To test the finest clock, scientists built it a twin

A clock has become so sensitive that raising it by a few millimetres can measurably change its tick. That is not a defect: Einstein’s general relativity predicts that time runs slightly faster higher in a gravitational field. It is also why judging the clock requires much more than asking whether it agrees with the clock on the wall.

Researchers at Singapore’s Centre for Quantum Technologies built two independent clocks, each based on a single electrically charged atom of lutetium-176. They then made the clocks test each other. After 200 hours of comparison, their frequencies agreed within 5.7 parts in 10¹⁹, the team reports in Nature.

The result addresses a deceptively simple puzzle: if no demonstrably better clock exists, how can anyone know the best clock is right?

An atom supplies the tick

A mechanical clock counts swings of a pendulum. Each lutetium clock instead uses a laser tuned to an energy change inside its trapped ion. The laser light has a wavelength of 848 nanometres, just beyond the deepest red visible to human eyes. Its oscillations provide the regular beat.

Atoms of the same isotope should share the same internal transition. But a working clock is also a physical apparatus sitting in a warm, magnetised and imperfect world. Temperature, magnetic fields, motion, laser light, collisions with stray gas and gravity can all shift its measured frequency.

The researchers therefore made an uncertainty budget for each clock: a quantified assessment of how much every known effect could move its tick after corrections. The two clocks had evaluated fractional uncertainties near 1 × 10⁻¹⁹. That is accuracy—the estimated risk that a clock’s frequency is systematically offset.

Stability is different. It describes how much repeated readings wander and how quickly averaging can reveal the underlying frequency. A clock can be stable but consistently wrong, or accurate on average but noisy over short measurements.

For scale, a fractional uncertainty of about 1.2 × 10⁻¹⁹ converts to roughly one second in 264 billion years:

1 second ÷ 1.2 × 10⁻¹⁹ ≈ 8.33 × 10¹⁸ seconds ≈ 264 billion years

That is an illustration of the ratio, not a claim that anyone watched the clock for billions of years.

Two clocks, several attempts to catch an error

Agreement between twins is useful only if they are not quietly sharing the same mistake. The team separately measured environmental effects in the two systems and deliberately changed the largest correction, caused by magnetic fields, to check that the clock shifted by the predicted amount. Lutetium is particularly helpful because its chosen transition is comparatively insensitive to magnetic fields and room-temperature thermal radiation.

The clocks were compared using correlation spectroscopy. Both ions were interrogated with the same clock laser, allowing common laser noise to be rejected so that the comparison more directly tested the atoms. Eleven measurements added up to 200 hours. The observed relative difference was consistent with zero, with a statistical uncertainty of 5.7 × 10⁻¹⁹ and a systematic uncertainty for the difference of 1.0 × 10⁻¹⁹.

Those numbers should not be collapsed into one extravagant claim. The near-10⁻¹⁹ figures describe each clock’s evaluated systematic uncertainty. The larger 5.7 × 10⁻¹⁹ figure describes how closely the experiment actually resolved their agreement after finite averaging. The comparison remained limited by measurement statistics; it did not prove that either clock was exactly correct.

Gravity makes the test stranger still. The National University of Singapore’s account says the researchers measured the relative heights of the two ions to below one millimetre. A five-millimetre height difference would produce a gravitational frequency shift around the same scale as the comparison’s uncertainty. At this precision, “Are the clocks level?” becomes part of “Do the clocks agree?”

Why distant clocks still matter

A separate European campaign connected seven optical clocks at four national laboratories through stabilised fibre links. Its best comparison—between independently developed ytterbium-ion clocks in Britain and Germany—reached an uncertainty of 7.7 × 10⁻¹⁸, according to the accepted manuscript. That is less precise than the same-site lutetium comparison, but it answers a complementary question: can different laboratories, instruments and clock designs reproduce one another across national distances?

Remote comparisons must also reckon with imperfect knowledge of gravitational height. At sufficiently fine precision, a frequency difference between distant clocks measures their difference in gravitational potential as much as it tests their machinery.

The lutetium result is therefore best understood as an unusually demanding audit. Two clocks agreed; known disturbances were measured; the largest correction was stress-tested; and the remaining statistical limit was stated plainly. It strengthens lutetium’s case as scientists consider optical clocks for a future redefinition of the second—while showing why the final standard must rest on many clocks, places and kinds of evidence.