The Missing Nine Seconds: Why Two Ways of Timing a Neutron Disagree

Introduction

A free neutron, pulled out of an atomic nucleus and left on its own, does not last. Within roughly a quarter of an hour it decays into a proton, an electron, and an antineutrino. That number, the neutron mean lifetime, is one of the most heavily measured constants in physics. It sets how much helium the universe made in its first few minutes, and it feeds directly into tests of the Standard Model.

Physicists measure it two ways. The first watches a beam of neutrons and counts the protons that appear when some of them decay. The second pours very slow neutrons into a trap, waits, and counts how many are left. Both are straightforward in principle, and both have been refined over decades by teams at national laboratories in the United States, Europe, Russia, and Japan.

They do not agree.

The best proton-counting beam measurement, from the National Institute of Standards and Technology (NIST), gives about 887.7 seconds. The best trap measurements, led by the UCNτ experiment at Los Alamos, cluster near 877.8 seconds. The difference is close to nine seconds, about 1 percent, and it has held at roughly four standard deviations through more than a decade of improvements on both sides.

The conventional explanation is that one method, most likely the beam, carries a systematic error that nobody has yet identified. The less conventional explanation is that both methods are right, and they are measuring different things: the bottle counts every way a neutron can disappear, while the beam counts only the decays that produce a proton. If about 1 percent of neutrons decay into something other than a proton, the two numbers would disagree by exactly this much.

Background context

The neutron was discovered by James Chadwick in 1932. Its decay was first observed in the years after the Second World War, once nuclear reactors made intense neutron sources available. Early lifetime estimates scattered over hundreds of seconds. By the 1990s the two modern techniques had matured, and the field expected them to converge.

The beam method sends a cold neutron beam through a magnetic and electric trap. When a neutron decays inside the trap, its proton is caught and later counted. Separately, the experiment measures how many neutrons pass through, using a thin foil that absorbs a known fraction of them. The lifetime follows from the ratio of neutrons present to decays observed. The NIST result, published by Yue and colleagues in Physical Review Letters in 2013, reported 887.7 ± 1.2 (statistical) ± 1.9 (systematic) seconds. The paper's main advance was a direct measurement of the neutron monitor's efficiency using a second, fully absorbing detector, which cut the systematic uncertainty from an earlier 3.2 seconds and nudged the lifetime up by 1.4 seconds.

The bottle method uses ultracold neutrons, moving at only a few meters per second, slow enough to be confined by material walls, magnetic fields, or gravity itself. Researchers fill the trap, hold the neutrons for varying times, then empty it and count the survivors. The lifetime comes from how the survivor count falls with storage time. The main hazard is that neutrons can leak or be absorbed by the walls, so modern traps avoid walls where they can.

The leading bottle experiment is UCNτ at Los Alamos National Laboratory, which levitates neutrons above an array of permanent magnets and uses gravity as the lid. It reported 877.75 ± 0.28 (statistical) +0.22/−0.16 (systematic) seconds in Physical Review Letters in 2021, and a combined analysis including later running in Physical Review C in 2025 reported 877.83 ± 0.22 (statistical) +0.20/−0.17 (systematic) seconds, a total uncertainty under 0.3 seconds.

The Particle Data Group, which maintains the reference values used across physics, now quotes a neutron mean life of 878.3 ± 0.4 seconds in its 2026 listing. That average uses only the eight trap measurements. The beam result is listed but left out, a quiet acknowledgment that the two populations of data cannot be combined cleanly. Even the trap results do not agree perfectly among themselves: the PDG inflates the uncertainty of its average by a scale factor of 1.8 to account for their scatter.

Evidence presentation

The case for a real discrepancy rests on how stubbornly it has held.

In 2018, William Marciano, Andrzej Czarnecki, and Alberto Sirlin compared the two approaches in Physical Review Letters and found a beam average of 888.0 ± 2.0 seconds against a trap average of 879.4 ± 0.6 seconds: a gap of 8.6 ± 2.1 seconds, about four standard deviations. Later trap results only tightened the bottle side. A 2025 overview presented at a Paul Scherrer Institute workshop dedicated to the problem put the gap near 9.8 ± 2.0 seconds.

Several features make simple explanations harder than they first appear.

• The bottles broadly agree across very different designs. UCNτ uses a magnetic trap with gravitational confinement. Pattie and colleagues, in an earlier UCNτ run published in Science in 2018, reported 877.7 seconds. Material-walled gravitational traps run by Anatoly Serebrov's group in Russia and at the Institut Laue-Langevin in France gave values from roughly 878 to 881.5 seconds. Different wall losses, different geometries, broadly the same answer, although the spread is wider than the quoted errors, with the 2018 Serebrov value the main outlier.
• The beam side has few independent entries. Nearly all of the beam weight comes from the NIST apparatus. Its central value has been re-examined repeatedly, and the 2013 recalibration moved it slightly further from the bottle value, not closer.
• The size of the gap is suggestive. Nine seconds out of roughly 880 is almost exactly 1 percent. That is the size of branch a hidden decay channel would need.

The disappearance-versus-appearance distinction is the heart of the matter. A bottle measures total lifetime, meaning every route by which a neutron vanishes. A proton-counting beam measures partial lifetime, meaning only decays that yield a proton. In the Standard Model those are essentially the same, because almost every neutron decay produces a proton. If they are not the same, the beam would read long and the bottle would read short. That is the direction of the observed gap.

There is also a third, much cruder line of evidence from space. Neutrons knocked out of planetary surfaces by cosmic rays decay on their way up to orbiting spacecraft, so the flux a spacecraft sees depends on altitude and on the neutron lifetime. Wilson and colleagues used data from NASA's MESSENGER flybys of Venus and Mercury to estimate 780 ± 60 ± 70 seconds in Physical Review Research in 2020, then used archival Lunar Prospector data to reach 887 ± 14 +7/−3 seconds in Physical Review C in 2021. The uncertainties are far too large to take sides, but they show the lifetime can be measured by a method sharing no hardware with either laboratory technique.

Contradictory evidence complicates the picture in both directions.

The most important recent entry comes from Japan. A team at the J-PARC spallation source measures the beam lifetime differently, by counting decay electrons in a gas chamber instead of trapping protons. An early result, published in 2020, was too imprecise to matter, at 898 ± 10 +15/−18 seconds. An improved analysis posted by Fuwa and colleagues in December 2024 reports 877.2 ± 1.7 +4.0/−3.6 seconds, consistent with the bottles and about 2.3 standard deviations from the proton-counting beam average. If that holds up, it suggests the problem is specific to how protons are counted, not to beams as such. It is, however, still an unpublished preprint, its systematic uncertainty is about four seconds, and it has already been challenged. A July 2026 analysis by Desai in the European Physical Journal A pointed to large chi-squared values and an apparent dependence of the J-PARC lifetime on gas pressure, and argued that pressure-dependent detector effects could plausibly explain the result.

Decay correlation measurements give a second indirect check. The Standard Model ties the lifetime to the ratio of the neutron's axial and vector couplings, a quantity measured separately. The PERKEO III experiment reported that ratio with high precision in Physical Review Letters in 2019, and its value points toward the shorter, bottle-like lifetime. A competing measurement from the aSPECT experiment leans toward the beam, so this route has not settled the question either.

Alternative perspectives

Mainstream view: an unidentified systematic error. Most physicists working on the problem expect an instrumental explanation, and the beam method is the usual suspect. It depends on two absolute counts, neutrons and protons, each with its own efficiency. A small, unmodeled loss of protons before detection would make the lifetime look long. Serebrov's group proposed in 2021 that protons could be lost through charge exchange with residual hydrogen gas in the NIST trap. NIST researchers tested that directly, and a 2025 study by Caylor and colleagues in Physical Review C concluded the NIST result was unlikely to have been significantly affected. The bottle side has its own suspects, including neutrons in marginally stable orbits that leak out slowly, but UCNτ was designed to remove them and reports its largest remaining systematic at a fraction of a second.

Dark decay. In 2018 Bartosz Fornal and Benjamín Grinstein at the University of California, San Diego proposed in Physical Review Letters that about 1 percent of neutrons decay into a dark matter particle. In the version where the decay emits a photon, keeping ordinary matter stable requires the dark particle's mass to fall in a very narrow window, between 937.900 and 938.783 million electron volts, just below the neutron itself. The idea was testable almost immediately, and it was tested within months. Tang and colleagues at Los Alamos looked for that photon and excluded a branch large enough to explain the anomaly at more than four standard deviations. The UCNA collaboration ruled out the version producing an electron-positron pair as the dominant channel. Astrophysics added the sharpest objection: Baym and colleagues and McKeen and colleagues showed that, in the simplest version, neutrons converting to such a particle inside neutron stars would soften them so much that no star above about 0.7 solar masses could exist. Astronomers routinely observe neutron stars near two solar masses.

Those results closed the simplest models but not all of them. Fornal's 2023 review notes that dark sectors with self-interactions, or decays into several dark particles, can still accommodate a 1 percent branch and heavy neutron stars. Critics regard those extensions as increasingly contrived.

Mirror neutrons. Zurab Berezhiani suggested neutrons might oscillate into a hidden mirror copy, with the effect enhanced in the strong magnetic fields of beam experiments. A dedicated search at Oak Ridge National Laboratory by Broussard and colleagues excluded that mechanism as the explanation in 2022, at 95 percent confidence over the range of parameters the model needed.

An excited neutron. Koch and Hummel proposed in Physical Review D in 2024 that neutrons might be produced partly in a long-lived excited state, which beams and bottles would sample differently. An analysis of UCNτ data, posted as a preprint in 2024, found no evidence for an excited state whose gamma-ray decay lifetime exceeds 139 seconds. That constrains the idea without closing it.

What remains unexplained is simple to state. No one has found an error large enough to move either the NIST beam value or the trap values by nine seconds, and no new-physics model has survived testing without added complexity.

Open questions

The puzzle is unusual in that clear experiments are already under construction to resolve it.

• Does a new beam apparatus reproduce the NIST value? NIST's next-generation BL3 experiment is designed to reach a precision below 0.3 seconds, with running at NIST planned from about 2027 according to collaboration presentations. The NIST research reactor restarted in August 2026 after a shutdown of five and a half years that followed a February 2021 incident, and plans to welcome outside users in late November. A new beam result that stays near 888 seconds with a better understood proton detector would strengthen the case for real physics. One that falls toward 878 would point to a systematic.
• Does electron counting agree with the bottles at higher precision? If J-PARC's result survives peer review and the gas-pressure critique, and its planned upgrade reaches about one second of uncertainty while staying near 877, suspicion will concentrate on proton detection specifically.
• Do independent traps confirm UCNτ? The UCNτ+ upgrade at Los Alamos targets an uncertainty of about 0.1 seconds. PENeLOPE at the Technical University of Munich aims for 0.1 seconds or better, and τSPECT at the Paul Scherrer Institute is in commissioning. These goals are announced, not yet achieved.
• Is there any trace of dark decay products? Any positive signal for an exotic channel near a 1 percent rate, in the lab or in neutron star observations, would change the question entirely.

Readers who want the raw material can start with the Particle Data Group neutron listing, which tabulates every measurement, and with Fornal's review for the theory side.

Conclusion

The neutron lifetime problem is not a fringe claim. It appears in the reference tables every physicist uses, in the form of a beam result that the averagers decline to combine with everything else. Two families of careful experiments, built by different teams with different hardware, give answers about nine seconds apart, and that gap has survived years of scrutiny.

The likeliest resolution remains an overlooked detail in how decays are counted. Early evidence from electron counting at J-PARC leans that way, though that result is unpublished and already contested. Yet the simplest new-physics reading, that 1 percent of neutrons leave by a door we cannot see, fits the size and the direction of the gap uncomfortably well, and not every version of it has been closed.

So the central question stands. Is the missing nine seconds a flaw in the stopwatch, or a message about what neutrons are doing when no detector is watching? The next generation of beams and bottles, now being built, is expected to answer within a few years.