Features

net.wars: Time bandits

"Time is an illusion. Lunchtime, doubly so." How we measure that illusion is about to become the focus of an international battle.

The Wall Street Journal reported last Friday that the US has secretly proposed to the International Telecommunications Union, which coordinates the broadcasting of official time signals, that leap seconds should be eliminated because they create trouble for systems such as GPS. If you're an astronomer, for whom time is inextricably linked to the physical rotation of the earth, this move would turn time into that illusion. If there's an analogy, it would be the end, in August 1971, of the gold standard, which severed the connection between the currency in circulation and its physical meaning as a fixed weight of gold.

Until the 1960s, a second was defined as 1/86,400 of the mean solar day. But decades before that, it was already known that the earth's rotation around the sun wasn't uniform. There are various factors at work – the moon gets further away and its gravitational pull on the tides changes; the earth's molten core shifts – and in 1967 the world shifted to atomic clocks, which use the decay rate of caesium to tick off the seconds with extreme accuracy.

 Too much accuracy: if you compare atomic time, known (TAI, for International Atomic Time) to the earth's actual rotation, TAI diverges as the earth's rotation slows down. UTC is corrected to stay within .9 seconds: periodically the International Earth Rotation and Reference Systems Service adds a leap second. At this point TAI is ahead of UTC by 32 seconds . GPS, which only started up on January 6, 1980, is in the middle: it doesn't correct for leap seconds, so it is 19 seconds behind TAI, and 13 seconds ahead of UTC. The next leap second is due at the end of 2005.

These differences are not as trivial as they may sound at first. The reason: clocks are navigational instruments, going all the way back to the 17th century and the longitude problem. They are also how we know where to look for orbiting satellites or where precisely to point radio telescopes.

Mike Hapgood, the geophysical secretary of the Royal Astronomical Society, puts it this way: "One second off will be within beam, but once you disconnect, within two to three years a spacecraft wouldn't be in the radio beam and you wouldn't be able to communicate with it."

A 2001 article for Metrologia explains the case for eliminating the leap second, arguing that our dependence on satellite navigation systems (instead of a sextant and the stars) means that the leap second may be "detrimental to some systems, possibly creating life-threatening situations".

Dave Mills, a professor at the University of Delaware and a major contributor to Network Time Protocol (NTP), which synchronies computers all over the Internet, says that NTP won't care much either way, though it runs on UTC and issues leap second warnings. (Only a very few operating system kernels – FreeBSD, Linux, and Solaris – can, he says, recognizee these warnings and insert the leap second precisely at the exact time.) In theory, he says, the Internet could change to TAI and send out a table of the UTC offset, but standard radio time broadcasts are not equipped to carry more than a very small value for that offset.

"Personally, I like the idea of running NTP on TAI and using NTP and/or the Web to retrieve the table of insertions," he said by email. However, he added, "I haven't heard anybody make the killer argument to abandon leaps."

Ignoring the political hostility that has everyone pointing fingers and calling each other "special interests", it seems clear that there is a genuine problem: people need different kinds of time. You can't please everybody all of the time.

For social time (or maybe "common time" would be a better term), we do actually prefer to be roughly synchronised with the earth's rotation. It's not just the romance of the farming lifestyle; few people really like getting up and going to work when it's dark. (Although there might be benefits to going to work at what we now call midnight, getting off at 8am, and having daylight "evenings" to socialise in.)

Most human beings naturally sleep when it's dark. Keeping that is easy: let the seconds mount up and every four years, add whatever you need to the length of February 29 to even things up. (This is an idea of my own, not one that forms part of the ongoing discussions, but it seems simple and obvious enough.)

Solving the problem for precise time is much harder. You've probably already thought of the obvious possibility: redefining the length of the second so that it stays the same fraction of the mean solar day. The Metrologia article gives that idea short shrift: "It would require a redefinition of all physical units and systems that depend on time, as it would affect all time scales, including GPS time." Not a goer, then. It seems most likely that there will have to be two systems, one that is coupled to the solar system and one that isn't. And a fancy network of computers sitting in between the two to do the translation.

So: when the computer strikes midnight, what time is it?