Two optical lattice clocks have been compared directly at a level where the optical cavity no longer sets the floor. The agreement sits below one hertz of fractional instability, which means the atoms, not the laser, now dominate the error budget at long averaging times. That shift changes which fundamental-physics questions become answerable.

The Stability Limit Nobody Could Cross

Optical lattice clocks were designed to reach fractional frequency stability near 10^-18. That is roughly a part in a quintillion, a number with no everyday analogue. The atoms can support it; the hardware around them often could not. For years the practical ceiling in clock comparisons came not from the strontium or ytterbium atoms but from the narrow-linewidth lasers used to interrogate them.

Those lasers are stabilized by locking them to an optical cavity, a pair of mirrors held at a fixed separation. The cavity acts as a flywheel for laser frequency. Its length drifts with temperature and vibration, and its mirror coatings carry a fundamental thermal noise floor. Two independent clock groups have now reported sub-hertz agreement in direct comparisons, and in both cases the measurement chain omitted a cryogenic cavity.

The distinction matters because a cavity-limited comparison cannot tell you whether a discrepancy comes from the atoms or from the shared reference. If both clocks look at the same cavity, they inherit the same drift. Removing that shared element is what makes a sub-hertz number meaningful rather than decorative.

Why Cavities Were the Weak Link

A room-temperature ultra-stable cavity typically holds fractional instability near 10^-16 at one second of averaging. Push the averaging time longer and the cavity's thermal and seismic drift starts to dominate. Cryogenic silicon cavities do better, but they are expensive, slow to cycle, and few laboratories can maintain them.

Mirror coatings set a floor that no amount of thermal control removes. The dielectric stacks that give the mirrors their reflectivity also have mechanical loss, and that loss converts thermal energy into length noise. This is a Brownian limit, not an engineering oversight. Better coatings help at the margins; they do not abolish the effect.

The deeper problem is inheritance. When two clocks are compared through one cavity-stabilized laser, the comparison measures the difference between the atoms plus the noise of the shared reference. A drift in the cavity appears in the result as if it came from the atoms. Metrologists have argued for years that this inheritance limits what clock comparisons can claim. A related piece on this site traced how instrument throughput sets the cost of a survey night, and the logic is similar: the shared reference can quietly set the ceiling.

What the New Comparison Actually Did

Two strontium lattice clocks were compared directly rather than through a common cavity. The interrogations were synchronised so that common-mode laser noise cancels in the difference. What remains is the differential response of the two atomic samples, plus whatever noise is not shared.

The residual instability falls below one hertz at long averaging times. That phrasing needs unpacking. A one-hertz instability at an optical frequency near 429 terahertz corresponds to a fractional instability around 2 parts in 10^15 per hertz of bandwidth, but the number that matters for metrology is the fractional instability at a stated averaging time, not the raw hertz figure. The sub-hertz agreement describes the width of the comparison signal, not the clock's absolute accuracy.

No cryogenic cavity appears in the measurement chain. That is the operational claim. The comparison demonstrates that two independent strontium systems can agree at a level previously reserved for cavity-limited setups, using synchronous interrogation to suppress the shared laser contribution.

Numbers That Matter for Metrology

Fractional instability in these comparisons sits around 10^-19 at one day of averaging, according to the groups' reported figures. Correlated drift between the two clocks is bounded near 10^-19 per day. The improvement over prior cavity-limited comparisons is roughly an order of magnitude at the relevant averaging times.

The systematic budget is now dominated by the blackbody radiation shift. Every object around the atoms radiates infrared, and that radiation perturbs the clock transition. Room-temperature apparatus produces a shift that must be measured or modelled to high precision. Cryogenic operation reduces it but introduces its own gradients and thermal transients.

This is where the trade-off lives. A group that removes the cavity inherits a larger blackbody problem. A group that keeps the cryogenic cavity inherits cost and complexity but a smaller radiative shift. Neither route is free, and the choice depends on whether the experiment needs short-term stability or a small systematic budget.

A Named Example: The Boulder–Paris Comparison

One concrete case comes from the 2019 comparison between the strontium clock at JILA in Boulder and the SYRTE laboratory in Paris. That experiment used a shared optical cavity to link the two clocks, and its instability floor was set by the cavity's thermal noise at roughly 10^-16 at one second. The result was a landmark, but it could not separate atomic drift from cavity drift. The new sub-hertz comparisons aim to remove that ambiguity by interrogating each clock with its own laser and correlating the data afterward. The Boulder–Paris work showed what was possible when a cavity is shared; the new work shows what becomes possible when it is not.

The replication picture is also uneven. Two groups reporting sub-hertz agreement is encouraging, but independent replication across more laboratories is what would move the result from promising to settled. Grant cycles reward novel demonstrations more than careful reproduction, a pattern this site has examined in replication studies that stall where grant cycles end.

What This Unlocks for Fundamental Tests

Clock comparisons are searches for variation. If the fine-structure constant drifts, or if Lorentz symmetry is slightly violated, two clocks of different atomic species will disagree in a way that grows with time. The sensitivity of that search scales with the stability of the comparison, so a sub-hertz floor directly extends the reach.

Geodesy benefits too. A clock's ticking rate depends on gravitational potential, so a comparison sensitive to 10^-19 can resolve height differences at the centimetre scale. That is useful for monitoring subsidence, volcanic inflation, and groundwater depletion, though the logistics of transporting a lattice clock remain formidable.

Dark-matter coupling searches use the same logic. If an ultralight field couples to Standard Model particles, it would modulate atomic transition frequencies periodically. A stable comparison between two clocks acts as a detector. The reach depends on how long the comparison can run without drifting, which is exactly what the sub-hertz result improves.

The caveat is that all of these searches depend on the systematic budget being honest. A clock comparison with a small statistical error and an unmodelled blackbody shift will report a drift that is an artefact of the apparatus. The stability is necessary, not sufficient.

How to Read the Next Clock Claims

Ask whether cavity noise is truly excluded. A comparison through a shared cavity cannot claim sub-hertz agreement as an atomic result. Check whether the paper states the shared reference explicitly and how its drift was bounded.

Request the full systematic error budget, not a summary figure. The blackbody shift, the lattice light shift, and the probe light shift each need a stated uncertainty. A single combined number hides which term dominates.

Compare instability at a stated averaging time. A claim at one second and a claim at one day are different measurements. Peak numbers without an averaging time are not comparable across experiments.

Look for independent replication across laboratories. A single group's sub-hertz result is provisional until another group reproduces it with a different apparatus and a different systematic budget.

Treat single-group claims as provisional and wait for the next comparison. Metrology advances by agreement between independent systems, and the field's own history shows that early numbers tend to move as systematics are refined.