IT'S ABOUT TIME

The origin of leap seconds, and why they should be abolished

How many seconds have there been since January 1, 1972?

While that number is extremely large, it is pretty simply calculated. Multiply the number of years (and don’t forget leap years!) by the number of days in a year by the number of hours in a day by the number of minutes in an hour by the number of seconds in a minute and voilà: 1,371,513,600 seconds as of midnight on June 29, 2015.

Except that that’s not right. We’re leaving out leap seconds. Leap seconds, you ask? Yes, leap seconds. There have been 26 of them, and number 27 is just around the corner: It will be added to global clocks at 1am London time on July 1.

The leap second is a near paradox. It befuddles all common definitions of time. Notionally, it’s a way to unify all our ways of measuring time. In reality, it’s just an attempt to preserve an old definition of time that has long since been superseded by newer methods. In the process, the leap second—through no fault of its own—puts at risk countless critical computer systems around the world. And that is forcing even the people who are charged with administering the world’s supply of leap seconds to consider getting rid of them altogether.

What is a second?

To understand the leap second you first have to understand what a normal second is.

We know there are 24 hours in a day, and 60 minutes in an hour, and 60 seconds in a minute. So effectively, a second is 1/86,400 of a day. But what is the length of a day?

In most parts of the world, it’s pretty easy to know the difference between yesterday and today. We remember that the sun was up and then it went down, we slept for a while and now we’re up and the sun is back in the sky. (Life in the Arctic and Antarctic circles can get trickier.)

Increasing the precision of this measurement isn’t too hard either. To mark the elapsing of a full day, track the sun in the sky until it stops rising and mark the moment before it starts descending. Give that moment a name—let’s say, “noon.” Every time we observe a noon we know that a day has elapsed.

Dividing up the time from noon to noon gets tricker. Civilizations have used all sorts of devices to do this—devices that relied on flowing water, or falling sand, or melting candles, or anchored weights, or unwinding springs, or something else. Let’s say we have one of these devices. Let’s also say that this device is perfectly accurate; it will always measure the exact length of time we’ve calibrated it to.

 Just as a figure skater spins faster or slower depending on the shape of her body, so does the earth. 

So we calibrate it to tick off every hour from noon yesterday to noon today. Every time we see a noon, that high point of the sun in its track, we tick off a day.

Thirty days in, however, we will realize something: the time it takes to get from noon yesterday to noon today has been slowly changing.

The reason is this: Observing the movement of the sun is, of course, a proxy for observing the rotation of the earth. And the speed of this rotation—the length of a day—varies.

Part—a large part—of the reason for this is that the earth’s orbit around the sun is not a perfect circle but an ellipse, so the amount by which earth has to spin each day for a given point on its surface to have the sun directly overhead again varies over the course of a year. However, changes in the atmosphere, glaciers, tectonic shift, ocean swells, and even soil moisture all constantly cause tiny variations in the earth’s shape. Just as a figure skater spins faster or slower depending on the shape of her body, so does the earth.

 “The problem is that the earth is a lousy clock.”—Demetrios Matsakis 

Tides also affect the earth’s rotation, through friction with the seabed, as well as through crashing waves and flowing swells. The discrete movements of the earth’s geological shells—the inner core, outer core, and mantle—spinning at different rates have their own effects too.

So herein lies the first problem. If we define the second as 1/86,400 of any one day, it won’t meet the definition on most days.

Scientists have known this for close to a century. As such, the length of a second was initially defined not as 1/86,400 of a day, but as 1/86,400 of an average day.

But the average day isn’t constant from year to year either. First, because, just statistically speaking, when each year is made of days of varying length, some years will end up with slightly longer days, on average, than others. And second, because the earth’s rotation is, very gradually, slowing down.

As Demetrios Matsakis, the chief scientist for time at the US Naval Observatory, puts it, “The problem is that the earth is a lousy clock.”

The new second

As soon as it was practical, scientists made it their goal to define a measurement of a second that would be consistent from now until the end of time. In 1967 they adopted an atomic standard.

When an atom jumps from one energy state to another, it emits radiation. For any given pair of energy states, that radiation always has exactly the same frequency, dictated by the most basic laws of physics. So it’s a reliably constant measure that holds true everywhere. In 1967 a second was defined as a certain number of cycles of the radiation from a certain kind of atom—cesium—jumping between two particular energy states.

But of course, to decide how many such cycles constituted a second, the scientists first had to choose a second to count them in. This turned out to be a somewhat arbitrary process.

 The length of a second is based on what the average day of 1900 was predicted to be in 1895. 

For historical reasons, they decided to use an average second in the year 1900—that is, 1/86,400 of the average day in 1900. Except they didn’t. Because all this happened several decades after 1900, the scientists couldn’t directly measure the length of the average day in 1900. Instead, they used the best figures available—which turned out to be what the length of the average day in 1900 was predicted to be five years in advance, in 1895.

So the modern definition of a second is 9,192,631,770 cycles of the radiation of a cesium atom jumping between two particular energy states, because that’s how many cycles it took to fill 1/86,400 of what the average day of 1900 was predicted to be in 1895.

The Master Clock at the US Naval Observatory is calibrated to this definition. The clocks in GPS satellites are calibrated to this definition. Nearly every clock in the world, be it inside a phone or an automobile, is in some way—directly or indirectly—calibrated to this definition of a second. And it is a definition that was plucked almost at random from the range of possible seconds offered up by a variably rotating earth.

Mo accuracy mo problem

With such a precise definition of a second, we have decoupled the earth’s movement from time. One day is no longer exactly 86,400 seconds. And given the arbitrary way in which the length of the standard second was chosen, it was pretty much inevitable that it would prove to be either ever-so-slightly shorter than 1/86,400 of the average day decades later, or ever-so-slightly longer.

It is, it turns out, on the short side.

Scientist knew this was going to happen when they were defining the specification for Coordinated Universal Time—a global standard for civil time. To compensate, they specified that this timescale, which goes by the abbreviation UTC, should never deviate more than 0.9 seconds from the rotation of the earth. They said that from time to time, as necessary, one second should be inserted or removed from clocks to realign them with the earth’s rotation. It has proven necessary to insert seconds, rather than remove them. Thus the leap second was born.

Chart showing the observed speed of earths rotation from the 1970s to today

The job of deciding when to add a leap second falls to scientists at the International Earth Rotation and Reference Systems Service (IERS). They determine the speed of the earth’s rotation by pointing radio telescopes around the world at the same quasar at the same time. By comparing the measurements from each telescope to the others, they can determine not only the rotation (and wobble) of the earth, but also tectonic shifts through triangulation methods.

Whereas with our primitive device before—measuring noon to noon—the variations we saw in the length of the day were mostly due to the earth’s elliptical orbit, the accuracy of IERS’s radio telescopes, combined with other methods (such as bouncing lasers off the moon and tracking GPS satellites), allows us to measure the length of a day with very high precision. When IERS detects that the earth is falling too far behind UTC, it calls for a leap second.

A chart showing the deviation of UTC from the Atomic Time

The deliberations over the inclusion of leap seconds into UTC have become lore amongst the people who keep time—scientifically at least—around the world. Apparently, the mid-century scientists thought they were doing us all a favor. Mariners and astronomers relied on accurate time to confirm the location of celestial bodies for navigation and study. Without leap seconds, their star charts would have gradually drifted out of sync with the heavens.

However as other consortia and organizations implement time-reliant systems, they have to decide to sync them to the atomic clock, to UTC, or to something else.

The GPS system picked something else. Every GPS satellite has an atomic clock on board counting fluctuations of the radiation from a rubidium atom. (Rubidium clocks are slightly less accurate but more portable than cesium clocks.) The GPS timescale started in 1980 in sync with UTC, but subsequently hasn’t implemented any leap seconds. On the other hand, Glonass, the Russian-built and -operated positioning system, a rival to GPS, is synced to UTC, and unlike GPS, it observes leap seconds.

A chart showing how GPS time used to be the same as UTC time but now runs parallel to both UTC and Atomic Time because of leap seconds.

An increasing level of precision

The increases in the precision of time came in response to the demands of an ever-more-organized planet.

There was a time where if a clock ran a five or ten minutes fast over the course of a day, it didn’t matter. Everyone’s clocks were this inaccurate, and most activities didn’t require hour-by-hour synchronization. A farmer doesn’t need to let the cows out at 6am. A farmer lets the cows out in the morning.

Even as urban life developed, time was a local determination. Noon was when the clock on the bank or city hall said noon. There wasn’t even a guarantee that the bank and city hall would agree to put the same time on their clocks. And as for a clock a few towns away, let alone on the other side of the country, it didn’t matter if it showed the same time, because there was no way to get there fast enough to see the difference.

 There wasn’t a guarantee that the bank and city hall would agree to put the same time on their clocks.  

The rise of railroads changed everything. Towns had to synchronize their clocks to the railroad’s clock so that passengers knew when to expect the trains. The railroad couldn’t be expected to publish timetables using the time listed on each town’s bank or city hall—they might be wildly out of sync and could be reset at any moment.

Since then the demand for accuracy has snowballed. Synchronizing traffic signals requires precision better than a minute. Telephones, smartphones, the internet, and GPS synchronize at a precision of less than a second. The NASDAQ recently announced that from this October it will increase the precision of its timestamps to include nanoseconds.

But here’s the thing: None of these devices depends on the position of the sun. The earth’s rotation and UTC might slowly move out of sync until 12 noon is in the middle of the night, but atomic clocks would still show time accurately, GPS receivers would still locate you precisely, and trains would still run to their schedules.

The case for ditching the leap second

So why is the leap second necessary? While it was once convenient to astronomers and mariners, both groups no longer rely heavily on UTC. “One second is too coarse for accurate astronomy,” says Matsakis, himself an astronomer. “The radio astronomers go to IERS” to know the earth’s precise rotation “instead of the clocks.” As for mariners, they, of course, now use GPS to navigate, which has no leap seconds.

  “There is this this false correlation between the sun and time.”—Elisa Felicitas Arias 

So the main point of adding leap seconds to UTC seems to be to make sure that what we humans perceive as noon—the sun at its highest point in the sky—is also what our clocks say noon is. But does that really matter?

Even the keepers of UTC seem to doubt that it does. “There is this this false correlation between the sun and time,” says Elisa Felicitas Arias, the director of the time department at the Bureau International des Poids et Mésures—the supervisors of the world’s time standard. Without the leap second, she says, “One day you will have breakfast at noon, but not me—this will happen in thousands and thousands of years.” The rate at which time creeps away from the earth’s rotation will pass completely unnoticed by anyone just going about their life. Children will experience near-identical timings of sunrises, noons, and sunsets to their parents, grandparents, and great-grandparents.

Indeed, most people’s local time is already well out of sync with the rotation of the earth and the position of the sun, because of time zones. Moreover, over the course of a year, the length of a solar day fluctuates by as much as 16 minutes.

Time, as most people experience it, is a political construction, not a scientific one.

Map of how much solar noon differs from local clock noon around the world

So if we already live in a world where that type of imprecision is tolerated, why not let the earth drift out of sync with the clocks by one second every few years? At this rate, 100 years from now we will have let the earth deviate just over a minute, and it will be 5,800 years before clocks deviate an hour—a shift that countries with daylight saving time currently put up with twice every year. That’s further in the future than the invention of writing is in the past.

One day our descendants will look back at our timing systems the same way we look at our ancestors’ cuneiform.

Bad programmers

It might seem, then, that leap seconds are at worst a harmless anachronism. But in fact they can be dangerous things. As more systems become digitized and interconnected, the risk grows of something bad happening when a leap second is inserted.

 It might seem that leap seconds are a harmless anachronism. But in fact they can be dangerous things.  A computer system that didn’t adjust for the leap second might confuse a computer that did by seemingly sending information that appears to come from the future. Many computer systems and programs weren’t designed to account for leap seconds; the programmers who developed them didn’t consider that a day could be longer than 86,400 seconds long. According to Matsakis at the US Naval Observatory, a one-nanosecond—that’s one billionth of a second—discrepancy on a GPS clock will yield an inaccurate location by a foot (0.3 m), so a botched leap-second insertion into a satellite-based positioning system could be disastrous.

Even the digital clock outside the US Naval Observatory—the official keepers of time in the US—won’t properly display the time for four hours after the leap second is inserted, due to the code written by the clock’s manufacturer, according to Matsakis. (Update: June 30: Matsakis tells Quartz that the GPS receiver in the clock has been replaced and is expected to display the correct time at the application of the leap second.)

Companies are dealing with the upcoming leap second on June 30 in many different ways. Some US exchanges are delaying the open or advancing the close (paywall) of certain trading markets as a precaution, so that the leap second doesn’t fall during trading hours. Amazon and Google are “smearing” the application of the leap second, which means that instead of inserting an extra second on their servers’ clocks between 23:59:59 UTC and 00:00:00 UTC they’ll make the seconds leading up to midnight UTC slightly longer, so that the leap second pushes their clocks back into synchronization with UTC.

 The programmers who developed them didn’t consider that a day could be longer than 86,400 seconds long. 

Japanese financial exchanges are smearing seconds leading up to the change too. Australian and South Korean markets are smearing the seconds immediately after the leap second. The Singapore markets are delaying implementation of the change until after its markets close. The Japanese time stamping authority—a service to authenticate digital documents—is shutting down momentarily while the leap second is applied.

And while exchanges can test their own systems to make sure they continue to function properly when the leap second comes, they cannot guarantee how third-party systems that connect to them will react. Those third parties are trying to avoid the fate that struck companies like Qantas, Gawker, Reddit, Foursquare, and Yelp during previous insertions of the leap second. All of suffered malfunctions, according to news reports.

A future without leap seconds

The continued application of leap seconds will be voted on this November at the World Radiocommunication Conference in Geneva. So it’s possible that this leap second could be one of the last, though Marek Kukula, an astronomer with the Royal Observatory Greenwich, told Wired that the debate could go on for for another decade or more.

If the leap second is abolished, noon might not be exactly when the sun is at its apex, but that’s pretty much how we’ve lived for a century already. And by the time the earth has drifted sufficiently out of sync with time for anyone to notice, we’ll probably be using a different time-keeping technology altogether.

home our picks popular latest obsessions search