JUNO’s First Result and the Neutrino Mass Ordering

Which neutrino is the odd one out

Neutrinos come in three mass states, labelled 1, 2 and 3. Oscillation experiments do not measure the masses. They measure the differences between the squares of the masses, and only two of those are independent.

What is known. States 1 and 2 sit close together, separated by a mass-squared difference near 7.5 × 10^-5 eV². State 3 sits far from that pair, separated by about 2.5 × 10^-3 eV², roughly thirty times larger. Solar neutrino measurements fixed the sign of the small splitting, so we know state 2 is heavier than state 1.

What is not. Nobody knows the sign of the large one. If state 3 sits above the close pair, the ordering is normal. If it sits below, the ordering is inverted. Oscillation probabilities in vacuum depend mostly on the squares of these differences, which erases the sign. Recovering it requires either matter effects or a very careful look at interference.

What 59 days already bought

JUNO began taking physics data on 26 August 2025. The collaboration reported two numbers in Nature, from events collected through 2 November.

• sin²θ₁₂ = 0.3092 ± 0.0087
• Δm²₂₁ = (7.50 ± 0.12) × 10^-5 eV²

Both come from 2,379 antineutrino candidates and 59.1 effective days. Both beat the combined precision of everything that came before by a factor of 1.6.

Those numbers are about the detector’s readiness, not its purpose. The engineering behind them, light yield, radiopurity, the full construction timeline, is covered separately in JUNO First Results: Record Precision in 59 Days. What follows here is the physics case: why the mass ordering is so much harder to extract than the two numbers above, and what answering it would settle.

One tension got sharper. Solar neutrino experiments and reactor experiments have long disagreed slightly on the value of Δm²₂₁, at about the 1.5 sigma level. JUNO’s reactor measurement confirms the gap rather than closing it. JUNO can eventually measure both solar and reactor neutrinos in one detector. That removes the cross-experiment systematics making the comparison awkward today.

How you catch a reactor antineutrino

Eight reactor cores at the Yangjiang and Taishan power plants pour electron antineutrinos into the surrounding rock. JUNO sits 52.5 km away, a 35.4-metre acrylic sphere holding 20,000 tonnes of liquid scintillator, watched by 17,612 twenty-inch photomultiplier tubes and 25,600 three-inch ones.

The signature is two flashes. An antineutrino strikes a proton and converts it into a neutron and a positron. The positron deposits its energy immediately and annihilates, giving a prompt flash. The neutron wanders for a couple of hundred microseconds, then gets captured by a hydrogen nucleus and releases a 2.2 MeV gamma ray, giving a delayed flash. Demanding both flashes close together, in space and time, removes almost all background, the same inverse-beta-decay tagging that lets Double Chooz count antineutrinos from a reactor that is not even running.

The prompt flash carries the energy information. Its brightness maps onto the antineutrino energy, minus a fixed 0.78 MeV. That mapping is the entire measurement, which is why the energy scale matters as much as the resolution.

Two oscillations at once

Antineutrinos leaving a reactor start as pure electron flavour. Along the way the three mass states drift out of step, and the chance of still finding electron flavour at the far end rises and falls with energy. At JUNO’s distance two separate oscillations are running simultaneously.

The slow one. Driven by the small splitting, it produces a single broad dip that bottoms out near 3.2 MeV and removes most of the signal. JUNO’s baseline was chosen to put that minimum right where reactor antineutrinos are most plentiful.

The fast one. Driven by the large splitting, it completes about ten full cycles between 2 and 8 MeV. These are the ripples riding on the slow dip. They are small, rapid, and they carry the answer.

The ordering is a phase, not a depth

Here is the subtlety that makes JUNO hard. There are really two fast oscillations, not one. State 3 beats against state 1 and against state 2, and those two beat frequencies differ by exactly the small splitting. The survival probability adds them with fixed weights: roughly 69 percent for the first and 31 percent for the second.

Where the sign hides. In normal ordering the heavier weight lands on the larger frequency. In inverted ordering it lands on the smaller one. The sum is still a fast oscillation of the same amplitude, but its phase creeps across the spectrum in opposite directions in the two cases.

The consequence is unforgiving. The two hypotheses predict spectra with the same depth, the same envelope and the same number of ripples. They differ only in where the ripples sit, by at most about 4 percentage points in survival probability. No amount of extra statistics helps if the detector smears the ripples together.

Two engineering numbers decide everything

Resolution better than 3 percent at 1 MeV. Blur the spectrum by more than that and neighbouring ripples merge into a smooth curve carrying no phase information. JUNO’s full detector simulation predicts 2.95 percent. A calibration-driven analysis reaches 3.02 percent, degrading to 3.12 percent once detector imperfections are included. The sensitivity to this number is brutal: at 3.5 percent, the time needed to reach 3 sigma more than doubles.

Energy scale non-linearity below 1 percent. Resolution controls whether you can see the ripples. The energy scale controls whether you can trust where they are. A systematic distortion in the conversion from light output to energy would shift the whole pattern and mimic the wrong ordering.

There is a third constraint, less discussed. Every reactor core must sit at nearly the same distance. Cores at different baselines produce ripple patterns offset from each other, and adding them washes out the interference. The site was chosen so that both power plants land within a kilometre or so of the same 52.5 km.

The reactor spectrum has to be measured separately

JUNO reads the ordering from wiggles in a spectrum. That only works if the spectrum leaving the reactor is smooth, or at least known. It is neither.

Reactor antineutrino spectra contain fine structure from thousands of individual beta decays, and previous experiments found a bump near 5 MeV that no model predicted. Structure in the source could easily be mistaken for structure from oscillation.

The fix is a second detector. The Taishan Antineutrino Observatory sits close to one Taishan core, near enough that oscillation has barely begun. It measures the reactor spectrum directly at sub-percent energy resolution and supplies JUNO with a reference shape. The published sensitivity figures assume a joint JUNO and TAO analysis, not JUNO alone.

Why JUNO’s answer is independent

Other experiments attack the ordering through matter effects. Neutrinos travelling through the Earth interact with electrons, and that interaction changes oscillation probabilities in a way that depends on the sign of the large splitting. Accelerator experiments and atmospheric neutrino detectors both use it.

The problem with matter effects. They are entangled with two other unknowns: the CP-violating phase and which side of 45 degrees the angle θ₂₃ falls on. This is why current global fits are stuck. NuFIT 6.0 finds inverted ordering disfavoured by a chi-squared difference of 6.1, and notes that the data have 2.5 to 3 sigma sensitivity in principle. In practice T2K and NOvA each prefer normal ordering on their own. They agree with each other better under inverted, and that drains the significance away.

JUNO has none of that. At 52.5 km through crust, matter effects are negligible. The collaboration’s own sensitivity study states the point directly: the measurement is completely independent of the CP-violating phase and of θ₂₃. That makes it an independent determination rather than another entry in the same tangled fit.

What the ordering decides

Whether neutrinoless double beta decay is findable. If neutrinos are their own antiparticles, that decay should occur at a rate set by an effective mass, the same Majorana-mass question that KamLAND-Zen and LEGEND-200 are already probing. Inverted ordering puts a floor under that mass, roughly 15 meV, within reach of the next generation of experiments. Normal ordering allows the contributions to cancel, so the effective mass can fall to zero. In that case a null result would prove nothing.

What cosmology is allowed to see. The ordering sets the minimum possible sum of neutrino masses: about 0.06 eV for normal, about 0.10 eV for inverted. Cosmological surveys are now pressing against the lower of those figures, so the two lines of evidence are converging on the same question from opposite directions.

What the accelerator programme can conclude. Measurements of CP violation in the lepton sector are correlated with the ordering, which is why DUNE and Hyper-Kamiokande’s long-baseline searches depend on JUNO settling it first. Fixing it independently sharpens everything downstream.

Where it stands

Established. The detector performs to specification. Two oscillation parameters are now measured better than the previous world combined, from two months of data.

Expected. A 3 sigma median determination of the ordering at about 6.5 years of exposure, roughly 7.1 years of running, from a joint JUNO and TAO analysis. Combining with atmospheric neutrino data would arrive sooner.

Unresolved. The ordering itself. Also the solar tension, which JUNO has now confirmed without explaining.

The contrast between the two timelines is worth sitting with. Precision on the solar parameters arrived almost immediately because those effects are large and slow. The ordering resists because it sits in a phase shift of a few percent. No quantity of events substitutes for being able to see the ripples at all.