Dark matter stopped losing, but it hasn't won yet
At the luminous, chaotic heart of our galaxy, a decades-old glow refuses to yield its secret. A team from the University of Vienna and Lawrence Berkeley National Laboratory has applied machine learning to over a million simulated gamma-ray observations, and in doing so has quietly unsettled what many considered a settled debate. The Galactic Center Excess—that broad wash of gamma rays emanating from the Milky Way's core—may yet be the fingerprint of dark matter, a possibility that had been largely set aside in favor of neutron stars. Science, it seems, has been reminded that premature certainty is its own kind of darkness.
- A mysterious gamma-ray glow at the Milky Way's center has divided astronomers for decades, with dark matter and spinning neutron stars each claiming the role of culprit.
- A new machine learning system, trained on over a million simulated observations and incorporating photon energy data for the first time, has sharpened the tools of detection beyond what earlier statistical methods could achieve.
- The pulsar explanation now strains under its own weight—generating the observed signal would require at least 35,000 millisecond pulsars, a figure that dwarfs all previous estimates by orders of magnitude.
- The dark matter hypothesis, once sidelined, has regained credibility: any unresolved point sources dim enough to fit the data would be nearly indistinguishable from what dark matter annihilation would produce.
- No winner has been declared—the study reopens the question rather than resolving it, leaving one of astronomy's most enduring mysteries more complex and more alive than the consensus had allowed.
At the heart of the Milky Way, something is glowing. For decades, astronomers have detected a broad wash of gamma rays from the galactic center—a signal stretching across thousands of light years known as the Galactic Center Excess. Two explanations have long competed: dark matter annihilating itself in the cosmic depths, or the accumulated radiation from thousands of rapidly spinning neutron stars called millisecond pulsars. For years, the pulsar explanation held the upper hand. Dark matter, never directly observed despite comprising much of the universe's mass, remained the more speculative candidate.
What changed is method. Researchers at the University of Vienna and Lawrence Berkeley National Laboratory trained a machine learning system on more than one million simulated gamma-ray observations. The key innovation was incorporating the energy of each detected photon—information that earlier statistical studies had largely overlooked. Analyzing both the origin and the energy of the gamma rays allowed the team to test competing explanations with unprecedented precision.
The results complicated the picture considerably. Any unresolved point sources dim enough to fit the data would be nearly indistinguishable from what dark matter annihilation would produce—undermining one of the strongest arguments previously used to rule dark matter out. The pulsar hypothesis, meanwhile, now requires at least 35,000 millisecond pulsars packed into the galactic center, orders of magnitude beyond earlier estimates of a few hundred to a few thousand.
Neither explanation emerges clean. As study author Florian List noted, the work does not prove dark matter is responsible—but it makes clear that dismissing the possibility is premature. The Galactic Center Excess remains a mystery, only now a more complicated and more interesting one than the field had recently been willing to admit.
At the heart of the Milky Way, something is glowing. For decades, astronomers have detected a broad wash of gamma rays emanating from the galactic center, a signal that stretches across thousands of light years. The question of what produces this light—known as the Galactic Center Excess—has divided the field. Is it the signature of dark matter annihilating itself in the cosmic depths, or the accumulated radiation from thousands of rapidly spinning neutron stars? A new analysis using machine learning suggests the answer may not be as settled as many believed.
The mystery has persisted because the galactic center is one of the most crowded and luminous regions of the gamma-ray sky. Sorting signal from noise there is like trying to identify individual voices in a stadium full of people. For years, the prevailing view among researchers favored the pulsar explanation—the idea that a large population of millisecond pulsars, those exotic spinning remnants of dead stars, could account for the glow. Dark matter, by contrast, remained a more speculative possibility, a form of matter that comprises much of the universe's mass yet has never been directly observed.
What changed is method. Researchers at the University of Vienna and Lawrence Berkeley National Laboratory trained a machine learning system on more than one million simulated gamma-ray observations. The crucial innovation was including a piece of information that earlier statistical studies had largely ignored: the energy of each individual photon detected. By analyzing both where the gamma rays were coming from and how energetic they were, the team could test competing explanations with greater precision than before.
The results shifted the landscape. If the glow comes from unresolved point sources—bright objects too distant or faint to resolve individually—those sources would need to be extraordinarily dim, so faint that they would be nearly indistinguishable from what dark matter annihilation would produce. This finding undermines one of the strongest arguments that had been used to rule out dark matter. Meanwhile, the pulsar hypothesis faces its own problem. For millisecond pulsars to generate the observed signal, there would need to be at least 35,000 of them crammed into the galactic center. That is orders of magnitude more than the few hundred to few thousand sources that earlier estimates had suggested.
Neither explanation now looks clean. The dark matter scenario remains viable, but unproven. The pulsar scenario requires an implausibly dense population of sources. The machine learning analysis does not declare a winner; instead, it reopens a question that many thought had been settled. Florian List, one of the study's authors, put it plainly: the work does not show that dark matter is responsible for the signal, but it demonstrates that it is premature to dismiss the possibility. One of astronomy's most enduring mysteries remains exactly that—a mystery, but one that now looks more complicated and more interesting than the consensus had suggested.
Notable Quotes
The origin of the Galactic Center Excess is one of the longest-running debates in astrophysics. 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.— Florian List, University of Vienna
The sources would have to be so faint that they would be almost indistinguishable from the emission expected from annihilating dark matter.— Nick Rodd, Lawrence Berkeley National Laboratory
The Hearth Conversation Another angle on the story
Why does the energy of individual photons matter so much? Couldn't you just look at where the light is coming from?
You could, and that's what earlier studies did. But if you only know location, you're missing half the story. Dark matter and pulsars produce gamma rays in different ways—different energy signatures. By ignoring that information, the old analyses were essentially working with one eye closed.
So the machine learning didn't discover anything new—it just used data that was already there?
Exactly. The data existed. But analyzing a million simulated scenarios by hand would be impossible. Machine learning let them test how well each hypothesis fit the full picture at once.
If you need 35,000 pulsars, why is that a problem? Space is big.
It is big, but the galactic center is a specific place. We have some sense of how many pulsars should exist there based on stellar physics. 35,000 would be wildly overcrowded—like fitting a city's worth of people into a small town.
Does this mean dark matter is winning?
Not quite. It means dark matter stopped losing. The study weakened the case against it, but didn't prove it's there. The mystery is still open.
What happens next?
More data, better telescopes, new ways of looking. This analysis is a checkpoint, not a destination.