In 2015, physicists at JILA in Colorado built a strontium clock so stable that, if it had started running 15 billion years ago, before the universe itself existed, it would still be within one second of correct today. That number sounds like marketing. It isn’t. It comes from a measured fractional uncertainty small enough to make “drift” almost the wrong word for what these instruments do. But drift is still real, and it comes from two completely different places: the tiny imperfections inside the clock itself, and the fact that time, according to physics, doesn’t pass at the same rate everywhere to begin with.
What “One Second in a Million Years” Actually Means
No lab has run a clock for a million years to test it. Every accuracy figure for an atomic clock comes from a fractional frequency uncertainty, written as δf/f, a ratio describing how far the clock’s measured tick rate could plausibly stray from the true resonant frequency of the atom being used. A cesium fountain with an uncertainty of 1 × 10⁻¹⁶ means its rate could be wrong by one part in ten quadrillion. Multiply that fraction out against the number of seconds in a year and the result is the familiar “would neither gain nor lose a second in X years” claim. It’s a projection built from a short, intensely scrutinized measurement, not a literal stopwatch running across geological time. Labs verify these numbers by running clocks for a few weeks at a stretch and comparing results against the global reference-clock network coordinated through the BIPM in Paris.
How a Second Got Defined By an Atom, Not the Sky
Before 1967, the second was defined astronomically, as a fraction of Earth’s rotation or its orbit around the sun. That definition had a built-in problem: Earth’s rotation isn’t perfectly steady. In 1967, the 13th General Conference on Weights and Measures redefined the second in terms of the cesium-133 atom instead, tying it to something that doesn’t care about tides, earthquakes, or glacial rebound.
Every atomic clock built since, cesium or otherwise, exists to measure that number, or an equivalent one, as precisely as physically possible.
The Cesium Fountain Family: Three Decades of Getting Better
NIST’s line of cesium fountain clocks shows the improvement curve in concrete terms rather than abstract exponents.
| Clock | Year | Would Lose a Second In |
|---|---|---|
| NIST-7 (cesium beam) | 1993 | About 6 million years |
| NIST-F1 (launch) | 1999 | About 20 million years |
| NIST-F1 (improved) | 2005 | About 60 million years |
| NIST-F1 (final tuning) | 2013 | More than 100 million years |
| NIST-F2 | 2014 | About 300 million years |
| NIST-F4 | 2025 | About 140 million years (uncertainty basis) |
NIST-F2 needed a chamber cooled to about minus 193 degrees Celsius to hit its number, which cut down on stray radiation interfering with the cesium atoms. That cooling requirement made it expensive to keep running continuously, and NIST retired it in favor of newer, more practical designs within a year of its debut, a reminder that the most accurate clock on paper isn’t always the one that ends up doing the daily work.
Optical Clocks: A Different League Entirely
Cesium fountains measure a microwave transition. Optical lattice clocks, built around atoms like strontium and ytterbium, measure a transition that oscillates in the visible or near-visible light spectrum, hundreds of trillions of times per second rather than billions. That higher “tick rate” is what lets them reach fractional uncertainties around 10⁻¹⁸, roughly ten thousand times better than a cesium fountain. JILA’s strontium lattice clock hit 5 billion years of theoretical stability in a 2014 Nature paper, and an updated version reached roughly 15 billion years by 2015, edging past the estimated age of the universe itself. The international metrology community is now working toward redefining the second again around one of these optical transitions, with a decision expected by 2030.
Everyday Clocks, For Comparison
Almost nothing outside a national lab operates anywhere near these numbers, and putting ordinary timepieces on the same scale makes the gap easier to feel.
| Clock Type | Typical Drift |
|---|---|
| Inexpensive quartz watch | A few seconds per month |
| High-grade quartz movement | About 5 to 10 seconds per year |
| Commercial rubidium oscillator | Roughly 1 second per few thousand years |
| Cesium fountain (lab standard) | 1 second per tens to hundreds of millions of years |
| Optical lattice clock (record) | Less than 1 second over the age of the universe |
Consumer novelty items sit somewhere between rubidium and cesium fountain territory. A rubidium optical clock built for the watch market has been marketed at under 0.1 nanoseconds of daily error, equivalent to less than a second of drift per million years, a genuinely lab-grade number squeezed into a piece of jewelry, though it still trails a proper cesium fountain by two orders of magnitude.
The Drift That Has Nothing to Do With Manufacturing
Even a theoretically perfect clock would still disagree with another theoretically perfect clock if the two sat at different altitudes or moved at different speeds, because relativity changes clock rates depending on gravity and motion. GPS satellites make this unavoidable and measurable every single day.
| Relativistic Effect | Cause | Effect on Satellite Clock |
|---|---|---|
| Special relativity | Satellite moves at about 3.87 km/s relative to the ground | Runs about 7 microseconds per day slower |
| General relativity | Weaker gravity at 20,200 km altitude | Runs about 46 microseconds per day faster |
| Net effect | Combination of both | Runs about 38 microseconds per day faster than a ground clock |
Left uncorrected, that 38-microsecond gap would translate into a position error of roughly 10 to 11 kilometers accumulating every single day, since GPS calculates location from the travel time of a radio signal moving at the speed of light. Engineers solve it before launch by tuning each satellite’s onboard oscillator to run slightly slow on the ground, at 10.22999999543 MHz instead of the nominal 10.23 MHz, so that once relativity speeds it up in orbit, it lands exactly on frequency.
In July 2020, researchers took the concept out of the satellite and into a building. A team led by Hidetoshi Katori placed two portable strontium optical lattice clocks about 450 meters apart in height, at the base and observation deck of the Tokyo Skytree tower, and directly measured the tiny difference in how fast time passed at each elevation, confirming general relativity’s prediction using instruments a person could carry up a staircase.
The most accurate clocks ever built haven’t just gotten better at measuring a second, they’ve proven that a single, universal “now” shared by everyone on Earth was never really true to begin with. worldtimedata
The Clock That’s Actually the Unreliable One
For most of history, people assumed the sky kept better time than any machine could. That assumption has now fully reversed. Since 1972, whenever UT1, the time standard based on Earth’s actual rotation, has drifted more than 0.9 seconds away from atomic time, the world has inserted a leap second to realign them. It is Earth’s spin that wanders, not the atomic clocks.
In November 2022, the world’s metrology authorities voted to stop this practice by 2035, letting atomic and astronomical time quietly diverge by more than a second for the first time in over half a century. The reason was practical rather than scientific: leap seconds had become a headache for modern computing systems, a bigger problem than the tiny astronomical mismatch they were meant to fix. The planet, it turns out, was always the less punctual party in this relationship.
