KamLAND: testing a solar mystery with reactors on Earth

Change the source, keep the question

The solar neutrino problem invited a persistent objection: perhaps something in the Sun was not understood. SNO had already made the flavor-change interpretation much stronger. KamLAND offered a further test by replacing the Sun with nuclear reactors.

Reactors produce electron antineutrinos through the beta decays of fission products. Their power, fuel composition and distance give an estimate of the signal at a detector, and those inputs carry different uncertainties from a solar model.

KamLAND, in Japan’s Kamioka mine, received antineutrinos from many reactors at a typical distance of about 180 kilometres. There was no single beam at one exact baseline, so the analysis had to combine the separate reactor contributions.

Sources: [1] · [2]

A long journey made the comparison possible

An oscillation pattern depends on travel distance relative to energy. Reactor antineutrinos have energies of a few MeV. KamLAND’s baseline, much longer than those of earlier short-distance reactor experiments, let them probe the small mass-squared splitting implicated by solar measurements.

Inside a large volume of liquid scintillator, an electron antineutrino could interact with a proton and produce a positron and a neutron. The positron gave a prompt flash of light; the neutron capture gave a delayed one. Their closeness in time and position helped reject unrelated activity.

The coincidence technique picked out candidate interactions, but the detector still had to measure its energy response, count target protons and estimate backgrounds. How many candidates to expect with or without oscillations was a separate calculation.

Sources: [1] · [3]

The first sample was smaller than expected

The first result, submitted in December 2002 and published in 2003, used 145.1 days of data. After the analysis cuts, 54 candidate events remained. The no-oscillation prediction was 86.8 signal events, and the estimated background was about 0.95 events.

Subtracting that background and comparing with the predicted signal gave a ratio of 0.611, with statistical uncertainty 0.085 and systematic uncertainty 0.041. That is a significant deficit in the selected sample as a whole; it does not mean every reactor antineutrino had the same survival probability.

The result applies above a prompt-energy threshold of 2.6 MeV, which corresponds roughly to antineutrino energies above 3.4 MeV. The threshold belongs with the number: a different energy range changes both the expected spectrum and the backgrounds.

Sources: [1]

A missing fraction was only the first test

A lower total count can have more than one cause. An energy spectrum asks a harder question: is the shortfall spread across energies the way oscillation predicts?

With a larger exposure, KamLAND reported evidence for spectral distortion in a paper submitted in 2004 and published in 2005. The data preferred the changing shape expected from oscillations over an undistorted no-oscillation spectrum.

The measured shape combined reactor spectra, baselines, operating histories, interaction probability and detector response. It was not the probability curve for one neutrino energy and one reactor distance. Extracting oscillation information meant comparing the data with predictions that included all those effects.

Sources: [4] · [1]

What was shared with the solar explanation?

Combining reactor antineutrinos with solar neutrinos assumes that the neutrino and antineutrino oscillation parameters correspond, as they do in the standard framework with CPT symmetry. Under that assumption, KamLAND strongly supported the large-mixing-angle solution favored for solar neutrinos.

KamLAND did not recreate propagation inside the Sun. Solar neutrinos experience important matter effects there; KamLAND tested related parameters in a very different environment. The agreement was powerful because the sources and experimental difficulties were so different.

The reactor coincidence Lab shows how to recognize a prompt–delayed pair and estimate accidental overlaps. It does not reproduce KamLAND’s energy-spectrum fit. To follow that fit, separate two jobs: recognizing one interaction, and predicting how a whole population of interactions should be distributed.

Sources: [1] · [2] · [4]

Try it in the Lab

Primary sources & revision

1. KamLAND Collaboration · First Results: Evidence for Reactor Anti-Neutrino Disappearance (2002/2003)
2. Bahcall, Gonzalez-Garcia & Peña-Garay · Solar Neutrinos Before and After KamLAND (2003)
3. Particle Data Group · Neutrino Masses, Mixing, and Oscillations (2024)
4. KamLAND Collaboration · Evidence of Spectral Distortion (2004/2005)

First published 2026-10-09; last revised 2026-10-10. Original explanatory text and diagrams by Sandbox Physics. Illustrations are schematic; no experimental event records are reproduced here. This is an independent educational publication, not an official Nobel or experiment collaboration publication.