefore need not have identical signs or amplitudes. Averaging over the background’s wave directions and polarizations produces the angular curve, as developed in Romano and Allen’s explanation of the response and its limits.
What the angular pattern looks like
For two distinct pulsars, a common normalization writes the expected correlation as
Γ(θ) = 1/2 + (3/2)x ln(x) − x/4, where x = [1 − cos(θ)]/2.
Here θ is the separation angle as seen from Earth. At zero separation, use the limit x ln(x) → 0. These calculated values describe the ideal curve, not measured pulsar pairs:
| Separation | Expected correlation Γ | Interpretation |
|---|---|---|
| Approaching 0° | 0.50 | Nearby directions respond positively together. |
| 90° | About −0.145 | The average product can be negative. |
| 180° | 0.25 | Opposite directions become positively correlated again. |
The scale convention is arbitrary if the background amplitude is rescaled consistently. A pulsar correlated with itself is a separate case; it should not be confused with two distinct pulsars that happen to lie close together on the sky.
Why shared noise is not enough
“Red” timing noise has more power on long timescales. Several pulsars can favor a common red-noise spectrum without displaying the cross-correlations predicted for gravitational waves. A model can share the spectrum’s parameters while leaving each pulsar’s particular fluctuations independent.
A clock error offers another comparison: shifting the receiving clock affects all directions similarly, producing a monopolar pattern. Errors in the solar-system reference calculation can create a different directional pattern. Pulsar timing therefore compares explanations for both the time dependence and the correlations, while accounting for individual pulsar noise. A visually suggestive curve alone is not a complete statistical analysis.
An expectation with scatter
The Hellings–Downs curve is an average prediction under stated background assumptions. A finite set of sources, pulsars and observing years produces scatter around it, even before instrumental noise is included. An anisotropic background or an individually important source also requires more detailed treatment.
The useful reading question is therefore twofold: does an analysis find a shared low-frequency process, and does it separately favor the predicted angular correlations over alternatives? Pulsar timing gains its gravitational-wave evidence from that combination, rather than from a single unusually regular star becoming a little less regular.
