Inside the Gears: How the Antikythera Mechanism Actually Works

In Part 1, we covered what the Antikythera mechanism could do: predict eclipses, track planets, model the Moon's variable speed, and compute the dates of the Olympic Games. We met the engineers who built the tradition it emerged from. And we confronted the 1,400-year gap between this device and anything remotely comparable.

Now we go inside the machine itself.

Eighty-two corroded fragments sit in the National Archaeological Museum in Athens. They contain 30 surviving gearwheels, over 3,500 characters of ancient Greek text, and engineering secrets that took three generations of researchers and 120 years to decode. What follows is how those gears actually connect, what the hidden inscriptions say, and why the mathematics behind it all is stranger than the hardware.

Two Faces of the Machine

The mechanism had two faces, front and back, and they served different purposes. The front told you where things are right now. The back told you what's coming.

The front face displayed two large concentric rings. The outer ring was an Egyptian solar calendar: twelve months of 30 days each, plus five extra days at the end, totalling 365. The inner ring was the zodiac, divided into twelve 30-degree segments. This was the celestial coordinate system. Pointers swept across both rings showing the current positions of the Sun, Moon, and all five planets known in antiquity: Mercury, Venus, Mars, Jupiter, and Saturn.

The Moon pointer carried a special feature: a small ball, half silver and half black, that rotated as the pointer moved. It displayed the current lunar phase. The rotation was driven by a differential gear that read the angle between the Sun and Moon pointers. When the Moon was opposite the Sun, the silver half faced outward: full moon. When the Moon was near the Sun, the black half showed: new moon. An analogue display of a continuous function, built in bronze.

The back face held two large spiral dials. The upper spiral had five turns and tracked the Metonic cycle: 235 lunar months compressed into exactly 19 solar years, the period after which the Moon's phases repeat on the same calendar dates. A small subsidiary dial at the top tracked the four-year cycle of the Panhellenic Games. Inscriptions on this dial name specific festivals: Olympia, Pythia, Isthmia, Nemea, Naa, and Halieia. This wasn't abstract astronomy. It was a sports calendar.

The lower spiral had four turns and tracked the Saros cycle: 223 lunar months (roughly 18 years and 11 days), the period after which solar and lunar eclipses repeat in nearly identical patterns. A subsidiary dial below it tracked the Exeligmos: three Saros cycles (54 years), providing a time correction for eclipse predictions across multiple cycles.

Everything was driven by a single main gear, designated B1 by researchers. One full rotation of B1 equalled one tropical solar year. Its coaxial partner B2 had exactly 64 teeth and directly drove the mean Sun pointer. Every other computation in the device, every planetary position, every eclipse prediction, every lunar phase display, derived its fundamental frequency from that single gear.

3,500 Characters Nobody Could Read

The mechanism's bronze plates and protective cover doors were covered in engraved text. Over 3,500 characters of ancient Greek, tiny and precise, packed into every available surface. For decades after the device's recovery, most of this text was invisible, buried under two millennia of corrosion and mineral deposits. Early researchers could make out scattered words but couldn't read the bulk of it.

That changed in 2005. The Antikythera Mechanism Research Project (AMRP) deployed two technologies that cracked the inscriptions wide open. The first was Polynomial Texture Mapping (PTM), which photographs surfaces under dozens of different lighting angles and computationally enhances surface details invisible to the naked eye. The second was an 8-tonne X-Tek Systems "Bladerunner" Microfocus X-ray CT scanner, originally designed to detect hairline cracks in aircraft turbine blades. It operated at 450 kilovolts with sub-0.1mm resolution. For the first time, researchers could read text on internal surfaces of the mechanism that hadn't been seen by human eyes since the device sank.

The inscriptions fall into several categories. The Front Door Inscription is the most extensive. It references the planets Venus (ΑΦΡΟΔΙΤΗ) and Mercury (ΕΡΜΗΣ) and describes their retrograde motion, the apparent backward movement planets make against the background stars. The Greek word ΣΤΗΡΙΓΜΟΣ (stationary point) appears multiple times, referring to the exact moments when a planet appears to stop before reversing direction. This confirmed that the mechanism modelled planetary motion, not just the Sun and Moon.

The Back Cover Inscription functions as a user manual, describing how to operate the dials and interpret the outputs. Eclipse inscriptions ring the Saros dial, with alphabetical index letters linking specific month positions to detailed prediction blocks.

Most remarkable: the CT scans revealed the Greek numerals ΨΞΒ (462) and ΨΜΒ (442) in the Front Door Inscription. These encode the period relations for Venus and Saturn, respectively: 289 synodic cycles of Venus in 462 years, and 427 synodic cycles of Saturn in 442 years. These numbers would prove to be the key that finally unlocked the mechanism's planetary display, but that breakthrough wouldn't come for another 16 years.

One more detail buried in the inscriptions: researchers identified two distinct handwriting styles in the engraved text. This wasn't one genius working alone. It was a workshop. A team.

The Pin-and-Slot Trick

Here is where the mechanism stops being impressive and starts being genuinely strange.

The Moon doesn't travel at constant speed. It moves faster when it's closer to Earth (at perigee) and slower when it's farther away (at apogee). The Greeks called this lunar anomaly. They didn't know the orbit was elliptical; that insight wouldn't arrive until Kepler in the 17th century. But they knew the speed varied, and they built a mechanical system to model it.

The system uses four gears, each with exactly 50 teeth. Run the math on the gear ratio: 50/50 multiplied by 50/50 equals 1. No net speed change. The output turns at exactly the same rate as the input. So what's the point?

The point isn't the ratio. It's the geometry.

The accuracy is startling. Modern analysis shows the mechanism modelled the Moon's angular velocity variations to better than 1 part in 200. The offset distance controlling the speed variation matches double the Moon's actual orbital eccentricity to within 4%. This is a physical implementation of the geometric model proposed by Apollonius of Perga in the 3rd century BCE: an eccentric circle (an orbit shifted off-centre relative to Earth) produces exactly the kind of sinusoidal speed variation that the pin-and-slot mechanism generates.

The Greeks didn't know the orbit was an ellipse. They used circular geometry. But the mechanical result is nearly identical to what Kepler's second law predicts. And it's more accurate than Ptolemy's own written account of Hipparchus's lunar theory, published roughly 300 years after the mechanism was built.

Predicting Eclipses by the Month

The lower back dial is the eclipse computer. Its four-turn spiral covers 223 lunar months, one complete Saros cycle, driven by a gear train with the ratio 940/4237 (derived from the relationship between 19 solar years and 235 synodic months).

At specific month positions on the spiral, small glyphs mark predicted eclipses. Each glyph packs dense information into a tiny space:

• Σ (Sigma, for ΣΕΛΗΝΗ, Moon) marks a lunar eclipse
• Η (Eta, for ΗΛΙΟΣ, Sun) marks a solar eclipse
• A number indicating the hour of the eclipse, so you know when to look up
• A notation for "of the day" or "of the night", telling you whether the eclipse will be visible from your location

Each glyph also carries an alphabetical index letter (A, B, C and so on) that references a separate inscription block engraved around the periphery of the Saros dial. These blocks describe the eclipse in detail: the direction of shadow obscuration (which limb of the Sun or Moon darkens first), the magnitude (how much of the disk is covered), the expected colour of the eclipsed Moon, and the Moon's angular diameter.

The four-turn design wasn't arbitrary. Research published in PLOS ONE in 2014 demonstrated that each quarter-turn of the Saros spiral corresponds to one Full Moon Cycle: the roughly 14-month oscillation in the Moon's apparent diameter. The Full Moon Cycle determines eclipse duration and magnitude. By encoding four FMCs in four turns, the builders gave the eclipse characteristics their quantitative basis. The spiral's shape was the data model.

After one Saros cycle (18 years, 11 days), eclipses repeat but shifted by roughly 8 hours. The subsidiary Exeligmos dial below the main spiral corrects for this. It displays 0, 8, or 16 hours to be added to the times shown in the Saros glyphs, allowing accurate eclipse time prediction across multiple Saros cycles spanning over a century.

The People Who Cracked It

The mechanism sat in that Athens museum for over half a century before anyone seriously attempted to understand its gear trains. Three waves of researchers, each building on the last, each correcting the previous generation's errors, gradually decoded the device across 120 years.

Derek de Solla Price (1959-1974)

Price, a British science historian at Yale, was the first scholar to treat the mechanism as a precision astronomical instrument rather than a curiosity. Working from photographs and early X-rays taken in collaboration with Greek physicists Charalambos and Emily Karakalos, he published his landmark monograph Gears from the Greeks in 1974.

Price got the big picture right. He correctly identified 30 gears across the fragments. He recognized the 19-year Metonic cycle and the 223-month Saros cycle as the primary functions of the back dials. He spotted the 127-tooth gear and understood its significance: 127 is half of 254, the number of sidereal lunar orbits in one Metonic cycle. He established, permanently, that this was a computational device.

But his detailed reconstruction was substantially wrong. He proposed a differential gear turntable that later analysis showed was mechanically implausible. He believed the back dials were concentric rings, not spirals. And he misidentified the upper subsidiary dial as a 76-year Callippic cycle tracker. It was actually the Panhellenic Games dial, confirmed decades later by inscriptions naming Olympia, Pythia, Isthmia, and Nemea.

"Nothing like this instrument is preserved elsewhere. Nothing comparable to it is known from any ancient scientific text or literary allusion." Derek de Solla Price, who spent 15 years proving that sentence true.

Michael Wright (1990s-2002)

Wright, curator of Mechanical Engineering at the Science Museum in London, conducted a second X-ray campaign in the 1990s using linear tomography. His corrections were fundamental. The back dials were spirals, not concentric rings. The pin-and-slot mechanism for lunar anomaly was his identification. And in 2002, he proposed something Price had never attempted: a complete planetary display using epicyclic gear trains with pin-and-slot mechanisms for all five planets.

Wright built a full working reconstruction in bronze and wood. For the first time since antiquity, someone turned a hand crank and watched the Antikythera mechanism's gears move.

The AMRP and UCL (2006-2021)

The 2005 CT scanning campaign produced the data that resolved decades of debate. The Antikythera Mechanism Research Project (AMRP), led by Tony Freeth, Mike Edmunds, and others, published a landmark paper in Nature in 2006 that confirmed the spiral back dials, resolved the gear topology, and decoded the inscription text. A second Nature paper in 2008 added the Olympiad dial discovery and further eclipse analysis.

But the front planetary display remained unsolved. The gear system needed to fit within the mechanism's known physical dimensions, match the surviving evidence (including a mysterious 63-tooth gear in Fragment D), and reproduce the period relations found in the inscriptions.

In March 2021, Freeth's team at University College London published the solution in Scientific Reports. The key was those numbers from the inscriptions: 462 for Venus and 442 for Saturn. These didn't come from Babylonian astronomy. They matched a Greek mathematical method attributed to Parmenides of Elea and reported by Plato: combining known period relations iteratively to derive more accurate ones.

The numbers worked because they were factorizable into gear tooth counts under 100 (the practical maximum for hand-cut bronze gears). Venus's 462 factors into 2 x 3 x 7 x 11. Saturn's 442 factors into 2 x 13 x 17. Shared prime factors across different planetary trains allowed gears to be shared, fitting everything into the mechanism's compact frame. Fragment D's unexplained 63-tooth gear finally found its purpose: a key component of the Venus epicyclic system.

Bronze, Saws, and Steady Hands

Knowing what the mechanism computed is one thing. Understanding how someone actually built it with ancient tools is another.

The gears are bronze with approximately 5% tin content. Chemical analysis in 2018 identified three distinct alloys in different components, suggesting either different manufacturing batches or different workshop sources. The frame plates use a different alloy than the gearwheels themselves.

The teeth are equilateral triangles, not the involute profiles used in modern gears. Each tooth was hand-cut. The average circular pitch (distance between teeth) is 1.6 mm. Average wheel thickness is 1.4 mm, with a 1.2 mm air gap between meshing gears. The smallest gears have 15 teeth. The largest has 223 teeth and spans roughly 13 cm across.

The manufacturing process was almost certainly: cast or cold-forge a thin bronze blank into a circular plate, scribe the tooth divisions using a compass and dividers, saw out the triangular tooth spaces, then file each tooth to refine the form. No evidence of any specialized gear-cutting machine has been found from this period. The lathe was available for creating circular parts and shafts. Beyond that, it was hand tools and steady hands.

The triangular tooth profile introduces periodic speed variation as each tooth engages. But analysis shows that manufacturing inaccuracies (not tooth geometry) are the main source of error. The sub-millimetre tolerances achieved are remarkable. The mechanism likely suffered from pointer wobble that would have obscured the finer correction mechanisms, but the core computations remained sound.

Cicero, writing around 54 BCE, describes a device built by Posidonius of Rhodes that showed the daily motions of Sun, Moon, and five planets. He also describes a similar "sphere" made by Archimedes, captured by the Roman general Marcellus at Syracuse in 212 BCE. The astronomical analysis of the mechanism's eclipse dial (epoch 205 BCE) and star positions best matches observations from Rhodes, supporting the idea that the device originated from Posidonius's philosophical school there.

These references confirm what the two handwriting styles in the inscriptions already suggest: the Antikythera mechanism wasn't a one-off miracle. It came from a workshop tradition that spanned centuries and multiple cities. Bronze was routinely melted down for reuse. The only reason any example survives is that this one sank with a ship.

What the Gears Tell Us

The Antikythera mechanism is not a metaphor. It's not a symbol of lost wisdom or a prompt for mystical speculation. It's a piece of engineering, built by people who understood gear ratios, astronomical cycles, and the limits of their materials. They knew what they didn't know (the orbits aren't circular) and they built workarounds that produced accurate results anyway (the pin-and-slot trick).

The inscriptions prove it was meant to be used, not displayed. The two handwriting styles prove it was built by a team. The three bronze alloys prove it was manufactured, not sculpted. The Saros epoch of 205 BCE places it in a specific historical moment. The Games dial, naming Olympia and Nemea, places it in a specific culture.

This was a product. A tool. Built by professionals, for professionals, in a workshop that probably made more than one.

And then the workshop closed. The tradition broke. The bronze got recycled. And for 1,400 years, nobody on Earth could build anything this good.