Leap seconds

Why does the whole world have to argue about inserting a single second?

One second. Not a leap day, which reshuffles everyone's calendar and provokes no argument at all. A single second, inserted at midnight, roughly every eighteen months on average. And yet it took forty years of committee work, two deferred international votes, and finally a resolution of the General Conference on Weights and Measures in November 2022 to decide its fate.

So the puzzle is not really about seconds. Something about this particular second makes it far harder to agree on than its size suggests. What is it?

Start by noticing that we are trying to serve two masters with one number. Since 1967 the second has been defined by a caesium transition — a fixed, uniform interval that knows nothing about the Earth. But the day is defined by the Earth turning, and the Earth is not a good clock. Tidal friction is slowing it over geological time; the atmosphere, the oceans, and motions in the core shift it about on scales of weeks to decades.

So you have two time scales. International Atomic Time, TAI, simply counts caesium seconds. Universal Time, UT1, tracks the actual rotation of the planet. They drift apart, and the drift is not predictable more than a few months out.

Now ask the design question. If you want one civil time scale, which do you break? Civil time is UTC, and the choice made in 1972 was: keep UTC's seconds identical to atomic seconds, and whenever UTC threatens to run more than 0.9 seconds ahead of the Earth, insert an extra second so it falls back into step. Twenty-seven such seconds have been added since 1972, all of them positive, the most recent at the end of 2016. TAI now runs 37 seconds ahead of UTC.

Here is where the trouble begins, and it is worth being precise about the mechanism. A leap second is not a small error; it is a discontinuity. The minute 23:59:59 is followed by 23:59:60, a timestamp that a great deal of software simply does not believe exists. Time appears to stand still or repeat. Anything that assumes clocks increase monotonically — log ordering, database replication, lock timeouts, scheduling loops — can behave strangely, and in 2012 a Linux kernel bug triggered by the insertion took down a range of well-known services. The correctness of a distributed system frequently rests on the clock never going backwards.

Now who bears which cost? This is the heart of it. Astronomers, and anyone pointing an instrument at the sky or navigating by it, need a time scale tied to the Earth's actual orientation, because a second of rotation is about 0.46 kilometres at the equator. Operators of networks, satellites, finance systems and telecoms want a smooth, predictable, never-discontinuous count. Neither group is being unreasonable. They want different things from the same shared number.

And here is the property that makes it an argument rather than a decision. A time standard is only valuable because everyone else uses it. Its worth comes entirely from agreement, so no single party can improve its own position by defecting — a laboratory that unilaterally stopped applying leap seconds would not be more accurate, it would merely be wrong relative to everybody. That means the choice cannot be made by anyone locally; it can only be made collectively, and any collective body contains all the parties with conflicting costs.

There is a further wrinkle that people find surprising. Since about 2020 the Earth has been spinning slightly faster, and several of the shortest days on record have occurred in recent years. That raises the prospect of a negative leap second — deleting a second rather than adding one — which has never been done, and which essentially no software has ever been tested against. If insertion is risky, deletion is an untested risk.

The resolution, when it came, was characteristic of a coordination problem: not to solve the disagreement but to widen the tolerance. The 2022 resolution calls for the maximum permitted difference between UTC and UT1 to be increased by or before 2035, which in practice retires the leap second without abolishing the link to the Earth outright. The difference will simply be allowed to accumulate, and how it will eventually be handled is left to be settled later.

The misconception worth correcting is that this is a fight between precision and convenience, with the astronomers being fussy. It is not. Both camps want precision; they want it in different quantities. Atomic time is precisely uniform and imprecise about where the Earth is pointing. Rotational time is precise about the Earth's orientation and non-uniform in interval. You cannot have one number with both properties, so the question was never technical — it was about which of two irreconcilable needs the shared standard should serve, and who absorbs the mismatch.

Notice too what the 2022 decision does not do. It does not sever civil time from the Earth. It lets the two drift further before intervening, which converts a frequent small discontinuity into a rare large one. That is a trade, not a fix, and the eventual large correction is a problem deliberately handed to a later generation.

• How distributed systems now sidestep the discontinuity with leap smearing, and why the smears are mutually incompatible.
• What a negative leap second would actually require, and why nobody wants to be first.
• How UT1 is measured in the first place, given that it depends on watching distant quasars.

Every term the collection defines is gathered in the glossary.