Dark matter at the Milky Way’s core: machine learning reopens a decade-long debate

For more than a decade, a faint but persistent glow has emanated from the center of our galaxy, defying clean explanation. Known as the Galactic Center Excess (GCE), this roughly spherical halo of gamma-ray light stretches thousands of light-years around the galactic nucleus and was first detected by NASA’s Fermi Gamma-ray Space Telescope in the early 2010s. It has divided astrophysicists ever since. Is it the long-sought fingerprint of dark matter particles annihilating one another into high-energy radiation? Or is it simply the combined, unresolved glow of tens of thousands of faint neutron stars?

A study published in Physical Review Letters on June 12, 2026, by an international team from the University of Vienna and Lawrence Berkeley National Laboratory has reopened the debate in decisive fashion. Not by providing a final answer, but by dismantling one of the strongest arguments that had been leveled against dark matter as the source.

The Galactic Center Excess: a signal with no consensus

The GCE is not a marginal anomaly. It is a statistically robust feature of the gamma-ray sky that has withstood years of independent scrutiny. The signal peaks at energies of a few GeV — fully consistent with theoretical predictions for the annihilation of weakly interacting massive particles (WIMPs), among the most extensively studied dark matter candidates in particle physics. When two such particles meet and annihilate, they produce gamma rays with a characteristic energy spectrum. The spatial distribution of the GCE — spherically symmetric and centered on the galactic nucleus — matches what models of dark matter halos predict.

And yet, identifying dark matter as the source has remained deeply controversial, because the same signal can also be explained by a far more familiar class of objects.

Millisecond pulsars: the competing hypothesis

Millisecond pulsars are neutron stars — the extremely dense collapsed remnants of massive stars — that rotate hundreds of times per second and are powerful natural emitters of gamma radiation. A sufficiently large population of them concentrated near the galactic center could, in principle, produce an aggregate glow whose properties resemble those of the GCE.

This pulsar hypothesis gained substantial traction over the years through a particular line of evidence. Statistical analyses of the GCE’s spatial structure — employing techniques based on photon counting statistics — found that the emission appeared «clumpy,» pointing to many discrete faint sources rather than a smooth diffuse background. That clumpiness was consistent with an unresolved population of point sources, not with the homogeneous emission expected from dark matter distributed throughout a halo.

This spatial argument became one of the primary reasons many researchers had effectively set aside the dark matter interpretation. For several years, the pulsar hypothesis appeared to be settling the question.

The missing variable: photon energy

The new study, led by Florian List, Yujin Park, Nicholas L. Rodd, Eve Schoen, and Florian Wolf, identified a critical limitation in all prior analyses: every earlier statistical assessment had examined only the spatial distribution of gamma-ray photons — where they came from in the sky — while discarding an equally important piece of information: the energy of each individual detected photon.

This was not a conceptual oversight but a genuine computational obstacle. Incorporating spectral information — the full energy distribution of the detected photons — alongside spatial data, and doing so simultaneously at the scale of real Fermi observations, required tools that simply had not existed before.

The team overcame this barrier by constructing a neural network simulation-based inference approach: a machine-learning algorithm trained on more than one million simulated gamma-ray observations. For the first time in the history of the GCE debate, the model was able to process spatial structure and spectral data simultaneously. What it found changed the picture substantially.

What the energy information reveals

When photon energies are included in the analysis, the spatial argument that had favored pulsars does not disappear — it transforms into something that undermines the pulsar hypothesis itself.

Previous purely spatial analyses had inferred the presence of point sources with comparatively bright individual fluxes. The implication was that a few hundred to a few thousand moderately luminous millisecond pulsars could account for the excess. The new analysis overturns this reading. With spectral constraints applied, the inferred point sources are not moderately bright. They are extremely faint — so faint, in fact, that their individual gamma-ray output would be nearly indistinguishable from the diffuse emission produced by dark matter annihilation.

This has a stark consequence. To generate the full GCE from sources that dim, the Milky Way’s central region would need to harbor a minimum of 35,000 millisecond pulsars — an estimate that is orders of magnitude larger than the few hundred to few thousand assumed in most previous pulsar models. That figure is not physically ruled out, but it represents a far more extreme demand on the pulsar hypothesis than had previously been recognized, and it substantially erodes the evidentiary case for pulsars as the dominant explanation.

«Our new analysis shows that the sources would have to be so faint that they would be almost indistinguishable from the emission expected from annihilating dark matter,» said Nick Rodd, study co-author and physicist at Lawrence Berkeley National Laboratory.

The debate is not over — and that is the finding

Precision matters here. This study does not demonstrate that dark matter is responsible for the GCE. It does not detect a dark matter signal. What it accomplishes is something structurally important: it removes one of the most cited empirical arguments against the dark matter interpretation, revealing that argument to have been built on incomplete data.

«The origin of the Galactic Center Excess is one of the longest-running debates in astrophysics,» said Florian List, lead author and researcher at the University of Vienna. «Our work does not show that dark matter is responsible for the signal. However, it suggests that it is still too early to rule out this possibility.»

Both explanations — dark matter annihilation and a hidden population of extremely faint pulsars — now rest on roughly equal epistemic footing. Neither is confirmed, and neither can be dismissed on the basis of the current data alone. The result is scientifically meaningful precisely because it restores genuine uncertainty where premature certainty had begun to settle in.

What comes next

Resolving the GCE will almost certainly require the next generation of gamma-ray observatories. The Cherenkov Telescope Array Observatory (CTAO), currently under construction at sites in northern Chile and southern Spain, is designed to detect gamma rays at angular resolutions and sensitivities far exceeding those of Fermi. It is expected to deliver science-quality data by approximately 2027. With those capabilities, CTAO will be positioned to determine whether discrete gamma-ray sources are present in the galactic center region, how bright they are, and whether their spectral signatures match the predictions of the dark matter annihilation scenario, the pulsar scenario, or neither.

The work by List and colleagues is more than a methodological advance. It is a reminder that in astrophysics, the strength of an argument is only as sound as the completeness of the method used to test it. For a decade, purely spatial analyses had been examining half the available information. The other half — the energy of each photon — is now part of the analysis.

And it says the question is still very much open.

If this analysis opened questions for you about dark matter, the Milky Way’s structure, or the tools modern astrophysics uses to probe the unknown, share it with someone who looks at the night sky and wonders.

© 2026 SKYCR.ORG | Homer Dávila Gutiérrez, FRAS. All rights reserved. Total or partial reproduction is prohibited without express authorization. More information: Physical Review Letters (2026). DOI: 10.1103/dkcq-6y4f. On arXiv: arxiv.org/abs/2507.17804