The Neutron Lifetime Puzzle

9.9 Seconds Apart

Los Alamos National Laboratory’s UCNτ holds ultracold neutrons in a magnetic field and counts how many are left after a while; its final result, published in 2025, is 877.83 s, with a statistical error of ±0.22 s and a systematic error of +0.20/−0.17 s. The National Institute of Standards and Technology (NIST) sends a beam of cold neutrons through a proton trap and counts the protons the decays make; its 2013 result is 887.7 s, ±1.2 s statistical and ±1.9 s systematic.

The two differ by 9.87 s; their errors combined come to about ±2.27 s, and the gap is 4.4 times that (this page’s arithmetic). Averages tell the same story: the Particle Data Group’s 2025 value, which averages only bottle results, is 878.4 ± 0.5 s; the two proton-beam experiments average 888.0 ± 2.0 s, 9.6 s higher. How many standard deviations depends on which results are compared; the usual figure is about four.

bottle 877.8 s · beam 887.7 s · 9.9 s apart, about 4.4σ (this page’s arithmetic)

A Free Neutron Decays

A neutron is 1.293 MeV heavier than a proton (CODATA 2022). On its own it turns into a proton, giving off an electron and an electron antineutrino: n → p + e⁻ + ν̄ₑ. Less the electron’s mass, each decay releases about 0.782 MeV, shared among the three. Tritium’s beta decay is the same event inside a nucleus; see the plate “Weighing the Neutrino”.

A free neutron’s mean lifetime is about 878 s, or 14.6 minutes; half of a sample is gone after about 10.2 minutes (this page’s arithmetic). A neutron in a stable nucleus does not decay: what the decay would leave behind weighs more than the nucleus it started from, so there is no energy to pay for it. A neutron in a deuteron, for example, is about 1.44 MeV short (this page’s arithmetic from CODATA 2022 masses).

n → p + e⁻ + ν̄ₑ · about 0.782 MeV released · mean lifetime about 14.6 minutes

Why the Seconds Matter

In the universe’s first minutes, protons and neutrons had to wait for the temperature to fall before they could join into nuclei, and while they waited, neutrons decayed. Nearly all that survived ended up in helium-4. A longer lifetime leaves more neutrons and so more helium: by Pitrou and colleagues’ 2018 calculation, each extra second raises the primordial helium mass fraction Yₚ (about 0.247) by about 0.0002, so 9.9 s by about 0.002 (this page’s arithmetic).

The lifetime is also tied to the constants of the weak force. With Czarnecki and colleagues’ 2018 relation τ(1 + 3gA²) = 5,172.0 s, the gA measured by PERKEO III gives 878.4 ± 0.7 s, close to the bottle result (this page’s arithmetic). The measured values of gA do not all agree with one another either.

Yₚ: about 0.0002 per second · τ(1 + 3gA²) = 5,172.0 s · PERKEO III → 878.4 ± 0.7 s

Slow Enough to Bottle

Ultracold neutrons carry less than about 300 neV of kinetic energy and move at under 8 m/s; many materials bounce them back from any angle. Rising 1 m costs 102.5 neV, so a 300 neV neutron climbs about 3 m at most. A neutron is a small magnet, 60.3 neV per tesla, and a strong field pushes away the neutrons of one spin direction (this page’s arithmetic from CODATA 2022).

UCNτ’s ultracold neutrons come from the Los Alamos Neutron Science Center: 800 MeV protons strike a tungsten target, and the neutrons slow down in solid deuterium. A 5.5 T superconducting solenoid lets through only one spin, and a spin flipper turns it into the kind a field pushes away. The trap keeps only neutrons below about 50 neV: about 3.1 m/s, too slow to clear 50 cm.

ultracold: below 300 neV, under 8 m/s · gravity 102.5 neV / m · magnetism 60.3 neV / T

Counting What Is Left

A run goes like this: neutrons come in through a door in the floor for 150 to 300 s; the cleaner sits 38 cm above the floor and for 50 s removes neutrons with too much energy, which would leak out slowly; then it rises 5 cm, the door closes and the neutrons are stored for 20 to 1,550 s; finally the detector is lowered in to count how many are left.

The lifetime comes from a short and a long hold: τ = (t₂ − t₁) ÷ ln(N₁ ÷ N₂), 1 the short one and 2 the long one, each count with its background removed and scaled by the monitors. The 2017 and 2018 runs counted about 38 million surviving neutrons; the 2025 result combines five years of data, 2017 to 2022.

stored 20⁠–⁠1,550 s · a short and a long hold, in pairs · about 38 million in 2017⁠–⁠2018

Cut Open: the Magnet Bathtub

The tub is lined with 5,310 neodymium-iron-boron magnets, each 5.08 cm long, 1.27 cm wide and 2.54 cm deep, with about 1 T at their surface. They form a Halbach array: along the tub each row’s magnetisation is turned 90° from the last, so the field gathers on the inner side and fades within about a centimetre. Neutrons below about 50 neV bounce off this skin of field; the top is open, and gravity is the lid.

The shape is two half-tori: along the tub the floor is an arc of radius 1.5 m; across it, one side rises on a 0.5 m circle and the other on a 1 m circle, meeting smoothly at the bottom. Cut off 50 cm up, the opening is about 2.24 × 1.37 m and the volume about 670 L (692 L by this page’s arithmetic). The asymmetry scrambles the neutrons’ orbits, so the too-energetic ones leave sooner. Ten coils outside add a holding field of about 6 to 12 mT, so that no neutron flips its spin where the field would fall to zero.

5,310 blocks · about 1 T at the surface · 50 cm deep · about 670 L · holding field 6⁠–⁠12 mT · arrows on the cut: each block’s magnetisation, schematic

Lowering the Detector

The detector that counts them is called the dagger: an upright acrylic paddle coated on both faces with zinc sulfide and boron-10, 750 cm² a side. A neutron that meets boron-10 is absorbed and throws out an alpha particle and a lithium-7, which make the zinc sulfide flash; wavelength-shifting fibres carry the light to photomultipliers.

It goes down in steps: 40 s at the cleaners’ height, 20 s in the middle, then 150 s with its edge 1 cm above the floor (the 2021 paper). At the bottom it takes up the stored neutrons with a time constant of about 8 s and an efficiency of about 96 %. The neutrons are counted where they are stored, without draining them out of the trap.

750 cm² a side · down to 1 cm above the floor · time constant about 8 s

Counting What Appears

The beam method turns it round: instead of the neutrons that are left, it counts the protons the decays make. In a stretch of beam of known length, the decays each second equal the neutrons in it divided by the lifetime, τ = N ÷ (dN/dt). So two things must be known: how many neutrons are in that stretch, and how many decay each second.

NIST’s neutrons come from a 20 MW research reactor and are slowed in a liquid-hydrogen cold source at 20 K, to a mean wavelength of about 0.45 nm, about 840 to 930 m/s; nickel-58-coated guides carry them about 68 m. The beam is only 7 to 8.4 mm across; some 20 to 50 million neutrons pass each second, and the trap counts about 4 protons a second.

τ = N ÷ (dN/dt) · about 2⁠–⁠5 × 10⁷ neutrons a second · about 4 protons

Cut Open: the Proton Trap

The beam runs along the axis of a superconducting solenoid, where the field is 4.6 T. On the axis sit 16 gold-coated fused-quartz ring electrodes, each 18.6 mm long with a 26 mm bore. The first three, the door, and the last three, the mirror, are held at +800 V; the 3 to 10 grounded ones between them are the trap. A decay’s proton has at most 751 eV, too little to climb the 800 V walls, and the field winds it round the axis, so it stays.

After about 10 ms the door drops to ground and the protons follow the field lines out. At the door end the lines bend by 9.5° and lead the protons to a silicon detector beside the beam, held at −25 to −35 kV, which counts nearly every one. The trap’s length is changed from run to run: the ends have the same effective length for every trap, so following how the count changes with length removes them.

4.6 T · 16 electrodes · door and mirror at +800 V · a 9.5° bend · −25 to −35 kV

Counting the Neutrons

How many neutrons are in the beam is counted by the neutron monitor downstream of the trap: a 50 mm silicon wafer with a deposit of lithium-6 fluoride 38 mm across. A neutron that strikes lithium-6 breaks it into an alpha particle and a triton, and four silicon detectors count them through precision apertures.

The chance that lithium-6 catches a neutron falls as the neutron gets faster, and so does the chance that it decays in the trap, which grows with the time it spends there. Dividing one by the other, the beam’s spread of speeds cancels out. In 2005 this was the largest uncertainty (2.7 s); in 2013 the Alpha-Gamma device measured the monitor’s efficiency directly to 0.057 %, which moved the lifetime by +1.4 s, to 887.7 s.

⁶LiF, 38 mm across · efficiency ∝ 1/v · calibrated to 0.057 % in 2013

Decades of Readings

In 1980 the Particle Data Group’s average was 917 ± 14 s; in 1990, 888.6 ± 3.5 s; from 2002 to 2012 it stood at 885.7 ± 0.8 s; today it is 878.4 ± 0.5 s. The bottle results since 2005 all lie between 877 and 883 s; the proton-counting beam results of 1996 and 2013 are 889.2 and 887.7 s.

The Particle Data Group now averages only the eight results made with ultracold neutrons; they do not fully agree with one another either, so its error is enlarged 1.8-fold. Including NIST’s beam result gives 878.6 ± 0.6 s, with the factor up to 2.2.

1980: 917 s · 1990: 888.6 s · 2002⁠–⁠2012: 885.7 s · today: 878.4 ± 0.5 s

What Could Be Missing

The possible answers come in two kinds: a decay without a proton, which a bottle counts and a beam misses, or an effect not yet accounted for in one of the methods. To explain 9.9 s the first way, about 1.1 % of decays would have to be invisible (this page’s arithmetic). In 2018 a decay into a dark-matter particle and a photon was proposed; experiments since found neither the photon nor an electron–positron pair, and the measured gA limits invisible decays to about 0.27 % (95 % confidence).

Another idea was that neutrons turn into “mirror neutrons” in NIST’s 4.6 T field; a 2022 search at 6.6 T excluded it. Both methods’ systematics have been checked: in a bottle, too-energetic neutrons that leak out slowly make the lifetime short if they are not cleaned out (about 15 s short with no cleaning at all); a beam must count neutrons and protons exactly, and a 2025 study of charge exchange with leftover hydrogen found it unlikely to matter. None of it is settled.

needed: about 1.1 % unseen · limit from gA: about 0.27 % · photon, electron pairs, mirror neutrons: none found

What Comes Next

At Japan’s J-PARC a beam experiment counts not protons but the electrons from decays, in a time projection chamber with a trace of helium-3; its 2024 preprint gives 877.2 s, ±1.7 s statistical and +4.0/−3.6 s systematic: in line with the bottles, 2.3 standard deviations from the proton-beam average. It can be measured from space too: the Lunar Prospector’s neutron spectrometer gave 887 ± 14 s, not yet enough to choose.

NIST has repeated its beam experiment with an improved apparatus on a new beam line (BL2), not yet published; its successor BL3 is to run from 2027 and aims for better than 0.3 s. UCNτ is moving to “elevator” loading, with 5 to 10 times more neutrons and a goal of about 0.1 s; UCNProβe will count the decay electrons inside a bottle, to compare beta decay with every way a neutron disappears.

J-PARC 2024: 877.2 s (preprint) · Lunar Prospector: 887 ± 14 s · BL3: aiming below 0.3 s

Taken Apart

The detector swings away, and the cleaners, the magnet bathtub and the coils lift one by one out of the vacuum jacket; on the other side the solenoid rises off the proton trap.

Rest the pointer on a part (tap it on a phone) and it lifts on its own, named, while the rest fades.