Antikythera Mechanism – An Engineering Audit of the Academic Record

A half-century of academic literature catalogued the Antikythera bronze as a precision astronomical device. Two engineers in Argentina then took the accepted structural model and applied standard tolerance testing in April 2025. Their simulation locked up the physical components in under three simulated months.

The Discovery of the Antikythera Mechanism

Sponge divers found the wreck first. In 1900 and 1901, a team led by Captain Dimitrios Kontos was working the seabed 45 metres down off the Greek island of Antikythera when they hauled up a Roman-era cargo of bronze and marble. The pieces went to the National Archaeological Museum of Athens and sat unread for two years.

Then Spyridon Stais, a former Greek Minister of Education, walked past a corroded lump in May 1902 and spotted a gear wheel inside it. That was the moment the calcified debris became a puzzle.

The bronze itself was doing the work of destruction. Two thousand years of seawater had converted most of it into atacamite, a brittle green mineral that shrank as it dried and split the object open across 82 surviving fragments. Early scholars, including the German researcher Albert Rehm, argued the object was an astronomical calculator, but nobody could see the inside to confirm it.

June 1959 changed the story. Derek J. de Solla Price, an English physicist at Yale University, published ‘An Ancient Greek Computer’ in Scientific American. The seed of the modern consensus was planted in a magazine article.

Price’s 1974 Differential Gear Theory and Structural Discrepancies

Price came to the artefact through the American Philosophical Society. In 1958 the Society funded his first trip to Athens to examine the surviving fragments. By 1971 he had brought in Charalampos Karakalos, a radiographer at the Greek Atomic Energy Commission, to produce the first gamma-ray scans of the bronze.

The scans were flat, two-dimensional shadows.

In 1974 Price published ‘Gears from the Greeks‘, a 70-page monograph. He counted the visible teeth on each gear and matched the counts to the 19-year Metonic cycle, a known astronomical period containing exactly 235 lunar months. Price argued the object contained a differential gear, an assembly that would subtract solar motion from lunar motion to output the moon’s phase.

That claim became the anchor of the modern consensus.

Michael Wright, later a curator at the Science Museum in London, went back to the radiographs and examined the physical fragments themselves. He observed pillars rising from the bronze wheels that ran straight through the space where Price’s differential assembly would have to sit. The central arbour was fixed to the base plate, and the geometry made a differential physically impossible.

What Wright documented was an epicyclic gear train, a simpler arrangement in which outer gears rotate around a central one.

The catch was the format of the 1974 monograph itself. It maps astronomical gear ratios and notes divisional errors in the tooth counts. No physical tolerance simulation appears anywhere in the 70 pages.

That absence was published, peer-reviewed and accepted by the American Philosophical Society without a public engineering review. Whether a single referee flagged the physical impossibility of Price’s differential gear cannot be answered from the public record. The 1974 peer review correspondence remains uncatalogued for this audit and sits, as far as can be established, in the Society’s archive files.

The 2006 BladeRunner CT Imaging Findings

Everything changed in 2005 when the X-Tek BladeRunner arrived in Athens. This was an eight-ton computed tomography scanner running at 450 kilovolts, purpose-built for looking inside dense industrial objects. It rotated the fragments through thousands of X-ray projections and reconstructed the interior at 50 micrometres of resolution.

The imaging team was not working alone. Tom Malzbender of Hewlett-Packard Labs turned 50 sequenced flashbulbs on the surface of the object and captured its topography through polynomial texture mapping, a method that pulls faint inscriptions out of corroded metal.

The results reshaped the record. Freeth and colleagues published in Nature in November 2006, outlining a 37-gear interlocking model that tracked the solar year, lunar phases and eclipse prediction through a 223-month cycle. Thousands of characters in the inscriptions were deciphered and read as a user manual for the machine.

A subsequent 2008 Nature paper by the same consortium added the Olympic Games timing dial to the picture.

Scans also showed something the theoretical papers noted only in passing. Every one of the 30 surviving gear wheels had been cut by hand. Their teeth were shaped one at a time, not milled to a common tolerance.
None of the 2006 or 2008 Nature papers ran a physical tolerance simulation on the reconstructed 3D model. Whether the imaging team ran any friction or torque check internally and chose not to publish it is not on the public record.

Mike Edmunds’ 2011 Accuracy Assessment

Mike Edmunds sat with the 2005 imaging data for six years. In August 2011 the Cardiff University astrophysicist published ‘An Initial Assessment of the Accuracy of the Gear Trains‘. He mapped the spacing of the hand-cut teeth from the CT reconstructions and logged what he found.

The teeth were not evenly spaced. Random errors sat next to systematic ones across every fragment he measured. Several gear axes were slightly off-centre, producing eccentricity errors that make a rotating wheel wobble and vary its output speed within every turn.

Printing his findings in August 2011, Edmunds logged a specific caveat alongside the physical spacing measurements. Cumulative flaws in the off-centre gear axes meant the physical bronze could never sustain the necessary mathematical accuracy. He theorised the original maker built the object for basic display rather than precise prediction.

The suggestion sat in the paper. No historical journal picked it up as a challenge to the working consensus.

Not a single mechanical engineering group was commissioned by any journal to run a tolerance simulation on the error metrics Edmunds had just made public. The 2011 paper became a footnote in a series of publications that carried on treating the artefact as a functioning calculator.

That paper sat in the public record for nearly fourteen years before a mechanical engineer ran the tolerance simulation the metrics demanded.

The Academic Consensus Gap – Historical Drift (2012–2024)

Nobody in the engineering faculties picked up the 2011 Cardiff metrics. Museum curators and university presses simply spent the next twelve years releasing physical models of a smoothly operating device. Not one of these subsequent structural papers cited the spacing flaws Edmunds had placed on the record.

The drift is visible in the physical bibliography. MDPI Heritage Journal printed a 2021 paper explicitly titled ‘The Prove of the Accuracy of the Astronomical Calculations Based on It‘.

A full decade after the off-centre gear axes were measured and catalogued, peer review was still passing theoretical mathematics as physical proof.

No mechanical tolerance simulation appears in the historical, astronomical or museum literature across the entire twelve-year span. Public museum installations and television documentaries in the same period presented the artefact as a working precision calculator. The Edmunds caveat sits in the supporting literature, unrebutted and uncited.

The 2025 Antikythera Mechanical Simulation

Szigety and Arenas uploaded arXiv preprint 2504.00327 from the National University of Mar del Plata in April 2025. Taking the raw 2011 spacing measurements, the two engineers mapped those physical flaws onto the flat triangular teeth of the accepted model. They simply ran the load calculation that the previous papers omitted.

Their teeth are the physical hinge of the finding. Modern industrial gears use an involute profile, a smoothly curved shape that keeps the contact point and the speed of motion steady as one gear drives another. On the Antikythera artefact, the teeth are simple flat triangles.

That difference matters more than it sounds.

When a flat triangular tooth pushes against another, the point of contact slides up and down the straight face. Effective gear radius changes with every fraction of a turn, and the driven wheel jerks between acceleration and deceleration within a single rotation.

The catch was the drive layout. A single hand crank powers all the interconnected gear trains at once, and a single jam anywhere brings the whole machine to a halt.

Feed the Edmunds error data into that geometry and the wheels stop meshing. The simulation shows the teeth either clash head-on and jam, or slip past each other and disengage. Failure runs at roughly 90 per cent, and the device locks up within a few simulated months of solar movement.

Szigety and Arenas state the finding as a binary. Either the object never functioned as intended, or Edmunds’ 2011 error measurements are wrong. Full computational output tables that would let another team check the 90 per cent figure are not appended to the arXiv PDF; they remain in the authors’ working archive.

As of this audit, no major historical or astronomical journal has published a formal response to the engineering paper.

Why the Consensus Overlooked Mechanical Reality

Handling of the primary fragments passed through three distinct academic groups over seventy years. Yale historians of science constructed the initial theory. Cardiff astrophysicists and mathematicians later took over the CT data to map the gear ratios.

Mechanical engineers arrived last, in Szigety and Arenas.

Each centre carried partial expertise, and the disciplines barely cited each other. Astrophysicists read the gears as vessels for mathematical ratios rather than as load-bearing physical objects subject to friction and wear. Mechanical engineering journals stayed out of the picture entirely.

Editorial architecture baked the disciplinary split into the academic record. Major historical journals routinely accepted schematic models of this object, asking authors for zero baseline friction or physical load tests in return. A ratio that balanced on a blackboard became acceptable evidence that the corroded bronze wheels actually turned.

Price himself drove much of this early momentum. He arrived at the Yale archives with a background in scientometrics, actively mapping the historical growth rate of human discovery.

An advanced ancient Mediterranean computer was exactly the data point his wider theory required.

The 1974 peer review correspondence for ‘Gears from the Greeks’ is held by the American Philosophical Society and has not been examined for this audit. Whether any 1974 referee flagged the physical impossibility of Price’s differential gear cannot be answered without those files.

Source

Sources include: Derek J. de Solla Price’s 1974 American Philosophical Society monograph ‘Gears from the Greeks’; the 2006 and 2008 ‘Nature’ publications by Freeth et al. detailing the X-Tek BladeRunner CT imaging; Mike Edmunds’ August 2011 assessment of gear train accuracy; Michael Wright and A.G. Bromley’s structural re-evaluations documenting an epicyclic gear train; and the April 2025 National University of Mar del Plata engineering tolerance simulation by Szigety and Arenas (arXiv preprint 2504.00327).