How the First Photo of a Black Hole Was Actually Made

The first picture of a black hole was not taken with a camera, and no single telescope ever saw it. It is a reconstruction, assembled from petabytes of radio data recorded at seven observatories scattered around the globe, then flown across the planet on hard drives. Even the famous orange glow is a decision: the color was assigned by scientists to show brightness, because the radiation itself is invisible to your eyes.

On April 10, 2019, the Event Horizon Telescope collaboration announced the image in six simultaneous press conferences around the world, backed by six research papers. What flashed across every screen was a fuzzy orange ring wrapped around a dark center, the shadow of a supermassive black hole named M87*, sitting at the heart of the galaxy Messier 87 about 55 million light-years away. It looked simple. Getting there was one of the most complicated feats of data collection in the history of science.

A Telescope the Size of the Planet

No dish on Earth is large enough to see a black hole. The target is absurdly small in the sky. M87* spans roughly 40 microarcseconds, which is the angular size of an orange sitting on the surface of the Moon as viewed from your backyard. To resolve something that fine, you would need a telescope thousands of miles across.

So the collaboration built one, sort of. The technique is called very long baseline interferometry, or VLBI, and the idea is to point many separate radio dishes at the same object at the same instant and later combine their recordings as if they were fragments of one enormous dish. The size of that virtual dish equals the distance between the telescopes. Spread your stations across the planet and you get an aperture nearly the diameter of Earth.

In April 2017, the array consisted of eight telescopes at six geographic sites: ALMA and APEX in the Chilean Andes, the IRAM 30-meter dish in Spain, the James Clerk Maxwell Telescope and the Submillimeter Array in Hawaii, the Large Millimeter Telescope in Mexico, the Submillimeter Telescope in Arizona, and the South Pole Telescope in Antarctica. Over four nights, on April 5, 6, 10, and 11, seven of them observed M87* (the South Pole Telescope cannot see a target that far north, so it sat this one out) at a wavelength of 1.3 millimeters, a frequency of 230 gigahertz, which sits in the radio and microwave part of the spectrum, far outside anything the human eye can register.

Pulling this off took the better part of two decades and a slow campaign to enlist each dish. Several of these telescopes were built for entirely different science and had to be fitted with new receivers and recording hardware to join the network. The array only works when the same target rides above the horizon at enough sites at once and the weather cooperates at all of them, which is why the whole effort collapses into a short window each spring and why one cloudy night can knock out a station.

Because the dishes never physically touch, timing is everything. Each station carried a hydrogen maser, an atomic clock stable enough to keep the recordings synchronized to a fraction of a billionth of a second. Every telescope stamped its incoming radio waves with that precise time, wrote them to disk, and waited. The image did not exist yet at any of them. It existed only in the relationship between their separate streams of numbers.

Half a Ton of Hard Drives

Recording the raw radio signal at that bandwidth generates a staggering amount of data. Across the April 2017 campaign, the EHT captured roughly 3.5 petabytes (3,500 TB) of raw data. Stacked up, the drives weighed more than half a ton.

You cannot email a dataset that size. The internet is far too slow, so the collaboration shipped the physical disks by air freight to two supercomputing centers, the MIT Haystack Observatory in Massachusetts and the Max Planck Institute for Radio Astronomy in Bonn, Germany. Dan Marrone, an astronomer at the University of Arizona, put it plainly at the announcement: there is no internet connection that can compete with petabytes of data loaded onto an airplane.

One station could not ship at all for months. The South Pole Telescope was locked in by the Antarctic winter, and the drives could not leave the continent until the weather warmed enough for planes to fly in and out, which pushed the last of the South Pole data to late 2017, data that fed the later Milky Way portrait, since the South Pole cannot see M87 at all. Once the disks for a given target were in hand, the correlators could begin the real work: lining up the seven time-stamped streams, accounting for the fact that a radio wave hits Chile a few milliseconds before it hits Hawaii, and cross-multiplying them into the raw interference measurements that a black hole image is built from.

The Image Is a Reconstruction, Not a Snapshot

Here is where the word photograph starts to fall apart. An Earth-sized dish stitched from eight points is mostly holes. You have a handful of telescopes, not a solid mirror, so you sample only a sparse scattering of the information a real planet-wide dish would collect. That sparse data does not point to one single picture. An infinite number of different images are all mathematically consistent with what the array measured.

Choosing among those possibilities is an act of judgment, guided by algorithms that fill the gaps with reasonable assumptions, such as favoring smoother images over ones speckled with wild noise. And that is precisely the danger. Feed the process biased assumptions and you can conjure a ring that was never really there. The collaboration guarded against fooling itself by splitting into four independent imaging teams, isolated from one another, each using different software and methods. They even tested their pipelines on fake data, including images with no ring at all, to make sure they were not simply drawing the shape they hoped to find. When the four teams finally compared results, they had each independently produced the same asymmetric orange ring. That agreement is what turned a plausible reconstruction into a believable measurement.

The final reconstructions leaned on two independent families of methods, the traditional CLEAN approach run in DIFMAP and a newer regularized maximum likelihood technique built into the eht-imaging and SMILI software packages. Among the researchers who developed those newer imaging tools was Katie Bouman, who had built an earlier black-hole imaging algorithm called CHIRP as a graduate student at MIT and was a postdoctoral researcher by the time the image was released. In the days after the release, headlines framed her as the single person who made the image, and a photo of her reacting to the first reconstruction went viral. Bouman spent the following weeks correcting the record. "No one algorithm or person made this image," she wrote, giving credit instead to a worldwide team of scientists and to years of labor on the hardware, the calibration, and the analysis. That is the honest version. More than 200 researchers built the result, and Bouman's real contribution, developing imaging techniques and helping design the blind framework that kept everyone honest, is impressive without the myth. She is now a professor at Caltech.

Why the Ring Is Orange

The color you remember is not a color anyone could ever see. The EHT observed at 1.3 millimeters, deep in the radio band, so there is no literal hue to reproduce. The orange and yellow are a brightness map, a scale the team applied to show how much radio energy arrived from each patch of sky. Yellow marks the most intense emission, red is dimmer, and black is little or nothing. Swap the palette for blue or gray and the physics would be identical.

What the ring actually traces is light bending around a region of gravity so extreme that nothing escapes it. The dark patch in the middle is the black hole's shadow, and it is larger than the event horizon itself, because the black hole's gravity bends the light of the hot plasma surrounding it, including emission from the far side, around and into your view, inflating the silhouette to about 2.5 times the true horizon. From the size of that ring, the team measured the mass of M87* at 6.5 billion times the mass of the Sun, give or take a few hundred million. The ring glows brighter along its lower edge because of relativistic beaming: the gas on that side is whipping around the hole toward Earth at a large fraction of the speed of light, which boosts and focuses its radiation in our direction, while the side racing away is dimmed.

The Sequel at the Center of Our Galaxy

Three years later, on May 12, 2022, the same collaboration released a second portrait, and this one is closer to home. It shows Sagittarius A*, the supermassive black hole anchoring our own Milky Way, about 27,000 light-years away and weighing roughly 4 million solar masses. It came from the very same April 2017 observing run.

You might expect the nearer black hole to be the easy one. It was the opposite. M87* is so enormous that gas takes days to circle it, so it holds still like a patient portrait subject across a night of observing. Sagittarius A* is more than a thousand times smaller, and its surrounding gas laps it in minutes, meaning the source flickers and reshapes itself while the telescopes are still recording. Imaging it was like photographing a running dog with a long exposure, and the team had to average across thousands of possible frames to recover the ring, which measured about 51.8 microarcseconds across. That it produced the same fundamental shape, a bright ring around a dark shadow, was another powerful test of general relativity in the strong-gravity regime, matching what Einstein's equations predicted a century ago.

So the next time that orange ring scrolls past you, remember what you are actually looking at. Not a click of a shutter, but 3.5 petabytes of raw radio measurements, signal buried in noise, seven clocks ticking in lockstep across the planet, a fleet of hard drives on cargo planes, and years of argument among 200 people over which of infinitely many images the universe was trying to hand them. The picture is real. It just had to be built.

Lead image — The first image of a black hole, the supermassive black hole M87* in galaxy Messier 87, released April 10, 2019. EHT Collaboration, CC BY 4.0. Source.