In June 1955, at the National Physical Laboratory in Teddington, the steadiest timekeeper on Earth began to tick.
It was not the first atomic clock — that title belongs to an earlier machine in Washington, and we will come to it. It was the first accurate one, and the first built to keep time: a beam of caesium atoms disciplining a quartz oscillator, good to the equivalent of about one second in 300 years.
And the moment it ran, it had a problem no clockmaker had ever faced. A clock that good had nothing to be checked against. Louis Essen, who built it with Jack Parry, wrote it plainly a decade later: the standard “was much more precise than any astronomical unit and could be tested only in terms of an atomic unit.”
So how was it calibrated? Against the sky it was replacing — in units that were worse than the clock itself. The job took three years, two laboratories an ocean apart, and the Moon. And it ended with the measurement being retired into a definition.
First, which first?
On 12 August 1948, Harold Lyons at the US National Bureau of Standards ran a clock whose regulator was a cloud of ammonia molecules. It was a genuine atomic clock, unveiled publicly in early 1949 — and it was no more accurate than the conventional clocks of the day.
Nor was Teddington’s even the first caesium beam to operate. The Bureau had already demonstrated the potential of caesium as a frequency reference. The claim the NPL machine holds is precise: the first accurate caesium clock, and the world’s first operational source of atomic time.
The machine itself was a partnership. The caesium atoms did not tick out seconds; a quartz oscillator did the ticking, as quartz always does, drifting with temperature and manufacture like any physical object. The caesium resonance was the referee: any drift pulled the electronics off resonance and triggered a correction, steering the quartz back. Essen was so pleased with the first results that he dragged the NPL’s director to the laboratory, in his words, “to witness the birth of atomic time”.
The first calibration was a decision
The clock needed a number: how many cycles of the caesium transition make one second? In June 1955 there was only one second available to ask — the astronomical second of the day, kept by the Royal Greenwich Observatory and broadcast as time signals.
So that was the first calibration. The frequency was taken as 9,192,631,830 cycles per second, the value obtained against the Greenwich second of June 1955, and an atomic time scale was maintained with NPL’s quartz clocks. The atomic second was also disseminated through the UK’s radio time broadcasts.
Notice what kind of act that was. Not a discovery — a convention. The number was chosen so the atomic unit equalled the astronomical second then in use, and the two scales would start life in agreement. Physics World records Essen describing 3 June 1955 as “the death of the astronomical second and the birth of atomic time”. The obituary is premature by twelve years. The astronomical second had one act left.
The astronomers move the goalposts
While Essen was presenting the new clock, the International Astronomical Union, meeting in Dublin in August 1955, recommended a different fundamental unit of time: the ephemeris second, defined by the Earth’s motion round the Sun. In 1956 the International Committee for Weights and Measures redefined the second to match it; the 11th General Conference on Weights and Measures formalised the change in 1960. The second of Universal Time was no longer the fundamental unit.
On paper this was an upgrade. The ephemeris second was constant by definition — an ideal second, immune to the Earth’s wobble. In practice it was worse than the clock it was meant to govern: a unit “constant by definition but available to useful accuracy only after several years”. The world’s steadiest clock now ticked in the wrong unit, and the only instrument that could deliver the legal second was a camera.
Reading time off the Moon
At the US Naval Observatory in Washington, William Markowitz and R. Glenn Hall had been photographing the Moon since June 1952 with the dual-rate Moon position camera. The method: photograph the Moon against the background stars at a precisely known Universal Time, then enter the observed position into the Improved Lunar Ephemeris, which tabulates where the theory says the Moon should be at each moment of Ephemeris Time. Position in, time out. The difference between the two time scales — ΔT, the accumulated divergence between the ideal gravitational clock and the rotating Earth — is what the observations yield.
The two laboratories split the work. Markowitz and Hall supplied Ephemeris Time from the Moon; Essen and Parry, linked across the Atlantic by time signals, counted caesium cycles. The comparison ran from mid-1955 to early 1958 in the paper’s own accounting — 1955.50 to 1958.25 — and the arithmetic was done in two steps. First, the caesium frequency in terms of the astronomical second, using star observations and time signals. Then the Moon-camera data converted that frequency into ephemeris-time seconds.
Direction matters, because the obvious picture is wrong. Nobody in Teddington waited for a moon photograph to steer the oscillator. The caesium ran continuously; the Moon observations converted the unit — cycles per astronomical second, into cycles per ephemeris second. The Moon was not setting the clock. It was translating the clock’s language.
The number
In August 1958 the four authors published the result. The transition frequency of caesium at zero magnetic field: 9,192,631,770 ± 20 cycles per second of Ephemeris Time, quoted at a mean epoch of 1957.0.
The uncertainty deserves its original units. It is ±20 cycles per second — the internal estimate was ±10, doubled to ±20 to cover the systematic risks of a comparison chain running from Teddington to Washington to the Moon. The number is what it is; quote it as the authors did.
Now put the two numbers side by side: 9,192,631,830 cycles per astronomical second, from June 1955. And 9,192,631,770 cycles per ephemeris second, from 1958. Both are correct. The sixty-cycle gap between them is not an error being fixed. It is the difference between the two seconds themselves — the irregularity of the Earth’s rotation, showing up as a discrepancy in a frequency. The calibration’s most important output was a measurement of the wobble it was escaping.
The same data carried a quieter check. The Earth’s rotation, the comparison showed, is slowing — the shortfall grows by roughly 0.16 seconds per year, each year — matching what the caesium standard had already indicated on its own. The atom and the Moon agreed about the Earth. “The agreement between the moon camera and cesium is satisfactory,” the authors wrote, with 1958 restraint. There was no third referee; each was the other’s only check.
Even the epoch was a confession. The result was quoted at a mean epoch because, the paper says, “there is a possibility that the gravitational and atomic time scales may not be the same, and may change secularly”. Whether atomic time and gravitational time were even the same time was, in 1958, an open question.
The measurement becomes a definition
The standards bodies then ran the same playbook slowly. In October 1964 the International Committee on Weights and Measures assigned the frequency 9,192,631,770 hertz to the caesium transition — for temporary use, pending a new definition. Then the 13th General Conference on Weights and Measures made it the definition. Its 1967 resolution reads: “The second is the duration of 9 192 631 770 periods of the radiation corresponding to the transition between the two hyperfine levels of the ground state of the caesium 133 atom.”
The number was not chosen for roundness. It was chosen because the 1955–1958 comparison had found it — which meant the atomic second was made equal, by construction, to the ephemeris second it replaced. The measurement froze into the unit.
At that instant the questions inverted. Before 1967: how many caesium cycles are in a second? — an empirical question, answered with an uncertainty. After: exactly 9,192,631,770, forever, because that is what a second is. Ask instead how long the Earth takes to turn, and the uncertainty now belongs to the Earth. In 1997 the definition was explicitly idealised — a caesium atom at rest, at absolute zero — a clock that exists in no laboratory. The error bars did not disappear. They migrated from the clock to everything around it.
What your phone inherits
The number in every phone on Earth is the 1955–1958 Moon-camera experiment, digit for digit — a measured relation between a beam of atoms in Teddington and the orbit of the Moon, promoted from the best available answer to the answer. When the defined second and the rotating Earth drifted apart, the world bolted on corrections; that repair job is the leap second’s story, and it is now being retired.
Your stopwatch sits one step below all of this, and it is a good place to sit. A quartz oscillator, as we saw when two stopwatches drift apart, is an artefact that must be checked against something outside itself. Caesium is the thing outside. When you press start on TiCaNo Stopwatch, the seconds it counts are defined ones — each exactly 9,192,631,770 cycles of an invisible atom, a number that was once a three-year argument between two laboratories and the Moon.
The counting is the easy part now. The first accurate atomic clock was the last clock that ever needed the sky.
Sources
- Frequency of Cesium in Terms of Ephemeris Time — Markowitz, Hall, Essen & Parry, Physical Review Letters 1, 105 (1958), the primary paper: method, interval 1955.50–1958.25, adopted value ±20 cps, epoch caveat, deceleration cross-check
- Time Scales — L. Essen, Metrologia 4 (1968), the retrospective: the bootstrap paradox in Essen’s own words, the 9,192,631,830 convention
- An Atomic Standard of Frequency and Time Interval: A Cæsium Resonator — Essen & Parry, Nature 176, 280 (13 August 1955)
- Atomic Standards of Frequency and the Second of Ephemeris Time — Nicholson & Sadler, Nature 210, 187 (1966), the 1964 temporary assignment
- Resolution 1 of the 13th CGPM (1967) — BIPM, the definition of the second, the 1997 note, the abrogations
- The second (s) — National Physical Laboratory, on the first operational caesium clock and “one second in 300 years”
- A Brief History of Atomic Time — NIST, the Lyons ammonia clock and the Markowitz–Essen division of labour
- Obituary: Louis Essen — The Independent (1997), the RGO calibration, the Dublin IAU meeting, the NBS priority nuance
- Atomic time was born 60 years ago today — Physics World (2015), the 3 June date and Essen’s description, on secondary attribution
- Essen’s clock — Chemistry World, the servo-loop architecture and the “birth of atomic time” anecdote
- R. Glenn Hall (1921–2004) — BAAS, the USNO side: the two-step calibration and the period 1955.50–1958.25
- Milestones: First Atomic Clock, 1948 — IEEE History Center, the ammonia clock’s claim to “first atomic clock”