MeerKAT detects distant hydrogen without an optical map as a guide

The hydrogen that reached the MeerKAT radio telescope after traveling for roughly 4 to 5 billion years did not appear as a collection of recognizable galaxies. It emerged as an extremely faint statistical pattern buried beneath brighter radio sources and human-made interference. Sourabh Paul, Zhaoting Chen, Mario G. Santos, and Laura Wolz report the first direct detection of this signal’s auto-power spectrum in interferometric intensity-mapping mode, without an optical galaxy catalog to indicate where to look.

The signal comes from the 21-centimeter line of neutral atomic hydrogen, abbreviated H I in radio astronomy. Instead of separating every galaxy, intensity mapping adds the emission of many unresolved galaxies. The power spectrum measures how fluctuations in that intensity are distributed across spatial scales; it is neither a photograph nor a finished three-dimensional map. At these distances, detections had previously relied mainly on correlating radio data with optical surveys. The new result shows that radio observations alone contain recoverable statistical information.

The researchers reanalyzed about 96 hours of commissioning observations made in 2018, spread across nine datasets and using at least 58 antennas. They divided the L band into two windows about 46 megahertz wide, centered at 1,077.5 and 986 megahertz. Those frequencies correspond to redshifts z≈0.32 and z≈0.44: the greater the redshift, the more cosmic expansion has stretched the wave and the older the observed emission. Each window contained 220 channels drawn from the original 4,000 channels at 209-kilohertz resolution.

The decisive problem was separating hydrogen from radio-frequency interference, or RFI, and astrophysical foregrounds. The team split the data into even- and odd-time samples and calculated their cross-power, which removes the mean noise bias. It then applied two filters. The conservative baseline-flagging method, BLF, removed contaminated antenna pairs before calculation and detected the signal at 3.2 standard deviations for z≈0.32 and 3.5 for z≈0.44. The uv-delay-flagging method, UVDF, identified anomalous modes after arranging data in frequency-and-delay space; after excluding the first two scale bins, which were more vulnerable to systematics, the significances were 5.9 and 9.18 standard deviations. These values compare combined power with statistical uncertainty under a zero-signal hypothesis; they do not automatically include every possible systematic error.

The two analyses converged, but their different significances show how strongly the conclusion depends on contamination control. Jackknife tests, which removed each observation block in turn, found no dominant block. A null test crossed the two frequency bands, which should not share the same cosmological signal, and remained consistent with zero. Simulations also showed that filling removed channels loses some signal in particular mathematical directions, although the estimated loss remained below measurement error in the final one-dimensional spectrum. The authors still identify weak, broadband RFI as a risk.

The paper in The Astrophysical Journal Letters uses the spectrum to explore models of neutral-hydrogen mass inside dark-matter halos. The parameters became degenerate, however, when directional dependence was compressed into one-dimensional bins; preliminary constraints required outside information about the cosmic H I density. The data therefore do not yet measure dark energy or deliver the cosmological map envisioned for larger surveys. They demonstrate that a 64-dish instrument can extract the tracer without optical support, even from observations that were not collected for this purpose.

The South African Radio Astronomy Observatory and the University of Manchester announced the result on September 1; Robert Lea examined it in a report published by Space.com on September 4. For the future Square Kilometre Array, of which MeerKAT is a precursor, the achievement and the warning are inseparable. Wider fields and longer integrations could turn this proof of concept into cosmic cartography, provided the next generation of analyses can control still fainter terrestrial signals.