Skip to main content
Sandbox Physics
Pulsar timing arrays and the Hellings–Downs curve

How can a pulsar timing array tell gravitational waves from a bad clock?

Sandbox Physics ·

By looking at how the timing errors of many pulsars move together, and how that depends on where the pulsars sit in the sky. If the clock on Earth runs slightly wrong, every pulsar seems early or late by the same amount. If our map of the Solar System is slightly wrong, pulsars on one side of the sky seem early while pulsars on the opposite side seem late. A background of gravitational waves does something else: two pulsars close together in the sky drift together, two pulsars about a right angle apart drift slightly against each other, and two pulsars on opposite sides drift together again, weakly. That curve is named after Hellings and Downs, who worked it out in 1983. Since 2023, several pulsar timing arrays have reported evidence for it.

Clocks spread across the Galaxy

A millisecond pulsar is a dead star about the size of a city that spins hundreds of times a second. Each turn sweeps a radio beam past the Earth, so radio telescopes receive a pulse every few milliseconds. Over years these pulses arrive with a regularity that rivals the best atomic clocks. Jenet and Romano, 2014, Section I

Astronomers predict when each pulse should arrive. The prediction accounts for how fast the pulsar spins and how fast it slows down, where it is, how it moves across the sky, whether it orbits a companion star, and how the Earth moves around the Sun. What remains after subtracting the prediction is the timing residual: how early or late each pulse really was, usually less than a microsecond.

The NANOGrav collaboration timed 68 pulsars between July 2004 and August 2020 with the Arecibo Observatory, the Green Bank Telescope and the Very Large Array. Its search for gravitational waves used the 67 pulsars timed for at least three years, spread over 16.03 years. NANOGrav Collaboration, 2023, Section 2

What a gravitational wave does to the ticks

A gravitational wave slightly stretches space in one direction and squeezes it in the perpendicular direction, then reverses, over and over. A radio pulse crossing this changing space arrives a little early or a little late. The waves pulsar timing looks for take years to complete one cycle, so their frequencies are a few billionths of a hertz, called nanohertz.

The likely source is a large population of pairs of supermassive black holes slowly spiralling toward each other in the centres of distant galaxies. Their waves overlap into a random hum from every direction, a gravitational-wave background. No single pulsar can prove the hum is there. Pulsars have their own slow wobbles, and a long, slow drift in one pulsar looks the same whether space or the star caused it.

Three ways the ticks can move together

The way out is to compare pairs of pulsars. Something that affects all of them leaves a fingerprint in how the drifts of two pulsars match, and the fingerprint depends on the angle between them in the sky.

A clock error is the simplest case. Pulse arrival times are measured against clocks on Earth and then referred to an international time standard. If that chain runs slightly fast for a while, every pulsar appears late by the same amount at the same moment. Any two pulsars are perfectly correlated, whatever the angle between them. This is a monopole pattern.

An ephemeris error is the second case. Arrival times are converted to the centre of mass of the Solar System, which requires knowing where the Earth is relative to that point. If the planets' masses or orbits are slightly off, the assumed position of the Earth is wrong by a small vector. Pulses from pulsars in the direction of that vector appear early, pulses from the opposite side appear late, and pulsars at right angles are unaffected. The correlation follows the cosine of the angle between the two pulsars: a dipole pattern. Tiburzi et al., 2015, Section 2

A gravitational-wave background is the third case. Each wave changes the timing of a pulsar depending on how the line of sight to the pulsar lies against the wave's stretch and squeeze. Average over waves coming from every direction, and two pulsars correlate according to the Hellings–Downs curve.

Table 1.1 Correlation between two pulsars, by angle between them, for the three common signals.
Angle between the pulsarsClock errorEphemeris errorGravitational waves
close together110.5
49 degrees10.660
82 degrees10.14-0.15
90 degrees10-0.14
122 degrees1-0.530
opposite sides1-10.25

The gravitational-wave column uses the normalisation of the NANOGrav paper, where a single pulsar compared with itself would give one. Only the shape matters: it falls from one half, dips below zero between about 49 and 122 degrees, and rises to one quarter for pulsars on opposite sides. Jenet and Romano work out the same kind of curve for sound and radio waves first, which makes the gravitational case easier to follow. Jenet and Romano, 2014, Sections III to V

Why the curve is hard to see

Each pair of pulsars gives only one noisy estimate of the correlation. Most of the 2211 pairs in the NANOGrav array are individually dominated by noise. The curve appears only when many pairs at similar angles are averaged, and even then the points scatter. Part of that scatter cannot be removed with more pulsars: the gravitational waves reaching us come from a particular set of sources, and their interference makes our sky depart a little from the average curve. Bruce Allen calls this cosmic variance. Allen, 2022, Section I

The patterns also overlap. Real pulsars are not spread evenly over the sky, so a clock error can partly mimic the Hellings–Downs shape when only that shape is fitted. Analysts therefore fit all three patterns at once and check how often noise alone would produce the curve, by scrambling the data in ways that destroy any real correlation.

What NANOGrav found, and what it did not

In 2023 NANOGrav reported a common signal across its 67 pulsars whose correlations follow the Hellings–Downs pattern. A model with Hellings–Downs correlations was favoured over one with a common but uncorrelated signal by Bayes factors of 200 to 1000, depending on modelling choices. A test built directly from the pair correlations gave a false-alarm probability between about 5 in 100000 and 2 in 10000, roughly three and a half to four standard deviations. NANOGrav Collaboration, 2023, Abstract

That is evidence, not the five-standard-deviation level physicists usually call a detection. The fitted amplitude agrees with what populations of supermassive black-hole pairs could produce, but the collaboration did not rule out other sources. The data also showed a weak hint of an extra monopole near 4 nanohertz, which the authors could not confirm or explain. NANOGrav Collaboration, 2023, Section 5.3

To see how the timing residual is defined, start the course with what a pulsar timing residual measures. To read the 2023 result figure by figure, go to the guide to the NANOGrav 15-year evidence.

Sources

  1. Jenet and Romano, 2014, Section I
  2. NANOGrav Collaboration, 2023, Section 2
  3. Tiburzi et al., 2015, Section 2
  4. Allen, 2022, Section I

Keep reading

What this page links to directly

You are hereHow can a pulsar timing array tell gravitational waves from a bad clock?

Articles

See this series on the content map →

Pulsar timing arrays and the Hellings–Downs curve · Reading path

From one ticking pulsar to a Galaxy-sized detector

  1. You are hereHow can a pulsar timing array tell gravitational waves from a bad clock?
  2. ArticleWhat a pulsar timing residual measures
  3. ArticleDeriving the Hellings–Downs curve
  4. ArticleThe optimal statistic and its null tests
  5. ArticleRecompute a Hellings–Downs statistic from NANOGrav 15-year data
  6. ArticleReading the NANOGrav 15-year evidence for a gravitational-wave background
Something broken, confusing, or missing on this page?Tell us
What kind of feedback?

Sent with this page’s address, its Lab ID, the site language and your device type. Please leave out personal details. Feedback is kept for up to 12 months.Privacy policy