The Dark Flow

Summary

In September 2008 Alexander Kashlinsky and colleagues reported that galaxy clusters were moving together in a single direction at 600 to 1,000 kilometres per second, coherently across hundreds of megaparsecs, with no sign of the flow dying away at the edge of their catalogue. The signal was a residual dipole in microwave-background maps measured at the positions of 674 clusters. Its proposed cause, offered conditionally, was structure beyond the observable universe.

Four independent analyses found no significant signal. Planck set an upper limit of 254 kilometres per second on gigaparsec scales, called its result in clear contradiction with the claim, and attributed the original dipole to residuals of microwave-background origin in the filtered map. The claim has not been retracted, and the original group still defends it.

What is documented

The claim. The 2008 paper reported "a strong and coherent bulk flow" out to at least about 300 h⁻¹ megaparsecs, a unit scaled by the Hubble constant, which was "the limit of our catalog". It gave "a fairly high amplitude" of about 600 to 1,000 kilometres per second, and a coherence length that "shows no signs of convergence" at that limit.

The method. The microwave background is stacked at the sky positions of X-ray-selected galaxy clusters and a dipole is fitted to the residual. The physical effect relied on is the kinematic Sunyaev-Zel’dovich effect, in which hot gas moving relative to the background shifts the photons passing through it. Turning a microkelvin dipole into a velocity requires an estimate of how strongly the clusters’ gas scatters the background.

The data behind it. A combined catalogue of 782 clusters, of which 674 survived the sky mask, against three-year WMAP data in three bands analysed as eight separate differencing assemblies.

The significance, as the authors themselves stated it. The companion technical paper reports the dipole "at high-signifance [sic] level (~8 sigma in the outer bins)", and in a different section of the same paper, "we detect the dipole at ~2.5 sigma level at each channel". Its summary gives the amplitude as "600-1000 km/sec"; another section claims only a bulk velocity of "at least a few hundred km/sec". Direction was given as galactic longitude 283 and latitude 11 degrees, plus or minus 14.

The 2010 follow-up. Using five-year WMAP data and a larger catalogue, the group reported 934 plus or minus 352 kilometres per second for clusters out to redshift 0.12, 1,230 plus or minus 331 out to 0.16 and 1,042 plus or minus 295 out to 0.20, with the flow "consistent with approximately constant velocity" out to at least about 800 megaparsecs. The paper itself cautions that its calibration may overestimate the amplitude.

The first non-confirmation, 2009. Keisler reanalysed the same data: "We do not find a statistically significant detection of a bulk flow." Accounting for the fact that the primary microwave background is correlated across the eight WMAP channels, rather than independent, "decrease[s] the inferred significance of the detection to 0.7 sigma". Using Kashlinsky’s own best-fit dipole gives 0.8 sigma. The channel-to-channel correlation he measures is about 0.9, which he attributes to the background itself rather than to detector noise.

The second, 2011. Osborne and colleagues found "no significant power in the kinetic SZ signal at these multipoles with either filter", a null consistent with the standard cosmological model, with limits "a factor of ~3 more sensitive than the claimed bulk flow detection". They also quantified a trap: with a filter that lacks spectral dependence, a thermal Sunyaev-Zel’dovich dipole "would be mistakenly measured as a bulk motion of" roughly 2,000 to 4,000 kilometres per second.

The third, 2012. Mody and Hajian, using seven-year WMAP data with Planck and X-ray cluster catalogues, found no significant bulk flow and results consistent with the standard model.

Planck, 2014. Using 1,321 clusters from a larger catalogue: "No bulk flow is detected in any comoving sphere up to the maximum redshift of the sample." The blind upper limit is 254 kilometres per second at 95 per cent confidence for a radius of 2 h⁻¹ gigaparsecs. Planck states its result is "in clear contradiction with some previous claims", naming the 2008 and 2010 papers, and notes that "our error bars are a factor of about 2 smaller".

What Planck found at cluster positions. Planck did measure a dipole. Its argument is that the dipole is unremarkable once the real distribution of clusters on the sky is simulated rather than a random one: about 11 per cent of simulations and sky rotations produce a larger dipole than the data, so "the dipole measured for the real MCXC cluster positions is not peculiar". All measured dipoles were "compatible with zero", and "In no direction is the dipole detected at more than 2 sigma". Planck then reattributes the earlier signal, stating that the roughly three microkelvin dipole of the 2008 paper "should not be assigned to the clusters’ peculiar motion" but "to residuals (mostly of CMB origin) in the filtered map". It also notes that the direction it derives differs from both the microwave dipole and the original claim.

The limits are scale-dependent, and Planck says so. The 254 figure applies at 2 h⁻¹ gigaparsecs. The limits weaken to about 390 and 520 kilometres per second for spheres of about 350 h⁻¹ megaparsecs, and to about 2,000 for spheres of 90 h⁻¹ megaparsecs. The paper states explicitly that its 390 figure "does not permit us to confront claims at the level of 400-700 km s-1 within radii of 50-120" h⁻¹ megaparsecs.

The dissent. Atrio-Barandela, one of the original authors, argued in 2013 that Planck had detected the same dipole but assigned it an inflated uncertainty, that Planck had given the measurement a significance below 90 per cent, and that correcting biases in the error estimation raises this to about 99 per cent. A preprint by four of the original authors argues that the non-detections come from flawed filter implementations. The Registry found no independent analysis adopting either argument.

Leading explanations

The signal is residual microwave background in the filtered map (Planck, 2014; Keisler, 2009). That the dipole is not cluster motion but leftover primary background structure surviving the filter. Planck states this directly. Keisler found the dipole estimates from the eight WMAP channels correlated at about 0.9, and attributed that to the background itself rather than to detector noise.

The detection was never statistically significant (Keisler, 2009; Planck, 2014). That the original significance was overstated by treating correlated channels as independent. Recomputed, it falls to under one sigma. Planck independently finds that about 11 per cent of realistic simulations give a larger dipole than the data.

A real bulk flow caused by structure beyond the observable universe (Kashlinsky and colleagues, 2008; hypothesis, and the measurement it rests on was not reproduced). The original authors offered this conditionally, writing that the flow "may reflect far-away pre-inflationary inhomogeneities". Planck states that flows of the claimed magnitude "cannot be accommodated within a LCDM context", and that its own data contradict the claim.

Planck’s error bars are too large (Atrio-Barandela, 2013; not adopted in any independent analysis the Registry found). That Planck saw the same dipole and overestimated its uncertainty. Planck’s own position is that its error bars are about half the size of the original group’s.

Current status

Debunked. The classification refers to the claimed measurement, not to the people who made it. A coherent cluster bulk flow of 600 to 1,000 kilometres per second was reported in 2008 and extended in 2010. Four independent analyses found no significant signal: the significance was recomputed at 0.7 sigma once the correlation of the microwave background across detector channels was accounted for; a second analysis found no significant power with limits three times more sensitive; a third found a result consistent with the standard model; and Planck set a 254 kilometres per second upper limit at 2 h⁻¹ gigaparsecs, called its result in clear contradiction with the claim, and reattributed the original dipole to background residuals. The claim has not been retracted, and the original group argues that the non-detections stem from filter implementation and that Planck’s uncertainties are overestimated; the Registry found no independent analysis adopting that position. A measurement from a deeper, higher-resolution microwave survey against a larger and better-characterised cluster catalogue, reproduced by two independent pipelines with the real cluster sky distribution simulated, would reopen this record.

Last reviewed: November 2026. Records are provisional. Where the evidence changes, the entry changes. Found an error? Tell us.