How does a faint signal become a neutrino discovery?
A result connects recorded signals, calibration, event selection and tests of alternatives. No one striking event can replace that chain.
Begin with what the instrument actually recorded
A sensor does not record “a cosmic neutrino.” It records a response: for example, an electrical pulse with a time and amplitude. A scientific claim starts by preserving those measurements and understanding what they can and cannot tell us.
Calibration connects instrument signals to physical quantities. A clock offset can change an inferred direction; a sensitivity error can change an energy estimate. In an optical detector, the medium through which light travels also belongs to the measurement chain.
The first useful question is therefore concrete: would the same physical event produce the predicted signals throughout the array? Known light sources, ordinary particle samples and laboratory measurements help check this before unusual events receive a cosmic interpretation.
Define the sample before interpreting it
Event selection is a rule for deciding which records enter an analysis. It might require an interaction to start inside a detector, reject activity near its boundary, or demand a well-reconstructed track. Different rules answer different questions and retain different backgrounds.
The rules also change the signal efficiency. Rejecting more background can discard interesting events. This trade-off must be measured or simulated, with checks against suitable data; a clean-looking sample is not automatically an unbiased view of nature.
The 2013 IceCube starting-event analysis used the outer detector as a veto against entering particles. That strategy mattered because atmospheric muons could otherwise imitate energetic activity. The geometry and timing of an event were part of its eligibility, not just decoration in a display.
- 01Calibrate the record
Clocks, response and propagation
- 02Specify alternatives
Selection and background uncertainties
- 03Test the claim
Search scope and robustness checks
Give the alternative explanation a fair test
A background is a process that can enter the selected sample without the target phenomenon. Atmospheric neutrinos are real neutrinos, but they are background when the question asks about distant astrophysical emission. “Background” describes their role in a particular test.
Researchers need an estimate with uncertainties, rather than a single reassuring number. Control samples constrain some backgrounds; simulations connect physical processes to the selection. Checks of energy, direction and event shape can reveal an explanation that a total count misses.
Independent counts with a fixed expected mean follow a Poisson distribution. The equation below is a useful starting model, not the complete IceCube analysis. Uncertain background rates, correlated events or selection changes require additional treatment.
How many opportunities did the search have?
Suppose you inspect many sky positions or repeatedly vary a time window. Even an ordinary background has more opportunities to make one location look exceptional. A probability computed for one fixed test does not describe the entire search.
A local significance describes a specified location or hypothesis. A significance corrected for the search accounts for the tested alternatives. The correction depends on the actual procedure, including correlated trials; multiplying by an arbitrary number is not a universal solution.
A background probability also is not the probability that a proposed source is false. It asks how unusual the chosen statistic would be under a specified background model. Interpreting the source requires the model assumptions, systematic checks and other observations too.
Check again without pretending every check is independent
Changing a calibration model, using another event class or examining a control region can challenge a result. Those are valuable robustness checks. When they reuse the same records or assumptions, however, they are not new independent experiments.
A later observing period can test whether a signal persists. Another observatory can bring different instrumental errors. The relevant comparison depends on the claim: a transient need not repeat steadily, and a different energy threshold can make an apparently absent counterpart unsurprising.
The teaching Labs let you rehearse parts of this reasoning: estimate chance pulse pairs, or see how a wrong light model changes a fitted track. Their synthetic results illustrate a method. The published discovery rests on the experiment’s own records, controls and analysis.
Try it in the Lab
Primary sources & revision
- IceCube Collaboration · Instrumentation and Online Systems (2017; arXiv v3)
- IceCube · The end of the tenth polar season (5 March 2014)
- IceCube Collaboration · Evidence for High-Energy Extraterrestrial Neutrinos (2013)
- Particle Data Group · Statistics (2025)
- IceCube Collaboration · Evidence for neutrino emission from NGC 1068 (2022; arXiv v2, 2024)
- Reines et al. · Detection of the Free Antineutrino (1960)
- IceCube · Public data releases
First published 2026-10-10; 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.