SPHEREx sets early limits on axion-like dark matter

Dark matter’s gravitational effects are well established, but its ingredients remain unknown. One possibility is an axion-like particle, or ALP: a hypothetical particle that, in some models, could make up dark matter and decay into photons. Those photons might appear as a narrow spectral line. Its wavelength would depend on the particle’s mass, giving astronomers a way to search for it.

The authors of this study, Connor Hainje, Elena Pinetti and David Hogg, searched SPHEREx data for such a signal. Their question: do the observations show an unexpected line that could be attributed to ALP dark matter? If not, how strong can the particles’ interaction with light be without producing a signal the telescope should have detected?

Searching the Deep Fields

The researchers analyzed the first 16.5 months of SPHEREx observations, selecting the North and South Deep Fields. Together, the regions used in the analysis covered 213 square degrees. The team excluded measurements flagged as affected by known sources; about half of the measurements in these fields carried that flag. They grouped the remaining measurements into small sky bins and treated each bin’s data as a spectrum.

SPHEREx builds up its spectral coverage as it revisits the sky. Its detectors measure different wavelength ranges, and each pixel samples a slightly different part of a band. The researchers retained the continuous wavelength information rather than combining the measurements into broad spectral channels.

For each candidate particle mass, the team calculated the expected signal from ALP dark matter in the Milky Way, using an assumed model for the galaxy’s dark-matter density and the particles’ velocities. They compared that signal with the data while fitting a smooth, fourth-order polynomial to describe the local background separately for each sky bin and detector. The fit was made in a limited wavelength window around each possible line, and the researchers tested some alternative background choices on a small sky patch.

The analysis also included data-driven adjustments for uncertainties larger than the instrument’s reported errors. One adjustment increased uncertainties when apparent excesses recurred at nearby wavelengths. The final results are 95% power-constrained upper limits, a method intended to prevent unusually low fluctuations from producing overly strong exclusions.

Limits, not a discovery

The paper reports constraints across ALP masses from 0.496 to 3.32 electronvolts. Its tightest reported limit on the coupling to photons is below 2.10 × 10⁻¹¹ GeV⁻¹ at a mass of 0.606 eV. The median coupling limit is below 4.16 × 10⁻¹¹ GeV⁻¹. Expressed instead as a limit on the lifetime of a dark-matter particle decaying into two photons, the median result is a lifetime greater than 6.5 × 10²⁵ seconds.

These are bounds on what the data allow, not evidence that ALPs have been detected. The search found no unknown spectral line at high significance. Some of the strongest apparent excesses line up with expected features, including atmospheric helium emission and the 3.3-micrometre feature associated with polycyclic aromatic hydrocarbons, molecules found in space. The paper discusses these alignments as evidence for known features, but cautions that apparent excesses are not automatically identifiable as emission lines. Other features occur near detector boundaries or known lines. The authors also warn that their reported z-scores should not be read as ordinary “number of sigma” values: even after corrections, the residuals have heavier tails than a simple Gaussian-noise model predicts.

The signal calculation assumes that ALPs account for all dark matter and depends on the adopted model of the Milky Way halo. The analysis covers the Deep Fields, not the whole sky. Foregrounds, calibration, source masking and the background model can all affect sensitivity. In particular, the authors’ uncertainty corrections indicate correlated spectral residuals that their smooth background model does not adequately capture.

Room to improve

The paper compares its limits with two sensitivity benchmarks. The median gap to the calculated Cramér–Rao bound—a statistical benchmark under stated noise and modeling assumptions—is a factor of 39. The gap to the forecast for the completed 25-month mission is a factor of 55. These comparisons suggest room for improvement, not a guarantee that future analyses will reach those benchmarks. The authors point to improved calibration, more realistic treatment of spectral features and future work using the wider sky as possible steps.

This is a November 2026 arXiv preprint, not necessarily peer-reviewed research, so its conclusions should be treated as provisional rather than independent confirmation. The paper also says that some details of the pixel-level database are described in work still in preparation. As SPHEREx collects more data and its processing evolves, the limits may change. For now, the result is an early constraint—not a verdict on what dark matter is.