The Event Horizon Telescope

An Earth-Sized Telescope

How fine a detail a telescope can resolve depends on its wavelength divided by its width. The Event Horizon Telescope observes at a wavelength of 1.3 mm. In 2017, seen from the direction of M87*, the two telescopes furthest apart in the array were about 10,700 km apart.

1.3 mm divided by 10,700 km is about 25 microarcseconds. The papers give an effective resolution in the images of about 20 µas; the official comparison is reading a newspaper in New York from a café in Paris. The ring round M87* is 42 µas across. Why there is a ring at all is told in the plate “Where Not Even Light Escapes”.

1.3 mm ÷ 10,700 km ≈ 25 µas · images about 20 µas · ring 42 µas

Two Dishes, One Set of Stripes

The same wavefront reaches two telescopes at different moments. Line up the two recordings, multiply them and average, and out comes one complex number: how well a pattern of light and dark stripes on the sky matches the brightness round the black hole, and how far the stripes must shift to match it best.

The stripes’ spacing is the wavelength divided by the distance between the two telescopes, as seen from the source; they run at right angles to the line between them. One pair of telescopes measures one set of stripes, not a picture. Mark each set as a point, further from the centre for finer stripes: this map is called the u-v plane.

Stripe spacing = wavelength ÷ projected baseline · one pair = one point on the u-v plane (and its mirror)

The Earth Turns

The Earth turns. Seen from M87*, the line between two telescopes turns with it and changes length, so a pair draws an arc across the u-v plane. On the night of 11 April 2017, the seven telescopes that could see M87*, taken in pairs, drew these arcs. The telescope at the South Pole cannot see M87*.

The more arcs, the more of the stripes are measured, and the closer the rebuilt picture comes to the sky. The picture here is computed in the browser: a ring of known size stands in for the truth on the sky; only the cells the arcs pass through are kept, and a picture is rebuilt from them.

Seven telescopes see M87* · 21 pairs (two of them at one site, too short) · this page’s ring model, not real data

Filling the Gaps

A whole night still measures only a few thin arcs across the u-v plane, and most of it stays empty. Turned straight back into a picture, the measured part gives one full of stripy artefacts (left).

Countless pictures fit the data. An imaging algorithm picks the most reasonable one: no brightness below zero, no needless flicker between neighbouring spots, as plain as possible where nothing was measured. This page uses a small version of such a “regularised maximum likelihood” method (right).

Left: the measured cells transformed straight back · right: rebuilt by regularised maximum likelihood

Cut Open: One Station

This is APEX on the Chajnantor plateau in Chile, about 5,100 m up, one of the EHT’s eight telescopes in 2017. It was made from an ALMA prototype antenna; its main reflector is 12 m across with a focal length of 4.8 m, and the whole antenna weighs about 125 tonnes.

Radio waves strike the main reflector and head for its focus; on the way, a 0.75 m subreflector folds them back through a hole in the middle of the dish into the Cassegrain cabin, and mirrors send them along the elevation axis into a Nasmyth cabin at the side. The 1.3 mm receiver used for the EHT sits there.

Main reflector 12 m · focal length 4.8 m · subreflector 0.75 m · two Nasmyth cabins · about 125 tonnes

264 Panels

The main reflector is made of 264 aluminium panels in 8 rings, on a carbon-fibre backup structure of 24 sandwich segments. Adjusters behind each panel push it to its place on the paraboloid.

In 2017 the surface was accurate to about 17 µm (rms). By Ruze’s formula, that leaves 1.3 mm waves about 97 % of their efficiency; a 100 µm error would leave about 40 %. Early in 2018 the panels and subreflector were replaced, bringing the error to about 10 µm.

264 panels · 8 rings · 17 µm (2017) → 10 µm (2018) · efficiency = e^−(4πσ/λ)²

Bringing 230 GHz Down

A 1.3 mm wave swings about 230 billion times a second, far too fast to record. The receiver mixes it with a local signal at 221.1 GHz, which brings 226.1 to 230.1 GHz down to 5 to 9 GHz. The mixers are superconducting (SIS) and must be kept at a few kelvin.

APEX’s receiver is PI230, built by the Max Planck Institute for Radio Astronomy and installed in a Nasmyth cabin in 2015. The lowered signal is digitised at 2 bits: two bands 2 GHz wide, two polarisations, about 32 billion bits a second in all.

Local signal 221.1 GHz · 226.1–230.1 GHz → 5–9 GHz · 2 bits · 32 Gbit/s

A Clock at Every Station

No cable joins the telescopes. Every station has a hydrogen maser, an atomic clock that sets the receiver’s local signal and the time stamps on the data. APEX’s maser was installed in 2010, in a cabinet held to ±0.1 °C beside a container; its signal runs to the antenna on cables 85 m long.

The goal is a frequency that drifts by no more than 1.5 parts in 10¹⁴ over 10 seconds. At that, even at 0.87 mm, the signal keeps nearly all its coherence through an integration. GPS lines the stations’ clocks up to within tens of nanoseconds; what is left is solved for at the correlator.

σ(10 s) = 1.5 × 10⁻¹⁴ · GPS timing to tens of nanoseconds

Data by Air

Each station records 32 billion bits a second. In April 2017 the array recorded about 350 terabytes a day, about 3.5 petabytes of raw data in all (the EHT’s figure; its scientists have also written about 4 PB). Sending that over the internet would take too long, so it went onto helium-filled hard disks, flown to the correlators.

One recorder writes 16 billion bits a second, sliced in time across 32 disks in four removable modules. APEX’s recorders sit in a control container about 50 m from the antenna. With no flights to the South Pole from February to October, the disks from there reached MIT’s Haystack Observatory only on 13 December 2017.

32 Gbit/s · about 350 TB a day · about 3.5 PB · South Pole disks arrived 13 December

Line Up, Multiply

A wavefront reaches the stations at different times, up to tens of milliseconds apart. The correlator lines the recordings up with a model of the Earth’s geometry and each station’s clock, multiplies them in pairs and averages, giving one complex number every 0.4 s. In 2017 the work was split: MIT Haystack took the band at 227.1 GHz and the Max Planck Institute for Radio Astronomy in Bonn the band at 229.1 GHz, each on more than a thousand processor cores running the DiFX software.

The air above each station makes the phases jump. But add up the three phases between three telescopes, round the triangle, and each station’s error comes in once with a plus and once with a minus and cancels. That sum, the closure phase, depends only on the sky.

Closure phase = φ₁₂ + φ₂₃ + φ₃₁ · station errors cancel

Four Teams, Working Blind

Four teams took the same data from 11 April and imaged it for seven weeks without talking to one another: two with CLEAN, two with regularised maximum likelihood. All four got a ring, brighter in the south.

Then three pipelines (DIFMAP, eht-imaging, SMILI) each surveyed a thousand to tens of thousands of combinations of settings, keeping those that recovered known test pictures. Here, this page’s one set of simulated data rebuilt with three settings: the details differ, the ring is in each.

Four teams · seven weeks · three pipelines · about 49,000 parameter sets

Two Black Holes

M87*: shown on 10 April 2019, a ring 42 ± 3 µas across round a black hole of about 6.5 billion solar masses, 55 million light years away. The picture comes from one night, 11 April 2017; the other nights, imaged on their own, agree. In 2021 its polarised image showed the direction of the magnetic field round the ring.

Sagittarius A*, at the centre of our galaxy: shown on 12 May 2022, a ring 51.8 ± 2.3 µas across, about 4 million solar masses, some 27,000 light years away. Gas goes round it in minutes, so the ring changes within a night, which made it much harder to image. The South Pole telescope can see it. Telescopes in Greenland, at Kitt Peak and NOEMA have joined since; in test observations at 0.87 mm the finest stripes reached 19 µas.

M87* 42 µas (2019) · Sgr A* 51.8 µas (2022) · 0.87 mm: 19 µas

Taken Apart

The subreflector on its four legs, the 264 panels, the backup structure, the Cassegrain cabin, the two Nasmyth cabins, the yoke and the pedestal come apart, layer by layer.

Whichever part the pointer rests on is lifted out and the rest fade; left alone, the parts are all named.