About Antikythera Mechanism
A working model of the oldest known geared calculator, built in the Greek world around the 2nd century BC (dates from about 205 to about 100 BC have been argued). Turn the handle and every dial moves as the tooth counts make it move. Three colours run through the page: SURVIVES for what is in the fragments, INFERRED for what is missing but well supported by what survives, and RECONSTRUCTED for modern proposals of which nothing survives.
THE MECHANISM: the front and back faces. Drag the handle on the side of the case, type a date, step by a day, a month or a year, or START to turn it steadily (PAUSE stops it, RESUME carries on). INSIDE: the wheels, with what each one does and how sure we are of it, and a close-up of the pin-and-slot. THE CYCLES: why these numbers of months. WHAT SURVIVES: the fragments and the story of how they were read.
THE STORY
Sponge divers from Symi found a Roman-era wreck off the island of Antikythera in 1900, and salvage with the Greek navy brought up statues, pottery and, in 1901, a lump of corroded bronze and wood. In May 1902 Valerios Stais at the National Archaeological Museum in Athens noticed a gear wheel in it. The wreck is dated to about 70-60 BC. The mechanism itself has been dated from its lettering to the second half of the 2nd century BC; other studies argue for about 205 BC or for a date nearer the wreck. It is still argued.
Derek de Solla Price studied it from 1951, wrote "An Ancient Greek Computer" in Scientific American in 1959, and with Charalampos Karakalos made X-ray and gamma-ray images in 1971. His Gears from the Greeks (1974) showed the 19-year calendar gearing. Michael Wright, then at the Science Museum in London, worked through new X-ray images and in 2002 built the first working model with the Moon's phase and a planet display of his own design. In 2005 the Antikythera Mechanism Research Project took the fragments through high-resolution X-ray CT and surface imaging; their 2006 paper in Nature showed the pin-and-slot, the Saros eclipse spiral and twice as much readable text. In 2008 they read the month names of the calendar and the games dial. Front-cover text read in 2016 gave planetary periods, and in 2021 the team proposed a full model of the planet display.
HOW THIS MODEL WORKS
Every pointer is the main wheel's turn multiplied through tooth counts: 64/38 × 48/24 × 127/32 = 254/19 turns of the Moon a year, 64/38 × 53/96 × 15/53 = 5/19 turns of the Metonic pointer (235 months in 5 turns), × 57/60 for the four-year dial, × 15/60 × 12/60 for the Callippic dial, and 64/38 × 53/96 × 27/223 × 188/53 × 30/54 for the Saros pointer (4 turns in 223 months), then × 20/60 × 15/60 for the Exeligmos dial. The tests behind the page check these products against the cycles.
The pin-and-slot uses the published estimate: k1 and k2 on axles about 1.1 mm apart, the pin 9.6 mm from k1's centre, so the Moon pointer runs up to about 6.5° ahead of or behind its average place (6.58° by this page's sums). Because the pair rides on e3, which turns once in about 8.88 years, the fast and slow part follows the Moon's anomalistic month.
One turn of the main wheel is taken as 365¼ days, the year built into Callippus' 76-year cycle; the 365-day calendar ring was probably moved back a day every four years, which the page does for you, one hole at a time. Between moves the pointer can sit up to half a day either side of the true day, as it would on the real one. The mechanism has no clock of days, so this is the page's choice. Dates are in the Julian calendar up to 4 October 1582 and Gregorian after; years before AD 1 can be typed as BC.
The pointers are set to the mean Sun and Moon, from modern formulas, at the new moon of late April 205 BC. The two spirals start at proposed starting months: the Saros dial at the month after the new moon of 28 April 205 BC (Carman and Evans, 2014) and the calendar at the month after the new moon of 23 August 205 BC (Iversen, 2017). Months start a day and a half after the mean new moon, standing in for the first sight of the crescent. Both start dates are argued over. The four-year and Callippic dials start with the calendar; the four-year dial's Year 1 then holds the Olympic games of 204 BC.
Turn far from 205 BC and the gearing drifts from the real sky, as the real one would: 235 months are about 2 hours longer than 19 true years, and a 365¼-day year gains about three quarters of a day a century. The readout compares the pointers with today's formulas for the real Sun and Moon.
ECLIPSES
The real Saros dial carried glyphs in some months, giving lunar or solar eclipse predictions with times; 18 were known in 2008. This page does not copy them. The marks on its Saros spiral are worked out from mean motions: a month gets a lunar mark if the Moon is within 10.5° of a node at full moon, a solar mark if within 16.5° at the new moon that ends the month. They are labelled as this page's predictions. As a check, of 15 glyph months taken from the 2008 notes, 14 come out the same here and the other is one month off; but the start date was itself chosen by its authors to fit the glyphs, so this is no independent proof.
WHAT IT LEAVES OUT
- The planets are a modern reconstruction. Inscriptions show the planets were displayed; no planetary gearing survives. The page moves them with the 2021 period relations (Venus 289 synodic cycles in 462 years, Saturn 427 in 442, from numbers read on the front cover; Mercury 1513 in 480, Mars 133 in 284 and Jupiter 315 in 344, derived by the 2021 team) on simple circles, not their gear trains. No true Sun pointer and no node pointer are shown.
- When a spiral pointer comes to the end of its groove (the fifth turn of the calendar spiral, the fourth of the Saros), the page puts its follower back at the start for you. On the real one the user had to lift the follower out of the groove and set it back at the start by hand.
- The 365-day calendar ring is the long-standing reading, and the month names Pachon, Payni and Epiphi survive on it. Hole counts since 2020 suggest the ring may have had 354 or 355 divisions, a lunar year. The page keeps the 365-day reading and marks it as inferred.
- The Metonic cells held month names, with some months doubled; the page gives month numbers only, since which months were doubled is reconstructed.
- Pointers are thin lines; the real ones are lost. The case and the dial shapes are drawn for this page and are not copies of any published drawing. The Greek lines above and below the front dial follow published readings of the parapegma, a star calendar keyed to letters on the zodiac. Much of their wording is restored by editors from a few surviving words, and lines missing from the 2016 edition by Bitsakis and Jones are left out; their arrangement here is the page's, and the key letters on the zodiac ring are left out.
- The inside view is a plan with the wheels moved apart so each can be seen. Wheel sizes are squeezed; tooth counts are true. Most surviving wheels are broken, and their counts are estimates from the surviving teeth, checked against the cycles they make.
- The real Sun and Moon use the main terms of modern formulas and ignore the few hours by which the Earth's rotation has slowed since antiquity. They are a check, not an ephemeris.
Main sources: Freeth et al., Nature 2006 and 2008 (with supplementary notes); Freeth et al., Scientific Reports 2021; Carman and Evans, Archive for History of Exact Sciences 2014; Iversen, Hesperia 2017; Price, Gears from the Greeks, 1974. Astronomy from Meeus, Astronomical Algorithms.
FURTHER WATCHING AND READING