The Repository Absolute Magnitude Space

Sagittarius A* Moved While the Telescope Was Looking

September 9, 2026 · Runtime 12:34 · Watch on YouTube

The same eight telescopes, the same five nights of April 2017, two black holes. One picture took two years. The other took five, and the delay was in the object rather than in the people.

Chapters

Transcript

Two black holes, five nights, three years apart 0:00

In April two thousand seventeen, eight radio telescopes spent five nights pointed at two black holes. The first picture came out two years later, and you have seen it. The second picture took five years.

The delay was in the object 0:15

Nobody was being slow. The delay was in the object, not in the people. This episode ends on the one number that array returned for the black hole at the centre of our own galaxy, and on why it was so much harder to get than the famous one.

Last episode took apart the other image. An angle of forty two millionths of a second of arc, with two conversions stacked on top of it. Tonight: the same array, the same nights, and a problem that has nothing to do with distance.

Sagittarius A*, 27,000 light years away 0:48

The target is Sagittarius A star, the compact radio source at the centre of the Milky Way. It is about twenty seven thousand light years away, which by the standards of this subject is next door.

Less mass, much closer -- the two nearly cancel 1:01

Being next door has a consequence worth holding on to. Sagittarius A star has around fifteen hundred times less mass than the one in M eighty seven, but it sits about two thousand times closer. Those two nearly cancel, so on the sky the two black holes come out almost the same size.

5, 6, 7, 10 and 11 April 2017 1:21

So they were observed together. Sagittarius A star on the fifth, sixth, seventh, tenth and eleventh of April. M eighty seven on four of those same nights. Same dishes, same recorders, same hydrogen masers keeping time.

2019 for one, May 2022 for the other 1:40

One of those two data sets became a famous picture in April two thousand nineteen. The other was published on the twelfth of May, two thousand twenty two. Three years apart, from tapes recorded in the same week.

One pair of dishes measures one ripple 1:55

To see why, go back to what the array actually does. It never records a picture. Each PAIR of telescopes measures one ripple in the brightness of the sky, and which ripple it measures depends on how far apart that pair looks from the source.

Eight dishes make twenty eight pairs. Twenty eight ripples is not an image of anything. It is a handful of numbers where a picture needs thousands.

So you wait, and let the planet work 2:23

So you wait, and you let the planet do the work. As the Earth turns, every pair swings around and changes its length and its angle as seen from the source, and each one sweeps out an arc of new measurements it did not have before.

Earth rotation aperture synthesis 2:38

That is Earth rotation aperture synthesis, and it is the whole trick. You do not build the mirror. You let the planet drag your eight fragments through the places the mirror would have been, and you collect what they find along the way.

The price: the source must hold still 2:53

There is a price for that, and it is written into the method rather than into the hardware. To add up eight hours of measurements as though they all belonged to one mirror, the thing you are looking at has to hold still for eight hours.

M87* holds still, and mass is the reason 3:07

M eighty seven star holds still, and the reason is only its mass. It is about one thousand six hundred times heavier than the black hole at our own centre, and for a black hole, mass is the clock.

GM over c cubed: nine hours against twenty seconds 3:21

Every black hole has one natural unit of time: its mass, times big G, divided by the speed of light cubed. Roughly, the time light needs to cross it. For M eighty seven star that comes to about nine hours. For Sagittarius A star, twenty seconds.

Turn that into something you can picture. The tightest orbit a lump of gas can hold before it falls in takes about thirty days around M eighty seven star. Around Sagittarius A star, the same orbit takes about thirty minutes.

Eight hours is sixteen orbits 3:58

Now put one night of observing next to those two numbers. Eight hours is a rounding error for M eighty seven star. Nothing you can see moves. For Sagittarius A star, eight hours is more than a dozen complete orbits.

The picture moved while they were taking it 4:15

Which is the whole episode in one sentence. The picture moved while they were taking it. Not between nights. Inside a single track.

One hundred per cent, on some baselines 4:27

This is not a worry somebody raised in a meeting. It is in the data. On some pairs of telescopes the measured brightness changed by around one hundred per cent inside a single night.

And it survived the obvious excuses. It is larger than the measurement errors, and larger than the blurring our own galaxy's ionised gas smears across the source. Something at the black hole was genuinely changing.

Twenty-eight pairs, twenty-eight different skies 4:55

Here is what that breaks. If the source varies, then each of those twenty eight pairs sampled a slightly different sky. Hand that to software that assumes one fixed sky and it will still return an image. It just will not be an image of any single moment.

Reconstruct a movie, or widen the errors 5:13

There were two honest ways out. Reconstruct a movie, which is what the data would really prefer, or keep one static image and widen the uncertainties until they cover how much the source could have moved underneath.

The variability noise budget 5:27

The collaboration published the second, and gave it a name that says exactly what it is. A variability noise budget. An extra uncertainty added to every measurement before any image is made.

The size of that extra uncertainty is not a guess. It was calibrated on simulations of a black hole that does change, so a static reconstruction cannot mistake the movement of the source for structure in it.

DIFMAP, eht-imaging, SMILI, THEMIS 5:56

Then the same discipline that produced the first image. Four independent pipelines: DIFMAP, eht imaging, SMILI and THEMIS. One classical deconvolution, two regularised likelihood methods, and one full Bayesian sampler.

Each of them swept thousands of combinations of its own settings, and kept every image that fit the data within the noise. The picture that was published is an average over thousands of images, not the output of one run of one program.

Four families of shape, one not a ring 6:29

And when those surviving images were sorted by shape, they did not all agree. They fell into four families: three of them rings that differ in where the bright side sits, and one smaller group that are not rings at all.

That fourth group is worth noticing, because it was published rather than dropped. It is the honest way to say that the data prefer a ring without proving one.

51.8 plus or minus 2.3 microarcseconds 6:54

Across all of that, one number comes through. The bright ring has a diameter of fifty one point eight millionths of a second of arc, plus or minus two point three.

That is the measurement. An angle, and its error bar. Everything else you have read about this object either sits downstream of that number, or was already known before the telescopes were switched on.

A doughnut on the surface of the Moon 7:20

For scale, the collaboration's own comparison. That angle is roughly what a doughnut on the surface of the Moon would cover, seen from a back garden on Earth.

Why five years and not two 7:31

And that answers the question this started with. The five years were not spent staring at the tapes. None of the machinery that made this image possible existed when the first one was published: the variability budget, the simulations it had to be calibrated against, and the proof that a static reconstruction can survive a source that will not sit still.

The order matters, too. M eighty seven came first because it was the easy one, and the easy one is where you find out whether your method works at all.

Here the chain runs the other way 8:04

Now the conversion, and this is where the episode turns away from the last one. For M eighty seven star, the ring is how you get the mass. Here, the mass was already known, and known far better than the ring.

S2, a sixteen-year orbit, tracked since 1992 8:20

Since nineteen ninety two, telescopes have been following individual stars looping around the galactic centre. One of them, called S two, takes sixteen years to go round, and has now been tracked through roughly two and a half complete orbits.

4.3 million suns, to a quarter of a per cent 8:36

From four of those orbits fitted at once, the GRAVITY instrument on the Very Large Telescope Interferometer gives a mass of four point three million suns, to about a quarter of one per cent, and a distance of eight thousand two hundred and seventy seven parsecs.

So the ring is a test, not a measurement 8:52

So the ring is not measuring the mass here. It is being tested against it. General relativity says a hole of that mass, at that distance, must cast a shadow of one particular angular size. There is no free knob to turn.

Theta-g, and two root twenty-seven 9:10

The gravitational angle is big G times the mass, divided by c squared and by the distance. For those values it comes to five point one two five millionths of a second of arc. The shadow of a non spinning hole is two times the square root of twenty seven of those.

Which is fifty three point three microarcseconds. That is the prediction, and it was fixed before anyone looked.

Measured: 48.7 microarcseconds 9:37

The measurement, converted from the ring into a shadow, is forty eight point seven microarcseconds. Against a prediction of fifty three point three, that is about eight per cent low, which is the deviation the collaboration published.

Minus 0.08, uncertainty 0.09 9:53

And it passes, because the uncertainty on that eight per cent is nine. General relativity survives a test at a black hole four million times the mass of the Sun, using an image of something that refused to hold still.

0.1, 0.25, 4.4 -- and 14 per cent 10:08

But look at where the slack in that test actually is. The mass is known to a quarter of a per cent. The distance to about a tenth. The ring the telescopes measured, to four point four per cent. And the shadow that gets compared with theory, to fourteen.

The uncertainty triples in one step 10:27

Fourteen, out of four point four. The uncertainty roughly triples in one step, and that step is not an observation. It is the factor that turns a bright ring into a shadow, and it comes out of simulations.

It has two parts, and both are ranges rather than numbers. Between one and one point two for how much bigger the bright ring sits than the shadow itself, and up to one point three more, depending on which imaging method drew the ring.

The same lesson, from the other side 10:58

Last episode ended by saying the precision lost between the angle and the headline did not come from the telescope. This one lands on the same lesson from the other side. The stars handed over a mass to a quarter of a per cent, and by the time it can be checked against a picture, the check is only good to fourteen.

None of that weakens the result. Every one of those factors is published as a range, in the collaboration's own papers, and the variability was declared rather than quietly averaged away. The strength of a result like this is that you can take it apart at all.

Ask which link is the simulation 11:36

So the practical version, again. When a headline says a picture confirms a theory, ask how precise the confirmation is, and which link in the chain set that number. Usually one link is an instrument and one is a simulation, and it is the simulation that is wide.

Next: the orbit that weighed a black hole 11:55

Fifty one point eight millionths of a second of arc. A ring around a black hole whose innermost orbit closes in about half an hour, measured by an array that needed the Earth to turn for eight hours to see it at all. And the number it was tested against, four point three million suns to a quarter of a per cent, came from following one star around a sixteen year orbit, twice over, since nineteen ninety two. Next time: the orbit that weighed a black hole, and what a quarter of a per cent still hides. Absolute Magnitude Space. Every episode ends in a measured number.

Description and sources

Very long baseline interferometry works by letting the Earth turn: eight dishes give twenty-eight pairs, and eight hours of rotation sweeps those pairs across the coverage a far larger mirror would have had. The method assumes the source holds still while that happens. M87* does. Sagittarius A* is about 1,600 times less massive, so its gravitational timescale is around twenty seconds and its innermost orbit closes in about half an hour -- sixteen orbits inside a single track. The published answer was a variability noise budget, four independent imaging pipelines, and an average over thousands of images. What survives is one angle: a ring 51.8 +/- 2.3 microarcseconds across. And the chain runs backwards from last episode -- the mass was already known to about a quarter of a per cent from stellar orbits, so the ring is not a measurement but a test. It passes. But the ring is known to 4.4 per cent and the shadow it is converted into to 14, and that step is a factor out of simulations.

PRINT-READY, FROM THIS CHANNEL

The Conversion Chain, Vol. 01 — the method, and four numbers taken apart with it

https://therepository.gumroad.com/l/the-conversion-chain

PRIMARY SOURCES

EHT Collaboration (2022). First Sgr A* EHT Results. I. ApJL 930, L12. arXiv:2311.08680 -- the 51.8 +/- 2.3 microarcsecond ring (68% credible interval).
EHT Collaboration (2022). ... III. Imaging. ApJL 930, L14. arXiv:2311.09479 -- DIFMAP, eht-imaging, SMILI, THEMIS; the variability noise budget; four morphologies, one not a ring.
EHT Collaboration (2022). ... IV. Variability, Morphology, Mass. ApJL 930, L15. arXiv:2311.08697 -- GM/c^3 about 20 s; ISCO ~30 min against ~30 days; ~100% on some baselines.
EHT Collaboration (2022). ... VI. Testing the Black Hole Metric. ApJL 930, L17. arXiv:2311.09484 -- shadow 48.7 +/- 7 uas; deviation -0.08 +/- 0.09; alpha1 = 1.0-1.2, alpha2 up to 1.3.
GRAVITY Collaboration (2022). Mass distribution in the Galactic Centre. A&A 657, L12. arXiv:2112.07478 -- 4.30e6 solar masses to ~0.25%; R0 = 8,277 pc; from S2, S29, S38 and S55.
EHT Collaboration (2022). ... II. Observations and Calibration. ApJL 930, L13. arXiv:2311.08679 -- Sgr A* observed on 2017 April 5, 6, 7, 10 and 11.
EHT Collaboration (2019). First M87 EHT Results. I. ApJL 875, L1. arXiv:1906.11238 -- the 42 +/- 3 uas ring, published 10 April 2019.
Arithmetic redone here from the published mass and distance: theta_g = GM/c^2D = 5.125 uas; shadow = 2*sqrt(27)*theta_g = 53.3 uas; 48.7/53.3 - 1 = -0.086, reproducing the published -0.08.

ARCHIVE PLATES: all 15 are credited on screen in the video and listed in this project's manifest.

Absolute Magnitude takes one number the internet repeats and asks what an instrument actually returned. Every episode ends on a measured figure with its source on screen, and says plainly which part of it was converted afterwards.

No AI-generated imagery is used on this channel. Every frame of archive footage is real, publicly released material, credited on screen with its licence; every diagram is drawn in code from the numbers in the cited papers.

Corrections are pinned, not quietly edited.

#SagittariusA #BlackHole #EventHorizonTelescope #Astronomy