Dark matter around Sagittarius A*

How much invisible mass can hide around Sagittarius A* before a star on a 16-year orbit gives it away?

Welcome to FreeAstroScience. We read the GRAVITY Collaboration’s S2 paper end to end and turned its geometric units into astronomical units, so the null answer on scalar dark matter around Sagittarius A* arrives with distances you can picture.

The GRAVITY Collaboration used almost 30 years of orbital data on the star S2 to test whether a cloud of ultralight scalar dark matter surrounds Sagittarius A*, and found no substantial evidence for one. Where such a cloud would peak inside S2’s orbital range, its mass is bounded below 0.1 percent of the black hole’s 4.3 million solar masses. Every fitted cloud mass in the analysis stays consistent with zero to within 3 sigma.

Sensitivity to dark matter around Sagittarius A* does not behave the way you would expect. S2 pins a scalar cloud down hard for mass couplings between roughly 0.01 and 0.03, and barely touches it on either side of that window. Geometry rather than statistics is what sets those edges: a cloud built from an ultralight field piles its density up at one particular radius, and unless that radius falls between the star’s closest approach and its farthest point, S2 never walks through the part of the cloud that could push it.

What makes S2’s orbit a measuring instrument?

A 16-year period is short enough that two independent programs have now watched the whole loop more than once. S2 carries between 10 and 15 solar masses and shines at an apparent magnitude of K around 14. Membership of the S-cluster puts it among roughly 40 stars packed inside one arcsecond of the Galactic Center. At closest approach the star reaches about 7,650 km/s, close to one hundredth of the speed of light.

No single telescope built this data set.

Three decades of instruments feed the fit, and their precision improved by two orders of magnitude along the way.

One detail from the appendices is worth having. Light from S2 takes different amounts of time to reach us at different points on the orbit, an effect called Rømer’s delay, and averaged over a full orbit it comes to about 8 days. Getting it right mattered enough that the team flagged and corrected a sign error in the 2017 paper by Grould and colleagues whose equation they were working from.

• Two radial velocities from NIRC2 at Keck.
• SHARP on the New Technology Telescope contributed about 10 early astrometric points, accurate to some 4 milliarcseconds.
• NACO on the VLT ran from 2002 to 2019 and produced 118 positions at roughly 0.5 milliarcseconds.
• SINFONI on the VLT supplied 100 radial velocities between 2000 and March 2022, good to 10 or 15 km/s in decent conditions.
• GRAVITY on the VLT interferometer has added 76 points since 2016 at roughly 50 microarcseconds, and that precision is what makes the whole analysis possible.

We are writing this at the start of August 2026, and the newest thing in our own archive on Sagittarius A* is a piece from June by Denise Meloni. It asks where those stars came from rather than what they can weigh, and our June look at the origin of the S-star orbits sits well beside this one. The two run in opposite directions: hers is formation history, ours is what the motion of a single star rules out about everything else in there.

What is a scalar cloud, and how would one form?

Light bosonic fields turn up in string-inspired theories, and the oldest version of the idea is older than string theory. Peccei and Quinn proposed one in 1977 to explain why the neutron’s electric dipole moment is so small, and the axion came out of that work. Around a spinning black hole, small fluctuations of such a field can be amplified by superradiance until they settle into a bound state outside the horizon, a condensate that Brito, Cardoso and Pani showed in 2015 can carry up to roughly 10 percent of the hole’s mass. The paper tracks that structure with two numbers: lambda, the cloud’s mass as a fraction of the black hole’s, and alpha, a dimensionless coupling standing in for the field’s particle mass.

A mismatch sits inside this setup, and the authors are upfront about it. Kodama and Yoshino calculated in 2012 that for a hole of Sagittarius A*’s mass, superradiance can grow a cloud on a timescale shorter than the age of the universe only if the field’s effective mass lies between 10⁻¹⁸ and 10⁻¹⁵ eV. The band that would leave a detectable mark on S2’s orbit sits lower, from 10⁻²⁰ up to 10⁻¹⁸ eV. They barely touch. So if a cloud is there and S2 can feel it, superradiance is not how it got there, and the GRAVITY authors say exactly that before pointing at work by Cardoso and collaborators in 2022 showing that even a non-spinning hole can hold on to a cloud of primordial origin.

If the wave-like side of ultralight dark matter is new to you, our explainer on why axion dark matter behaves like a classical field covers the ground this paper assumes you already have.

Why the constraint only bites in a narrow window

Scalar density is not a smooth halo thinning outward. Instead it carries a peak, and the radius of that peak comes out of the model in closed form.

R_peak = 3M / α²

Here R_peak is the radius at which the cloud’s density is highest, M is the black hole’s mass written as a length in the units relativists use, and alpha is the mass coupling. Said in words: a heavier boson pulls the cloud in tighter, and it does so as the inverse square of alpha. Across the range explored in the paper that places the peak between about 3,000 and 30,000 M, which happens to be the span between S2’s closest approach and its farthest point.

Turning the paper’s units into distances you can picture

M as a unit of length is convenient for a relativist and useless for everybody else. So we converted it.

The fitted values in the paper give S2’s semi-major axis as 0.12497 arcseconds and the distance to the Galactic Center as 8,278 parsecs, which is about 27,000 light-years. An angle in arcseconds multiplied by a distance in parsecs is a length in astronomical units, so that semi-major axis is 1,035 AU. Feed in the fitted eccentricity of 0.88441 and the closest approach comes out at 120 AU, the farthest point at roughly 1,950 AU. That first figure is the one worth pausing on, because it is what the popular coverage quotes: the Italian write-up on Reccom gives about 120 astronomical units for the May 2018 pericenter passage, and the fitted time of periastron in the paper, 2018.379, lands in mid-May. Two routes, same number.

Now the geometric units fall into place. Periastron is quoted at about 3,000 M and apoastron near 50,000 M, so one M is close to 0.04 AU, and the peak radius of a scalar cloud stops being an abstraction.

At alpha = 0.01 the density peak sits at 30,000 M, some 1,200 AU out, in the far half of the orbit where S2 spends most of its time. Raise alpha to 0.03 and the peak pulls in to 3,333 M, roughly 133 AU, sitting just outside the star’s closest approach. Push it to 0.075 and the paper puts the peak at 530 M, about 21 AU, a region S2 never visits. That last case is why the whole constraint collapses at high alpha. Inside the model the cloud is perfectly real, and the star simply orbits outside it.

Our two conversions do not agree perfectly, and you should know that before you repeat them. Working through the orbital elements gives an apoastron of 1,950 AU. Going through the rounded 50,000 M gives about 1,995 AU instead. The 2 percent gap comes from the paper quoting its geometric figures to one digit, which is not a mistake in either place.

How much dark matter can sit around Sagittarius A*?

Lambda comes out small, and more to the point it comes out consistent with zero. At alpha = 0.01 the maximum likelihood value is 0.00361 ± 0.00147, and at alpha = 0.02 and 0.03 it falls to 0.00075 ± 0.00030 and 0.00073 ± 0.00029. Every value across the range is compatible with zero within 3 sigma. Converted into mass, the discussion section concludes that clouds with alpha between 0.015 and 0.045 are excluded whenever lambda exceeds 0.1 percent of the central mass, which for a black hole of 4.3 million solar masses means roughly 4,000 solar masses of scalar field.

Bayes factors deserve slower reading. On the Kass and Raftery scale the paper adopts, log₁₀ K above 2 counts as decisive evidence for the cloud model. In the paper’s own table the largest entry is 1.44. Most of the range sits near 1.3, which is mild, and a single row at alpha = 0.0035 drops to −10.58, decisive evidence against a cloud at that particular coupling. Why that one row falls off a cliff while its neighbors sit near plus one is not explained in the text, and we could not work it out from what is printed.

Table 1 — upper bounds on extra mass inside S2’s orbit (GRAVITY Collaboration, arXiv:2306.17215v4, 2023)

None of this challenges general relativity, whatever the headline over the Italian coverage says. That piece puts “sfida la relatività” in its title, the star that challenges relativity. S2 does the reverse. Its orbit is what delivered the Schwarzschild precession at 7 sigma, with the precession scaling factor coming out at 0.99 ± 0.15 where Einstein predicts exactly 1, and the very same orbit is what produces these dark matter bounds. This analysis fixes the factor at 1 and moves on. For the same kind of test run much closer to home, our report on the LARES-2 frame-dragging measurement covers the Solar System version.

One disagreement is worth flagging because it belongs to the authors themselves. Yuan and colleagues published a far tighter bound in 2022, below 10⁻⁴ of the hole’s mass for a scalar of 10⁻¹⁸ eV, and they used only publicly available data. The GRAVITY authors point out that the orbital parameters coming out of that work are not compatible within 3 sigma with the ones the collaboration published the same year, and they argue a discrepancy appearing that early makes the tighter bound unsafe to quote.

The cloud was pinned into S2’s orbital plane

Scalar potential in this model is not spherically symmetric, which creates a problem the authors state plainly. Because the cloud takes its shape from the black hole’s spin axis, each star’s inclination relative to that axis matters, and the same analysis cannot be carried out straightforwardly for several stars at once. They fixed S2’s initial angular position in the equatorial plane instead, theta equal to pi over two, and that choice maximizes the scalar potential. Doing so gives the cloud its best chance of showing up, which is the right way to go looking for something and not a neutral way to bound it.

Two inclinations were tested, not a range. Comparing theta = 0 with theta = pi/2, the largest relative difference in astrometry is about 25 percent at alpha = 0.01, and the largest in radial velocity about 15 percent at alpha = 0.045, both reached only during the two periastron passages. What the paper does not report is a scan across intermediate angles, so we cannot tell from it how far a genuinely misaligned cloud would loosen the bound.

Then there is the spin. Nobody has an agreed value for how fast Sagittarius A* rotates, and the paper lays the disagreement out rather than picking a side. Fragione and Loeb put the spin parameter below 0.1 in 2020. Qi and colleagues questioned that in 2021 and argued the astrometry is not yet good enough to fix it, while Kato and collaborators had already extracted 0.44 ± 0.08 from quasi-periodic radio oscillations in 2010, and the Event Horizon Telescope estimate from Broderick and colleagues in 2011 is 0.00 ± 0.64. Not knowing whether the hole spins fast enough for superradiance is precisely why a primordial formation route stays on the table.

We are leaving one thing out on purpose. Sengo and colleagues bounded the same fields in 2023 from the Event Horizon Telescope image rather than from stellar orbits, at roughly the 10 percent level, and the GRAVITY paper notes it improves on that considerably. Those are different instruments with different systematics, and that deserves an article rather than a paragraph here.

Where a null answer leaves Sagittarius A*

What the fit actually produced: a cloud fraction of 0.00361 ± 0.00147 at alpha = 0.01, every value consistent with zero at 3 sigma, and a Bayes factor that never reaches the decisive threshold of 2. The bound bites only where the cloud’s density peak lands between roughly 120 and 1,950 AU, and it was obtained with S2 forced into the black hole’s equatorial plane.

We wrote it out at this length because “no dark matter found” gives you nothing to argue with, and arguing is what this place is for. Turning a fourteen-parameter Markov chain into sentences you already own is the job, and we would rather you could check us than be impressed by us. Keep the lights on upstairs, because a mind that stops asking is where bad ideas move in. Go and disagree with our reading of that alpha = 0.0035 row. GRAVITY+ and ERIS are in commissioning now, and when they reach stars further in we will run the new numbers against these. FreeAstroScience, Rimini. Gerd Dani.

Sources

1. GRAVITY Collaboration: Foschi A., Abuter R., Aimar N., Amaro Seoane P., Cardoso V., Eisenhauer F., Garcia P.J.V., Genzel R., Gillessen S., et al. (2023). Using the motion of S2 to constrain scalar clouds around Sgr A*. Monthly Notices of the Royal Astronomical Society. Preprint dated 6 September 2023, arXiv:2306.17215v4. https://arxiv.org/abs/2306.17215
2. Meloni D. (2026). S2: la stella che sfida la relatività al centro della galassia. Reccom Network, Astronomia. Published 31 July 2026.