Cosmic rays are protons and atomic nuclei accelerated to energies far beyond anything achievable in human particle accelerators. They race through the Galaxy, deflected this way and that by tangled magnetic fields, so that by the time they arrive at Earth any clue about their point of origin has been scrambled beyond recognition. Yet when one of these particles collides with a proton in an interstellar gas cloud, the debris includes neutral pions that almost immediately decay into pairs of gamma rays. Those gamma rays, unlike their charged parents, travel in perfectly straight lines. By counting them, astronomers can effectively take an inventory of cosmic rays at the location of the cloud itself, no matter how far away it lies.

This idea is not new. The Fermi Gamma-ray Space Telescope pioneered the technique at gigaelectronvolt energies, measuring cosmic rays in nearby clouds and confirming that the flux in the local solar neighbourhood is broadly representative of the Galaxy on kiloparsec scales. Earlier this decade, the HAWC water Cherenkov observatory in Mexico found evidence for gamma-ray emission from passive molecular clouds at teraelectronvolt energies. What makes the new LHAASO measurement transformative is the energy reach. The collaboration reports spectral energy distributions consistent with gamma rays produced by cosmic rays interacting with the interstellar medium across the range from roughly 1 to 100 teraelectronvolts, meaning the clouds are revealing the population of cosmic-ray protons carrying energies up to several hundreds of teraelectronvolts and beyond, a regime previously accessible only through indirect and model-dependent methods.

The technical challenge is formidable. Individual giant molecular clouds are not bright gamma-ray sources at these energies; the emission is diffuse, extended across many square degrees, and must be extracted from backgrounds of charged cosmic-ray air showers that outnumber gamma-ray events by many thousands to one. LHAASO’s answer is a stacking analysis. Rather than trying to detect each cloud separately, the team aligned the data from five nearby clouds and combined their statistical weight, using the kilometre-squared detector array KM2A to reject the hadronic background and the water Cherenkov detector array WCDA to extend the reach toward lower energies. A likelihood-ratio test applied to the stacked maps produced a clear detection, with the test statistic profile showing a signal that could not plausibly arise from random fluctuations of the background.

The physical logic of the measurement rests on knowing how much target material each cloud contains. The gas mass is estimated from surveys of carbon monoxide line emission, converted to molecular hydrogen mass through a well-studied conversion factor, with corrections for dust opacity informed by Planck satellite maps that also account for so-called dark gas not traced by carbon monoxide. Combining the gamma-ray flux with the gas column density yields the cosmic-ray energy density and spectral shape in situ. The measured spectral energy distributions of the five clouds, when stacked, match what standard models of proton-proton interactions predict if the cosmic-ray spectrum at the clouds resembles the one measured near Earth. That consistency matters: it suggests that the cosmic-ray ‘sea’ is reasonably uniform, at least in the solar vicinity and at teraelectronvolt energies.

The really tantalizing part of the analysis is what it says about the cosmic-ray knee, the famous break in the all-particle cosmic-ray spectrum near a few petaelectronvolts, where the flux steepens dramatically and particles seem unable to reach higher energies. Discovered in 1958 and never fully explained, the knee is widely interpreted as the point where the Galaxy’s accelerators, most likely supernova remnants, run out of steam for protons. Direct measurements by balloon and space experiments such as DAMPE have mapped the proton spectrum up to a few hundred teraelectronvolts, and LHAASO’s own direct measurements have recently pushed precision into the knee region, but the exact break energy of the proton component remains contested. Gamma rays from the clouds offer an independent probe, because the knee in the proton spectrum must imprint itself as a softening in the gamma-ray spectrum at roughly one-tenth of the proton energy.

The collaboration therefore repeated its fit while introducing a deliberate break into the gamma-ray model, scanning over possible break energies and post-break spectral slopes. The result is a nuanced one. No clear knee signature is formally detected in the stacked gamma-ray data; the measured spectra remain compatible with an unbroken power law within the current uncertainties. However, the constraints derived from KM2A are broadly compatible with a proton cosmic-ray knee located at or above about 0.9 petaelectronvolts, although the team is careful to note that somewhat lower break energies cannot be excluded within the present error budget. This is broadly consistent with recent ground-based direct measurements of the proton spectrum in the knee region, adding an independent, calorimetric line of evidence to a decades-old puzzle.

Why does this matter beyond the specifics of one spectral feature? Because the method converts a class of ordinary astronomical objects into a distributed network of cosmic-ray detectors scattered around the Galaxy. Every giant molecular cloud within reach of an instrument like LHAASO becomes a point measurement of the local cosmic-ray density and spectrum, allowing astronomers to test whether the sea is truly uniform or whether it is modulated by nearby accelerators, by variations in the diffusion coefficient, or by the possibility that particle propagation changes character near the knee. Several theoretical works have argued that the diffusion of petaelectronvolt particles should be faster or differently structured than that of teraelectronvolt particles, and some models predict spatial gradients in the cosmic-ray density close to the Sun. The LHAASO clouds provide the empirical test bed for these ideas.

There are also systematic questions that the coming years of data will address. The gamma-ray flux from clouds is proportional to the assumed gas mass, so uncertainties in the carbon monoxide conversion factor propagate directly into the inferred cosmic-ray spectrum. The five clouds in the current sample were chosen for their proximity and high column density, and each additional year of exposure sharpens the stacked signal. With longer baselines, the collaboration could potentially resolve individual clouds rather than relying on stacking, break the sample into subsets at different distances from known supernova remnants, and tighten the limits on where the proton knee truly falls. The detectors themselves continue to operate at 4,400 metres above sea level, where staff maintain the water recycling and power systems year round to keep the arrays running.

For now, the result stands as proof of concept for one of the most elegant ideas in high-energy astrophysics: that the Galaxy’s most mundane structures, cold clouds of molecular hydrogen and carbon monoxide, can be read as detectors for its most exotic particles. In the same way that astronomers learned to treat interstellar gas as a tracer of magnetic fields and star formation, they can now treat it as a calorimeter for petaelectronvolt physics. If cosmic rays hold the secret of the Galaxy’s most powerful accelerators, the clouds, it turns out, have been recording that secret all along. LHAASO has just learned how to read the record, one gamma ray at a time, and the map of cosmic rays across the Milky Way is only beginning to take shape.

Subject of Research: Detection of ultrahigh-energy gamma-ray emission from giant molecular clouds to measure galactic cosmic-ray density and spectra

Article Title: LHAASO detection of ultrahigh-energy γ-ray emission towards giant molecular clouds

Article References: The LHAASO Collaboration, Cao, Z., Aharonian, F., Bai, Y. X., Bao, Y. W., Bastieri, D., Bi, X. J., Bi, Y. J., Bian, W., Blunier, J., Butkevich, A. V., Cai, C. M., Cai, Y. Y., Cao, W. Y., Cao, Z., Chang, J., Chang, J. F., Chen, E. S., Chen, G. H., … Zuo, X. (2026). LHAASO detection of ultrahigh-energy γ-ray emission towards giant molecular clouds. Nature Astronomy. https://doi.org/10.1038/s41550-026-02975-7

Image Credits: AI Generated

DOI: 10.1038/s41550-026-02975-7

Keywords: LHAASO, gamma-ray astronomy, cosmic rays, giant molecular clouds, cosmic-ray knee, PeV, particle astrophysics, KM2A, Nature Astronomy, cosmic-ray propagation, very-high-energy gamma rays, Milky Way