The Antikythera Mechanism
The next mechanical device of comparable complexity appears 1,500 years later, in medieval cathedral clocks. A hand-cranked bronze calculator built in Greece around 150–100 BCE, recovered from a Roman shipwreck in 1901, predicted eclipses, tracked two calendars at once, and modeled the Moon's varying orbital speed using a pin-and-slot gear that mechanically implements Hipparchus's geometry — centuries before anyone had a physical theory of why the Moon moves that way.
It sat on the seabed off the island of Antikythera at roughly 60 metres depth until sponge divers brought up 82 corroded fragments. Seven of those are mechanical. The rest is reconstruction, dispute, and a 124-year argument about what the ancient world actually knew how to build.
Key facts
- Recovered: 1901, Roman shipwreck off Antikythera, Greece (~60 m depth)
- Construction date: 150–100 BCE (best estimate); one study argues 205 BCE
- Size: ~34 × 18 × 9 cm, bronze gears in a wooden case
- Surviving gears: at least 30; full reconstructions propose 37–72
- Now held at: National Archaeological Museum, Athens
- Confidence: established that it exists and computes astronomical cycles; emerging on the exact gear count and the planetary trains
How it worked
Turn the crank. Gears with carefully chosen tooth ratios advance dials on the front and back faces.
The front shows solar and lunar position against the Greek zodiac, plus the Egyptian calendar date. The back carries two spiral dials. The upper is the Metonic spiral — 235 lunar months wound over 5 turns, equal to 19 solar years, the cycle that lets a lunar calendar stay in sync with the seasons. The lower is the Saros spiral — 223 months over 4 turns, which predicts the timing and type (solar or lunar) of eclipses.
The clever piece is the lunar mechanism. The Moon does not orbit at constant speed; it accelerates near perigee and slows near apogee. The builder modeled this with a pin-and-slot coupling between two gears on slightly offset axes. The result: a mechanical analog of Hipparchus's eccentric-circle astronomy, built into bronze. The device knows the Moon's variable speed without knowing gravity exists.
Who built it
Unknown. The technical vocabulary on the inscriptions traces to Rhodes, a 2nd-century-BCE centre of mathematical astronomy. The eclipse cycle data matches Hipparchus (~190–120 BCE). One of the secondary dials uses a Corinthian colonial month-name set that points to Syracuse — Archimedes's city. Cicero, writing in the 1st century BCE, describes seeing a planetary "sphere" attributed to Archimedes that sounds uncomfortably close to a relative of this device.
Most scholars favour a workshop tradition over a lone genius. The mechanism is too well-considered to be a first attempt; nothing similar survives, but a single artifact this refined implies several that did not.
The 2024 calendar resolution
A June 2024 paper out of Glasgow applied statistical methods developed for gravitational wave astronomy to the spacing of small holes around the Metonic dial. The question: did the ring have 354 holes (Greek lunar year) or 365 (Egyptian solar)? The Bayesian analysis came down firmly on 354. The calendar is lunar.
The method itself is the joke worth keeping: tools built to detect ripples in spacetime, redirected to detect ripples in 2,000-year-old hole spacings.
The 2025 jamming problem
A preprint by Esteban Szigety and Gustavo Arenas (arXiv:2504.00327, April 2025) simulated the surviving gear train and got an unsettling result. The teeth are triangular, not the involute curves modern machinists use, so meshing pairs accelerate and decelerate as each tooth engages. Combined with the off-centre axes and irregular tooth spacing visible in the fragments, the simulation predicted the device would seize after roughly 120 days of continuous operation — about a third of a year.
There is a real counter. Two thousand years of seawater converted the bronze into atacamite, a copper-chloride mineral that shrank and cracked when the fragments were dried. The geometry measured today is not the geometry that left the workshop. Whether the jamming is original design failure or post-mortem corrosion is, at this writing, unsettled.
What's contested
- Did it run? The 2025 simulation says no, for long. The corrosion argument says we cannot tell. No one has built a precision replica in workshop-fresh bronze and turned the crank for a year.
- Did it model the planets? Several reconstructions add gear trains for Mercury through Saturn based on inscription evidence and missing fragments. The physical gears for these are not in hand. Confidence is emerging, not established.
- One device or many? Cicero's account, the Archimedes connection, and the sophistication of the surviving piece all hint at a lineage. The archaeological record shows exactly one.
- The inscriptions — engraved instructions covering the cases, partially read — are still being analyzed at the Max Planck Institute for the History of Science through 2026. The user manual is incomplete.
Why this has to do with other realms
The mechanism is a hardware encoding of an abstract relationship. So is a tech jacquard loom punched card (1804), and so is a concept neuromorphic computing chip that represents a neural network as physical voltages rather than digital symbols. The Antikythera predates the loom by 1,900 years and the neuromorphic chip by 2,100, yet it sits in the same family: the medium is the mathematics. The device does not simulate Hipparchus's lunar model — it physically is the model, turned through its states by hand.
There is a quieter parallel with concept polynesian wayfinding: at roughly the same historical moment, Pacific navigators were tracking star positions, swell patterns, and bird flight to find islands across thousands of kilometres of open ocean. One civilization encoded its astronomy in bronze; the other in trained memory and chant. Both worked. Only one survives in a museum case.
An open question
If a workshop in Rhodes could machine differential gears in 100 BCE, what else did it make — and why does the archaeological record contain exactly one survivor? The answer is probably about the survival statistics of bronze (melted down for centuries) and shipwrecks (the only environment that preserves them), not about the rarity of the technology. Which means the question is really: how much classical engineering have we permanently lost?
Key sources
- Freeth et al., Nature (2006), "Decoding the ancient Greek astronomical calculator known as the Antikythera Mechanism" — the modern CT-scan reconstruction that opened the field.
- Freeth et al., Scientific Reports (2021), "A Model of the Cosmos in the ancient Greek Antikythera Mechanism" — proposes the planetary gearing on the front face.
- Woan & Bayley, University of Glasgow (June 2024) — Bayesian analysis of the calendar ring hole count. To verify: exact journal placement.
- Szigety & Arenas, arXiv:2504.00327 (April 2025) — gear-jamming simulation and the 120-day failure estimate.
- Cicero, De Re Publica I.21–22 — the ancient eyewitness account of an Archimedean planetary sphere.
Further reading
- Decoding the Heavens by Jo Marchant (2008) — narrative history of the mechanism's recovery and the decades of disputes that followed, written for non-specialists.
- The Antikythera Mechanism Research Project site (antikythera-mechanism.gr) — primary-source imagery, CT scans, and reconstruction videos.
- Michael Wright's working brass replicas — the most committed attempt to build the thing and turn the crank; videos exist on YouTube and demonstrate what "would it work?" looks like in metal.
- The Exact Sciences in Antiquity by Otto Neugebauer (1957) — the classical reference for Babylonian and Greek astronomical mathematics that the device implements.
See Also
- tech jacquard loom — programmable machine encoded in punched cards, 1,900 years later
- concept neuromorphic computing — modern return to "the medium is the mathematics"
- concept voynich manuscript — another artifact whose information survives but whose code is lost
- concept polynesian wayfinding — parallel ancient astronomy, encoded in memory instead of bronze
- event bronze age collapse — the long context for how Mediterranean technical knowledge was preserved and lost