There Might Not Be An Elegant Solution

On a recent trip to London I visited the National Maritime Museum, which had an excellent exhibit on determining longitude at sea. I didn't know this, but finding a ship's longitude was one of the biggest scientific puzzles in history. The British Parliament in 1714 instituted a prize for solving this problem, although people had been thinking about it long before then. Two practical solutions emerged eventually: the lunar distance method and marine chronometers. Both solutions kept track of time at a fixed point -- Greenwich. Navigators then determined longitude by the difference between the ship's local time and Greenwich time.

The lunar distance method measured the passage of time using the position of the moon with respect to other celestial bodies. There was no simple formula describing the movement of the moon with sufficient accuracy, so navigators relied on nautical almanacs, which were basically lookup tables precomputed by the government. Such almanacs were published starting from 1767, and required teams of human computers to carry out the calculations every year.

Marine chronometers were the first clocks accurate enough to keep time on a moving ship. The first such clock was Harrison's H4 in 1761. It had an incredibly complicated mechanism, including a 5 Hz mechanical oscillator, a bimetal strip to correct for temperature variation, and clever machinery that allowed the clock to keep running while it was being rewound. H4 took six years to build, if you discount the two decades Harrison spent on its predecessors. Marine chronometers took until ~1825 to become simple enough to be manufactured at scale.

Solving the problem of longitude took decades of focused work, and the dedicated careers of multiple scientists and clockmakers. Though the problem is simple to state, its first practical solutions required vast amounts of painstaking calculations, tinkering, and engineering. I often look at a problem and think that it ought to have a clean, elegant solution, and that anything less parsimonious than that is not worth spending time on. I hear the same bias from other people, when they speculate that progress in neural nets / solar cells / insert-favorite-subfield-here will solve AI / green energy / insert-favorite-problem-here. There might not be a shortcut. There might not be an elegant solution. Solving the problem might just require a ton of detailed effort and engineering. And the solution might take longer than the time scale of a single person's career.