Search for an Isotropic Gravitational-wave Background with the Parkes Pulsar Timing Array

If black holes merge, they should ring the universe like a bell — in gravitational waves so low and slow that no human-built detector could hear them. The frequencies involved are not the hundreds of hertz that the Laser Interferometer Gravitational-Wave Observatory detects from stellar-mass collisions. They are nanohertz: one oscillation every few years. To hear them, you need a detector the size of the galaxy. But here’s the thing — you can build one, if you’re patient enough, and if you know where to look. A team of astronomers in Australia did exactly that. And something is humming. The instrument is called a pulsar timing array, and it works like this. Millisecond pulsars — neutron stars spinning hundreds of times per second — are among the most stable natural clocks in the universe. When a gravitational wave passes between Earth and one of these pulsars, it very slightly stretches or squeezes the intervening space. The pulses arrive a little earlier or a little later than expected. That shift is tiny, but it’s real, and it’s measurable. The Parkes Pulsar Timing Array, or PPTA, run by Reardon and colleagues at the Parkes Observatory in Australia, has been monitoring 30 of these millisecond pulsars for 18 years. That’s the dataset at the center of this paper. The reason you need so many pulsars watched for so long comes down to what you’re actually looking for. Individual pulsars have their own noise — spin irregularities, fluctuations in the interstellar medium, all achromatic and local. A true gravitational wave background would leave a different kind of mark: a common signal across all pulsars simultaneously, with a very specific spatial pattern depending on the angle between any two pulsars in the sky. That pattern is called the Hellings-Downs correlation, and it’s the fingerprint of gravitational waves. Two pulsars pointing almost the same direction should be strongly correlated. Pulsars separated by about ninety degrees show a weaker correlation. And pulsars pointing in nearly opposite directions show a characteristic negative correlation. No other common astrophysical or instrumental effect produces exactly this angular dependence. So the Hellings-Downs curve is not just a detail — it’s the entire case for gravitational wave detection. To describe how a background's strength changes with frequency, Reardon and colleagues use what’s called a characteristic strain spectrum: the strain, h-sub-c, equals an amplitude A, multiplied by the ratio of frequency to one cycle per year, raised to a power alpha. Alpha is the spectral index — it tells you the shape of the signal across frequencies. For an isotropic background from inspiraling supermassive black hole binaries on circular orbits, theory predicts alpha equals minus two-thirds. That’s the canonical target. The corresponding power-spectral index, gamma, equals thirteen-thirds, or about 4.33. Now, how do you actually pull this signal out of 18 years of data from 30 pulsars? Reardon and colleagues used Bayesian inference — a framework that asks how probable each possible signal configuration is, given the data. Instead of returning a single best-fit number, it returns a probability distribution over the full range of plausible signal parameters. Two strategies were central. First, a factorized likelihood approach: if you assume the spectrum follows the expected power law with gamma fixed at thirteen-thirds, each pulsar’s data can be analyzed semi-independently, and the results multiplied together to give the array-wide answer. Second, they performed a free-spectrum analysis, where instead of assuming a power law, they let the power in each Fourier frequency bin vary independently. This is a way of asking: does the data actually look like a power law, or is the power scattered across frequencies in some other way? The free-spectrum result was telling. Most of the recovered power sat at the lowest frequencies the dataset is sensitive to — exactly what you’d expect from a red-noise process, one that gets stronger as you go to lower frequencies. There was also a localized excess at around 14 nanohertz, but the dominant story is a signal concentrated at the longest timescales. For the key numbers: with the spectral index left free, Reardon and colleagues measured a spectral slope gamma of 3.87 plus or minus 0.36, and a log-amplitude of negative 14.50, with uncertainties of plus 0.16 and minus 0.14. In linear terms, that amplitude A works out to roughly 3.1 times ten to the minus 15, with asymmetric uncertainties of plus 1.3 and minus 0.9. When the spectral index is fixed at the canonical binary black hole value of gamma equals thirteen-thirds, the recovered amplitude is A equals 2.04 times ten to the minus 15, with a 68 percent credible interval of plus 0.22 and minus 0.25 times ten to the minus 15. Those are small, dimensionless strains — a measure of how much space itself is being stretched by the background. Then comes the crucial test: do the pulsars show Hellings-Downs spatial correlations? This is where the paper becomes both exciting and honest. Using all 435 unique pulsar pairs — with an effective pair count of 355 — the team computed pairwise cross-correlations and compared them to the Hellings-Downs prediction. To assess whether any apparent correlation could be a statistical fluke, they generated ten thousand randomized sky configurations, shuffling pulsar positions to destroy any real spatial signal. The result: the observed support for Hellings-Downs correlations corresponds to a one-sided false-alarm probability of about 0.014. That’s roughly a two-sigma level. Suggestive. Not definitive. The log-likelihood ratio favoring Hellings-Downs over zero correlation was 1.1 — meaningful, but modest. And the per-pulsar picture was uneven. PSR J1909-3744 was the single most influential pulsar, dominating the recovered amplitude. Three others — J1744-1134, J1603-7202, and J1713+0747 — actually offered negative likelihood support for the canonical spectral model at the measured amplitude. That kind of disagreement between individual pulsars is a flag worth taking seriously. The team also examined whether the signal could be explained by something other than gravitational waves. Clock errors and solar system ephemeris uncertainties — errors in the precise mapping of planetary positions, which are needed to translate raw arrival times to a reference frame — can masquerade as common signals across pulsars. Reardon and colleagues marginalized over perturbations to planetary masses and Jupiter's orbital elements using a framework called BAYESEPHEM. They found weak but nonzero evidence for at least one ephemeris perturbation in every solar system model they tested. Marginalizing over those uncertainties steepened the recovered spectral index slightly, to gamma around 4.02, and shifted the amplitude modestly. None of this removed the common signal — but it underscored that disentangling real gravitational waves from systematic errors in our solar system map is genuinely hard. The spatial correlations are consistent with Hellings-Downs at the two-sigma level, but the full Parkes Pulsar Timing Array likelihood does not yet produce an unambiguous detection. If the signal is astrophysical, the picture it paints is vast. Hundreds of millions of supermassive black hole pairs — each at the center of a merging galaxy pair — slowly spiraling together across billions of years, each radiating gravitational waves, all of those waves adding up into an incoherent background hum that fills the universe. The Parkes Pulsar Timing Array measurement is one window onto that hum. But the Parkes Pulsar Timing Array was not working in isolation. Reardon and colleagues compared their amplitude to contemporaneous results from other arrays. Previous Parkes Pulsar Timing Array work by Goncharov and colleagues found a spectral index gamma of 4.11 and a log-amplitude around negative 14.55. The European Pulsar Timing Array found gamma of 3.78 and a log-amplitude of negative 14.29. The North American Nanohertz Observatory for Gravitational Waves reported a steeper median spectral index of 5.52 with a log-amplitude of negative 14.71. The International Pulsar Timing Array combination gave gamma of 3.90 plus or minus 0.90. These numbers are not identical, but they are broadly consistent — multiple independent arrays, watching different pulsars with different telescopes, all seeing roughly the same thing in the same frequency range. The path to certainty runs through the International Pulsar Timing Array, which combines all of these datasets. More pulsars, longer baselines, better cross-calibration between arrays — that’s how you accumulate enough pulsar pairs to push the Hellings-Downs detection above any reasonable threshold of doubt. What Reardon and colleagues have delivered is a careful, honest accounting of where things stood with the Parkes Pulsar Timing Array’s 18-year, 30-pulsar dataset: a common signal that looks like what a gravitational wave background should look like, spatial correlations that favor Hellings-Downs at the two-sigma level, and a set of open questions about individual pulsars and solar system systematics that the community will need to resolve. The hum is there. Whether it’s the one we’ve been waiting for is what the next generation of observations will decide. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.