A faint point of light showed up 15 milliarcseconds northwest of Sagittarius A* in the spring of 2023, not in a raw frame but in an image reconstructed from interferometry data. Its position and the speed of its motion made it quickly clear that it might be on a tight orbit.

They labelled it S301. It now holds the record for the shortest known orbital period around the Milky Way’s central black hole: one circuit in 8.7 years. Its closest approach is 136 or 142 Schwarzschild radii, around ten times nearer in than S2, the star that has anchored this field for two decades. The discovery was published in Nature on 19 August by the team behind the GRAVITY+ instrument.

A star that no fitting code would have found

The Galactic Centre has been watched with the GRAVITY instrument on the Very Large Telescope Interferometer since 2017. Observing runs come in roughly week-long campaigns, monthly between March and September, and use 80 to 100 hours of telescope time a year. That programme works by fitting models of the known stars to the interferometric data, and by design that procedure does not reveal new ones.

S301 came out of the other half of the analysis, in which the data are inverted into an actual image. The accepted manuscript is blunt about the difference: finding a star like this with a fitting code alone would be, in the authors’ phrase, de facto impossible. The fit settles into a local minimum, while the imaging explores the full parameter space. A second, more classical imaging routine recovered the same star independently.

Once a preliminary orbit existed, the team could work backwards and ask whether S301 ought to have been sitting in older pointings taken for other purposes. It was, strongly in 2021 and weakly in 2017. Nineteen astrometric positions now trace an ellipse across more than eight years.

More than eight percent of the speed of light

The orbit that comes out of those nineteen positions is extreme on every parameter:

  • a period of 8.68 years, against 12 years for S55, the previous record holder, and 16 for S2
  • an eccentricity of 0.9832 or 0.9821, which brings the star to 136 or 142 times the black hole’s Schwarzschild radius at closest approach, depending on which of two possible orbit orientations is the real one, where the equivalent figure for S2 is about 1,400
  • a speed at that point of 25,600 or 25,000 kilometres a second, which is 8.5 or 8.3 percent of the speed of light

On that fitted orbit, general relativity should advance the point of closest approach by 2.0 or 1.9 degrees each circuit, turning the ellipse once around inside its own plane every 1,560 or 1,630 years. The present data are consistent with that prediction and not yet sharp enough to confirm it. The star passed pericentre in early 2023, shortly before it was noticed.

The missing spectrum

S301 has never been caught in a spectrum. Deep integral-field spectroscopy has failed to detect it either as a continuum source or through any spectral feature, so there is no radial velocity. Without one, the orbit has two mirror solutions in three-dimensional orientation that the present data cannot tell apart. Their fit qualities are indistinguishable, and the paper reports the two as leaving semi-major axis, eccentricity and time of periastron unchanged within errors; what separates them is how the ellipse is turned in space, and the small residual difference in eccentricity is what produces the two pericentre distances above.

The star is faint even by the standards of this field: K-band magnitude 19.3, against a limit near 20, and roughly five magnitudes fainter than S2. Its mass has never been measured. From that single magnitude, an assumed extinction and an assumed distance, the authors infer an early F-type main-sequence star of somewhere between about 1.1 and 1.5 solar masses, the figure depending on the star’s age.

A second argument supports the inference without measuring anything. A giant star would have begun losing part of its envelope at a pericentre this deep, and that mass loss would have shown up in the orbit, which it does not. A main-sequence star of this size passes comfortably outside its own tidal radius and comes through untouched.

No spin has been measured

The spin is why the discovery matters, and it is the part of the paper that forecasts rather than reports. Nothing in this work constrains how fast Sagittarius A* rotates.

What the paper computes is how large the effect would be if the black hole were spinning as fast as it can, with its spin axis lined up with the star’s orbit. In that case the in-plane contribution of frame dragging would advance the star’s pericentre by 0.11 degrees per orbit, alongside the 1.9 degrees per orbit that ordinary relativistic precession contributes to the same ellipse.

To ask when a signal that size becomes measurable, the team built a synthetic dataset. The existing positions were extended with simulated observations from 2026 to 2035, ten points a year plus ten more around pericentre. The dimensionless spin parameter was, in the authors’ word, optimistically set to its maximum value of 1, and the spin axis was assumed to be roughly aligned with the orbit. The astrometry was assumed good to 100 microarcseconds and the radial velocities good to 1 kilometre per second. Under those conditions the spin would be recovered with an uncertainty below 0.2, separating a maximally spinning black hole from a non-spinning one at better than five sigma.

Several of those conditions do not exist. The astrometric precision is the expected final performance of the upgraded instrument, which it has not yet reached; the radial velocities would need the Extremely Large Telescope; and S301 has not been detected spectroscopically at all. The authors add that a robust spin constraint will require modelling to second post-Newtonian order, to avoid systematic biases at low spin.

Then there is the crowd. The authors model a population of stellar-mass black holes around the orbit and find it would tilt the orbital plane by about 0.65 arcminutes per revolution, roughly a hundredth of a degree. They call that typically subdominant to frame dragging, provided the black hole’s spin is moderate to high and favourably oriented. The proviso is load-bearing, because the size and orientation of the spin are exactly what the measurement is trying to establish.

The two effects are at least separable in principle, since frame dragging concentrates near pericentre while a granular stellar background shows up most near apocentre. One extreme configuration would defeat that: all the mass permitted by S2’s orbit packed into a disc inside S301’s, with the orbit lying a few degrees from the disc. There the Newtonian tilt might exceed frame dragging by up to an order of magnitude. For most orientations and less extreme mass distributions the paper expects it one or two orders of magnitude smaller.

The limits of the present dataset

On its own, the data in hand test relativity less sharply than S2 already does. The fitted strength of the first-order relativistic precession comes out at 0.94 with an uncertainty of 0.88, on the convention where 1 is general relativity and 0 is no relativistic precession at all. That is consistent with Einstein and far short of a detection.

The manuscript is also more careful than some of the phrasing around it. It never calls S301 the fastest star in the galaxy; what it claims is the shortest known period, the smallest closest approach and the most relativistic stellar orbit known. Its text gives that closest approach only in Schwarzschild radii, and never converts it into astronomical units or comparisons to planetary orbits.

The binary that was probably pulled in half

An eccentricity of 0.98 is difficult to arrive at gradually. Star formation that close to a supermassive black hole is unlikely, and a star of this mass lives about as long as it would take ordinary two-body encounters to drift it inward. The authors therefore argue that S301 was delivered from somewhere else.

Their preferred mechanism is the Hills process. A tight binary wanders too close, the black hole’s tidal field separates the pair, one star is flung outward as a hypervelocity star and the other is captured onto a long, thin orbit. Running the mapping backwards, the progenitor binary would have been separated by something like 0.05 to 0.2 astronomical units, with an orbital period of five to twenty days. Simulations of binary disruptions in the Galactic Centre, published in 2025 by a team overlapping heavily with this one, put the eccentricities of captured stars of 1.3 to 1.7 solar masses on orbits this size between 0.97 and 0.996 from the fifth to the ninety-fifth percentile, with a median of 0.985.

The authors leave the alternative open. If S301 has been in the Galactic Centre for more than a few tens of millions of years, its eccentricity would have thermalised and lost any memory of how it arrived. A thermal distribution throws up an orbit this eccentric roughly 3 percent of the time. A capture would leave a testable fingerprint, provided the binary had tidally synchronised before it was torn apart. The star should then be turning at 20 to 70 kilometres a second at its equator, which a sharp enough spectrograph could eventually test through the part of that rotation pointing along our line of sight.

Stars on orbits like this should not be unique. The paper gives two steady-state estimates a few sentences apart, of order a few such stars and about a hundred, and expects most of the larger number to be solar-mass stars too faint to detect.

Watching stars fall around Sagittarius A* has already pinned its mass down to better than a percent, because mass is what a black hole does to everything near it, and almost any star will report on it. Spin is what it does only to whatever comes very close. That is one reason a question about the shape of spacetime around the nearest supermassive black hole has waited this long on the arrival of one faint star. The instruments are the other.