LFM Labs
← Research
2026-04 · Preprint

Entangled Clocks: One-Way Speed of Light

Deposited asEntangled Cesium Lattices in Microgravity: A Quantum-Relativistic Consistency Test of One-Way Light-Speed Isotropy

Light’s speed has only ever been measured precisely as a round trip: out to a mirror and back. Its speed in one direction would need two distant clocks that agree on the time, and setting them already assumes the answer. This paper designs a different kind of check: linked atoms a metre apart, in orbit, testing whether light takes the same time in every direction. It still leans on a shared cable signal, so it does not settle the one-way speed itself. It is a plan for an experiment, not a result.

Prefer the original? Read the full paper · doi.org/10.5281/zenodo.19601871 →

The paper, in plain words11 min read

A question that sounds too simple

How fast does light travel? Most of us learned the answer at school: close to 300,000 kilometres every second. The paper uses the exact figure, 299,792,458 metres per second.

Here is the catch. Every precise measurement sends light out and brings it back. The few one-way attempts rely on clocks that were set using light. The measurement then divides the total distance by the total time.

To turn that average into a speed for each direction, you must assume something. You assume the trip out takes exactly as long as the trip back. No experiment that only exchanges light can check that.

This article walks through a paper that proposes an experiment aimed at that gap. It is a design on paper. Nothing in it has been built or run.

Why every measurement goes there and back

Picture a long corridor. You stand at one end, clap, and time the echo from the far wall. You learn how long the round trip took. You do not learn how that time splits between the way out and the way back.

Say the sound took a third of the time going out. The echo would arrive at the same moment. Your stopwatch cannot tell the two cases apart.

“Fine,” you say, “put a clock at each end.” Now you can note when the light leaves A and when it reaches B, and subtract. But that only works if the two clocks agree on what “now” means.

The usual way to make two distant clocks agree is to send a signal between them. The speed of that signal is what you were trying to measure. You would be using the answer to find the answer.

A · ROUND TRIPOne clock sends light out and gets it back.ONE CLOCKout: time unknownback: time unknownMIRRORThe clock reads only the total: out + back.B · ONE WAYTwo clocks: one stamps the start, one the arrival.CLOCK ACLOCK Blight, A to BA stamps the startB stamps the arrivalAssumed: both clocks agreeTHE HIDDEN ASSUMPTION
Figure 1A round trip needs one clock, and it only ever reads the total. A one-way trip needs two clocks, and the result depends on an assumption about whether they agree.

Round-trip tests are very precise. The paper cites laser tests that reach one part in a hundred quadrillion (10⁻¹⁷). But it says these are blind, by design, to a difference in one direction only. The few one-way tests that exist are limited by the very agreement they test.

Einstein’s agreement, and a dial called ε

In 1905 Einstein needed a way to say that two distant events happen “at the same time.” His recipe: send a flash from A to a mirror at B and back. Then declare that it reached B exactly halfway through the trip. That is a choice about how to label time. It is not something anyone measured.

The philosopher Hans Reichenbach, in a 1958 book the paper cites, showed the choice is not forced. Call the share of the round trip spent going out ε, the Greek letter epsilon. Think of it as a dial. Einstein’s setting is ε = ½: half out, half back.

Any other setting gives the same round trip. Light-signal methods cannot tell it apart from Einstein’s.

CLOCK ACLOCK BTIMEemitreturnsame timeε largeroutbound slowerε = ½Einstein’s pickε smalleroutbound fasterdistance d = 1.000 mε is the share of the round trip spent going out.ε0½1
Figure 2Three light paths with the same departure and the same return. They differ only in when the flash is said to touch B, which is what the ε dial sets. Einstein’s choice is the middle one.

Choosing ε = ½ is the same as assuming light is equally fast in every direction. No experiment that only exchanges light can say whether that assumption is true.

The idea: a pair of atoms linked as one

Imagine two coins that always land opposite ways: when one shows heads, the other shows tails, however far apart you throw them. Neither coin has a fixed result before it lands. Yet you know the two results will be opposite. Quantum particles can be linked like this. The link is called entanglement.

The paper uses two cesium atoms. Each one works as a clock. An atom holds a steady beat at 9,192,631,770 ticks a second. That beat is how the second is defined.

Where an atom is in its beat is called its phase. Think of the position of a clock hand. Reading the hand tells you how much time has passed.

The paper links the two atoms in a special way. As time passes, each atom ticks normally, but the pair is set up so that, while both tick at exactly the same rate, their ticking cancels out in the pair’s shared state. So the pair stays put while both atoms sit in the same conditions. The paper treats the pair as one frozen reference.

Here is the plan. Keep the pair together, then carry one atom a metre away. Send a single particle of light from station A to a detector beside atom B. Compare the two clock hands to see how long the light took.

A fair objection: carrying a clock across a room is the old trick, with problems of its own. The paper says the pair avoids them. During the move both atoms feel the same conditions, so the pair’s state does not change. Only differences between their surroundings matter, and the paper counts those below.

Both stations still read their clock hands against one shared steady signal, sent down a rigid cable. What the pair adds is a way to spot a bad run. If the link broke, the run shows it, and that run is thrown out afterwards.

Entanglement does not replace the shared reference. What it adds is a way to tell when a run has gone wrong, so bad runs can be thrown out.

How the experiment would run

Making the pair

First, laser light makes a pattern like an egg box. Each dip holds one very cold atom almost perfectly still. The dips sit a few micrometres apart. A micrometre is a millionth of a metre. The paper calls this an optical lattice.

To link the atoms, the paper proposes what it calls a “metronome” method. A laser lifts an atom into a very high-energy state. The atom swells up, and it stops its neighbours from joining it. This reaches about 8 micrometres.

Start at the rightmost atom. Sweep the laser settings from right to left. Each step pushes the link one atom further along, like a row of dominoes falling in turn.

THE METRONOME: HOW THE PAIR IS MADEEach step pushes the entangled region one site further.SITE 1 · STARTSTARTMIDWAYLATER………radius ≈ 8 µmthe sweep moves leftwardEstimated pair quality above 0.95 over 20 sites, under 5 µs.waitingentangledpushing nowno-join zone
Figure 3Each step pushes the link one atom further left; three moments are shown. The dashed circle shows how far an excited atom keeps its neighbours from joining in.

The paper estimates this makes a good pair, with a quality score above 0.95, across a 20-atom row in under 5 microseconds. The score runs from 0 to 1. It says how closely the pair you made matches the pair you wanted.

Why in orbit

Gravity gets in the way. On the ground, atoms sag in their traps. And a clock lower down ticks a little slower than one higher up. The paper puts the difference at about one ten-quadrillionth of the tick rate (1.09 × 10⁻¹⁶) for two clocks one metre apart.

On the ISS, both stations float at the same gravitational level. The station is about 400 kilometres up, which leaves a tiny difference. The paper puts it at about 3.3 × 10⁻²¹ radians of clock hand. The platforms in mind are the Cold Atom Lab on the ISS and its planned successor, BECCAL.

The two stations

Two stations sit 1.000 metre apart on a rail. Clock A holds atom A and a source of single particles of light, called photons. Clock B holds atom B and a very fast light detector. It stamps each photon’s arrival to within ten trillionths of a second. The stamp still shakes a little from shot to shot. That shake is called jitter, and it becomes the main source of noise.

TOP VIEW · NOT TO SCALEMICROGRAVITY PLATFORM · ISS / BECCALentangled pairCLOCK ACLOCK Bemittergentle readoutdetectordetector <10 psphoton, 3.336 nsSTARTatom B moves, ~30 msrigid cable: shared signaldelay about 5.05 ns1.000 mbaseline length watched to about 5 nm1231Vibration isolation under the trap mount2µ-metal magnetic shielding3Photon collected sideways, across the rail
Figure 4The layout seen from above. Light carries the timing signal between the stations; the cable carries the shared signal that both atoms are read against.

The moving part is atom B. The paper carries it the full 1.000 metre by steering the trap. The atom arrives as calm as if it had moved very slowly. The top speed is about 30 metres per second. The trip takes about 30 milliseconds.

This step needs the most care. The longest move the paper cites is a rubidium atom carried 77.5 micrometres in 87 millionths of a second. It arrived in good shape, with a score of 0.98 to 0.99. One metre is about 13,000 times further. The paper’s simulations give a score above 0.98 at 1.000 metre. These are estimates and simulations, not measurements.

One measurement, step by step

The paper lists eight steps for a single trial. They are numbered in Figure 5. First the pair is made and the shared signal connected (steps 1 and 2). Then atom B travels its metre (step 3).

Next comes the fast part, inside about five billionths of a second. Atom A’s hand is read, a photon is sent and caught, and atom B’s hand is read (steps 4 to 7). Last, offline, trials where the link failed are thrown away, and the rest are analysed (step 8).

setup: move takes ~30 mssteps 4–7: about 5 nslaterSCALE BREAKSCALE BREAKSHARED SIGNAL2ATOM A1pair prepared4read atom APHOTON5photon sent6detected3.336 nsATOM B3moves 1.000 m, ~30 ms7read atom BANALYSIS8drop bad runs, then find Δt
Figure 5The slow part, the 30-millisecond move, comes first. The measurement itself fits in about five billionths of a second. Analysis happens later, offline.

Each station records its atom’s clock hand against the shared signal. The gap between the two hands tells you how long the light took. For a photon crossing 1.000 metre, that is about 3.336 nanoseconds, and a nanosecond is a billionth of a second.

The two records meet only afterwards. How fast they travel does not matter, because both were written down on the spot. One honest detail: the final reading of each atom does destroy the link, as in any standard test of this kind.

What could go wrong, and how it is handled

A big known push

If light speed is the same in every direction, the measured time should not change as the rail points in different directions. The ISS helps. It circles the Earth in about 92 minutes, so the rail sweeps through many directions every orbit. If light had a favourite direction, the time would rise and fall like a wave over each orbit. A flat line would mean none.

One large effect is in the way. The station is moving and turning, and that shifts the clock hands. The paper puts it at up to 5.0 mrad, bigger than the 1 mrad target.

But it is predictable. GPS gives the station’s speed. Star trackers and gyroscopes give its direction. The paper subtracts the predicted shift point by point. It estimates the leftover uncertainty at about 25 billionths of a radian.

BEFORE CALIBRATIONTurning-motion shift: 5 mrad × cos θ.MRAD+50−5peak 5 mrad2.6 MRAD10⁵ TRIALS0°90°180°270°360°baseline angle θ to orbital velocity: one orbit, 92 minAFTER CALIBRATIONThe predicted shift is subtracted point by point.a ripple would mean direction mattersflat line means no favourite directioncalibration leaves about 25 nrad0°180°360°baseline angle to the preferred direction (illustrative)Ripple height is for shape only; the paper gives no size.
Figure 6Top: the large turning-motion shift the paper expects, with the size of one campaign’s uncertainty beside it. Bottom: what is left after that shift is subtracted. A flat line supports light being the same in every direction (see the cable caveat); a ripple with two humps per turn would point away from that.

The noise

The paper’s list of errors has a clear message. Timing jitter is the problem. The detector’s shake and a matching shake in the clock’s start signal each become 578 mrad of error in a single shot.

With the smaller terms added, the total is about 817 mrad per shot. One shot, on its own, says almost nothing.

The answer is to repeat. Averaging the detector shake over 100,000 independent trials leaves 31.6 femtoseconds. That is a millionth of a billionth of a second. Against the 3.336 nanoseconds the light takes, it is about one part in a hundred thousand. The paper calls this the 10⁻⁵ level per campaign.

In hand terms, the total error falls to about 2.6 mrad after 100,000 trials, and about 0.26 mrad after 10 million. The 1 mrad figure is a target for a whole campaign, not one shot. The paper expects a flat grid of atoms, swept diagonally to make many pairs at once, to supply extra trials. That takes the reach to about 10⁻⁶.

PHASE ERROR PER SHOT, LOG SCALE1 mrad targetDetector jitter578 mradTrigger jitter578 mradPhoton path jitter~17 mradMagnetic field~1 mradMove heating~1 mradLight shift, still~1 mradLight shift, moving~0.2 mradVibration~8.7 µradTurning, corrected~25 nradTidal gravity~10⁻²¹ radTotal817 mrad10⁷: 0.26 mrad10⁵: 2.6 mrad10⁻²¹10⁻¹⁵10⁻⁹10⁻³RADIANSper shotafter 10⁵ trials (10⁵ and 10⁷ for the total)
Figure 7The paper’s own error table on a log scale. Only the two timing terms matter per shot, and only averaging brings the total below the 1 mrad target.

And the paper is plain about where this stands. Round-trip tests reach 10⁻¹⁷, and this one “cannot compete on raw sensitivity.” Its value is a different way of timing, with a different set of possible errors.

What each result would mean

The paper sets out three outcomes. First, the time may stay the same at every angle once the known effects are removed. That supports the idea that light has the same speed in every direction. The paper expects this, and says it would reinforce the standard picture.

Second, a steady shift may remain that follows the direction. The paper says this “may indicate departure” from the standard picture. It calls such a result extraordinary, and says it would need independent confirmation.

Third, the readings may scatter widely. The paper reads that as a sign the link between the atoms was lost, or the clock readings were too coarse. The run is then inconclusive.

Measure Δt at many angles,subtract the known shift.QUESTION 1Is the scatter small enough?YESNOQUESTION 2Does the leftover vary by angle?YESNOUniformSupports thesame speed inevery direction.Expected.DirectionalMay indicate adeparture fromthe norm; needsconfirmation.InconclusiveWide scatter:the pair lost itslink, or readingswere too coarse.Caveat: if light and the cable respond to directionalike, as in any theory where all electromagneticsignals respond alike, a real effect cancels out.
Figure 8Which branch the data lands on depends first on whether the noise is low enough, then on whether the leftover changes with angle. The note below the first outcome is the paper’s own blind spot, covered next.

There is a catch, and it needs a plain explanation. The shared signal travels down a cable, where signals move at about two thirds of the speed of light. The one-way speed along that cable has the same open question as light itself.

Suppose the photon and the cable were affected by direction in exactly the same way. Then the two effects would cancel, and a real signal would be missed. The paper says this is what you would expect from any theory in which all electromagnetic signals respond to direction alike. It calls this a blind spot.

So the experiment tests the question “self-consistently.” It assumes light and cable respond differently. The paper calls this a weaker but well-defined claim.

What this paper does not claim

  • It reports no experiment. Every figure for this design’s performance is a calculation, an estimate or a simulation. The paper says the main parts, such as the detectors and short-range atom moves, have each been shown separately. Putting them together is “nontrivial but plausible.” The metronome method is the paper’s own proposal, and it has not yet been demonstrated.
  • It does not settle the ε dial. The paper says the method does not settle Reichenbach’s choice. The paper says it is “best interpreted” as a test of whether light is the same in every direction. It is not a direct measurement of one-way speed.
  • It cannot beat the round-trip tests. They sit at 10⁻¹⁷. This design aims at 10⁻⁵ per campaign of 100,000 trials, and about 10⁻⁶ with 10 million.
  • It does not fit today’s space hardware as it stands. The ISS Cold Atom Lab and BECCAL are built around rubidium and potassium, not cesium. The paper says so. It proposes two stages: a first proof of principle with rubidium, then a dedicated cesium module for the definitive test.
  • It used automated checks, and says so. The paper acknowledges AI-assisted technical audits and a software library for atom physics. These were used to verify its numbers and references.

The roadmap has three phases. Phase I, on the ground, tests making the pair and a short move. Phase II runs the full 1.000-metre chain on the ground with rubidium. Phase III flies cesium on the ISS.

Beyond that the paper lists possibilities, not promises. One is rails over 10 metres long, which would push sensitivity toward 10⁻⁷.

If you keep one thing, keep this. A metre, a photon and a pair of linked atoms could let us test whether light behaves the same in every direction. The route does not set the clocks with light. It still leans on the cable that carries the shared signal. The paper is the design of that test. The answer still has to be measured.

The full paperdoi.org/10.5281/zenodo.19601871 →

The DOI is the citable artifact and Zenodo holds the PDF. This is the concept DOI, so it always resolves to the newest deposit.

Also from the lab

lfmlabs.com — a polymathic laboratory →