A pendulum clock is wrong, and it is wrong in a specific and almost poetic way: it keeps worse time the wider it swings. The period of an ideal pendulum is supposed to be the famous T equals two pi root L over g, independent of how far it swings, and that independence is why Galileo and everyone after him thought the pendulum was the perfect timekeeper. It is not independent. That formula is only the first term of a series, and the next term depends on the swing:
T is larger than the ideal by a factor of about one plus theta-squared over sixteen, with theta the amplitude in radians.
That little theta-squared over sixteen is what three hundred years of clockmaking had to fight.
I worked the numbers rather than wave at them. Take the leading correction and turn it into seconds lost per day against the small-swing ideal. A pendulum swinging five degrees to each side runs slow by about 41 seconds a day. At three degrees it is about 15 seconds; at one degree, still 1.6 seconds. These are enormous errors for a clock. If you want to lose less than a second a day you need to hold the swing under about three quarters of a degree, which is a barely visible twitch.
But the absolute error is not even the real problem, because you can calibrate a constant error away: if the clock always runs 15 seconds slow, you just make the pendulum very slightly shorter and you are even. The real problem is that the error depends on the amplitude, so if the amplitude changes, the rate changes, and there is nothing to calibrate against. I computed that too: a pendulum whose swing drifts from three degrees to three and a half, the kind of drift you would get as the driving force weakened or the oil thickened in the cold, shifts its rate by about five seconds a day. A half-degree wobble you would never see by eye is five seconds. The clock’s accuracy is hostage not to how wide it swings but to how steadily it swings.
Christiaan Huygens, who built the first pendulum clock in 1656, understood this precisely, and in his 1673 Horologium Oscillatorium he did something that still reads as beautiful. He asked: is there a path, other than the circular arc a plain pendulum follows, along which the period is truly independent of amplitude? There is. It is the cycloid, the curve a point on a rolling wheel traces, and a bob constrained to a cycloid is a tautochrone: it takes exactly the same time to fall to the bottom no matter where on the curve you release it. Wide swing or narrow, the same beat. Huygens even built it, hanging the bob on a flexible cord and placing curved metal cheeks at the top so that the cord wrapped against them on a wide swing and shortened the effective pendulum by exactly the right amount to trace a cycloid instead of a circle.
It was the exact answer. And it was useless. The cord bending against the cheeks introduced friction that changed the timing more than the circular error it was correcting, and by the time the mechanism was built there were larger error sources, temperature, air drag, the push of the escapement itself, that swamped the thing the cheeks had so elegantly removed. Huygens had solved the stated problem completely and solved the real problem not at all.
What actually gave the world three centuries of accurate clocks was not elegance, it was a cruder and much better idea: do not try to make the period independent of amplitude. Make the amplitude constant. If the swing never changes, the circular error is just a fixed offset, and a fixed offset is exactly the thing you can calibrate away.
The amplitude is set by the escapement, the mechanism that feeds the pendulum a tiny push each beat to replace what friction steals. The ordinary anchor escapement of Huygens’ day gave that push in a way that also shoved the pendulum backward slightly on each tick, a recoil that made the amplitude jitter with every variation in the driving force. Around 1715 George Graham introduced the deadbeat escapement, which reshaped the contact so that the escape wheel stops dead against the pallet instead of recoiling. No recoil meant the pendulum was left alone to swing at a steady amplitude, and a steady amplitude meant the circular error stopped drifting and became a constant you could tune out. Graham’s clocks kept time to about a second a day, and that standard held until the twentieth century. The deadbeat escapement did not fix the theta-squared term at all. It made it hold still.
I find this a quietly instructive little piece of history. The elegant, exact, provably-correct solution, a tautochrone derived from first principles by one of the great minds of the age, lost to a reshaped piece of brass that ignored the real problem and controlled a different variable instead. Not because Huygens was wrong, he was exactly right, but because being exactly right about the stated problem is not the same as being useful about the real one, and the real problem had friction in it that his model did not. The craft that survived was the craft that engineered around the mess rather than solving through it.
None of it survives in your pocket, of course. A quartz clock does not swing; it counts a crystal’s vibrations, and there is no amplitude to drift, which is its own kind of answer to the whole three-hundred-year argument: if you cannot keep the swing steady, stop swinging. But every tower clock still ticking did it Graham’s way, holding a half-degree steady against the cold, losing its patient second a day, carrying a correction for a curve it never quite traced.