On this page
Long before mechanical clocks spread across Europe, a tower in Northern Song China used flowing water to turn a model of the sky and announce the time with moving figures. Its designer, Su Song, also wrote a book explaining it with drawings, which is the reason we can study it today. This post covers what the tower did, how its water-driven escapement worked, a worked example on reading its old measurements, and what a student can learn from it.
Who was Su Song?
Su Song lived from 1020 to 1101, according to a review of water-powered clocks by engineers at National Cheng Kung University in Mechanical Sciences (Hwang, Yan and Lin, 2021). He worked under the Northern Song dynasty, and the tower, according to the reconstruction paper discussed below, was originally used exclusively by the emperor.
His description of it survives in a book titled Xin yi xiang fa yao, which the same review translates as "New design for an armillary sphere and celestial globe" and dates to 1094-1096.
What did the astronomical clock tower do?
A paper on reconstructing the tower, presented at the CAADRIA 2017 conference by Chen and Liang, describes it as an automated astronomical demonstration device in three levels:
at the top, an armillary sphere for measuring star positions
in the middle, a celestial globe showing the movement of the heavens
at the bottom, a clock that told the time automatically, with wooden figures in a five-level pagoda-like structure
The same paper says Xin yi xiang fa yao gives a detailed account of more than 150 components with over 60 illustrations. The original tower was destroyed in 1127 during the Jingkang Incident. Su Song's son preserved the manuscript, but for a long time nobody could rebuild the machine from it.
How did the water-driven escapement work?
The heart of the tower was what historians call the waterwheel steelyard clepsydra. Hwang and colleagues describe it as a steelyard (a balance beam) combined with a waterwheel escapement, which behaves much like the escapement in a mechanical clock.
In simplified form:
Water flows at a steady rate into one scoop on the rim of a large wheel.
A balancing lever holds the wheel still while the scoop fills.
When the scoop is heavy enough, it tips the lever, the wheel is released, and it advances by one scoop.
The next scoop moves under the stream and the cycle repeats.
The review says the device had two kinds of feedback: one lever acts as a weight comparator, checking whether the scoop holds enough water to drive the wheel, and the flow rate into the scoop sets how often the cycle repeats. Compared with older float-driven clocks, it notes, the waterwheel could give a constant torque and drive the clock consistently.
Worked example: how tall was the tower?
Old Chinese texts give sizes in chi, a unit whose length changed over history. That makes the tower's size a real puzzle.
The CAADRIA paper gives the commonly quoted size: about 12 m high and 7 m wide. A study in Studies in the History of Natural Sciences (2011) argues instead that 1 chi for astronomical instruments was 24.5 cm, and that the recorded height was 35.65 to 36 chi and the width 21 chi.
Height: 36 x 24.5 cm = 882 cm, and 35.65 x 24.5 cm = 873 cm. So about 8.7 to 8.8 m.
Width: 21 x 24.5 cm = 514.5 cm, about 5.1 m.
Now check the older figure. For a 12 m tower to be 36 chi, each chi would be 1200 / 36 = 33.3 cm. The whole disagreement comes down to which length of chi you assume.
The same lesson appears inside the CAADRIA paper. It converts the inner diameter of the pagoda's first level, 6 chi 7 cun (6.7 chi), to about 2 m, which implies a chi near 30 cm. With 24.5 cm it would be 6.7 x 24.5 = 164 cm, about 1.6 m.
Common mistakes with units and levers
Assuming an old unit equals a modern one. Before converting, find out which standard the source used, and say so in your answer.
Dropping significant figures. 35.65 chi and 36 chi give different answers; carry both and report a range.
Forgetting that a balance compares torques, not weights. A small weight far from the pivot can balance a heavy scoop close to it. Sample numbers: a 2 kg counterweight 0.6 m from the pivot balances 2 x 0.6 / 0.2 = 6 kg of water 0.2 m from it.
If you want to see how balancing torques decide when something tips, the Fermi Simulation Lab has interactive simulations for ideas easier to see than read. For an earlier Chinese instrument with a clever trigger, read our post on Zhang Heng's seismoscope.
Why is the tower still debated?
The sources do not even agree on its dates. The CAADRIA paper says construction ran from 1086 to 1093, while the table in Hwang and colleagues' review lists the tower under 1088. Its height, as you saw, depends on the length of a unit. The CAADRIA authors note that parts that cannot be confirmed have to be treated as variables, so "there is no single result" for a reconstruction. Engineers keep rebuilding it at different scales to test which designs actually work.
What can a student take from how Su Song worked?
Write it down with drawings. The tower burned, but the illustrated book survived, and every modern reconstruction depends on it.
Turn a smooth process into countable steps. That idea, holding and releasing, is the core of every clock escapement.
Treat units as part of the evidence. A number without a reliable unit is only half a fact.
Practice routine: take three measurements from an old text or a historical problem in your book, convert each with two possible unit sizes, and write one sentence on how much the conclusion changes.
Next step: draw the four-step escapement cycle from this post from memory, then label where the torque balance happens.
