When can a few neutrino events become evidence?

Start with a simple counting experiment

Imagine a counting experiment with a fixed observing period and a selection decided in advance. Events are rare and independent, and the background mean is known exactly. Under these teaching assumptions a Poisson model describes the count. None of the numbers below are IceCube data.

Suppose five events are seen when the background alone should give one. The chance of five or more under that model is about 0.00366. If the background expectation is three instead, the same upper-tail probability is about 0.185. The count is the same; the strength of the evidence is very different.

The tail probability adds up all outcomes at least as extreme as the one observed, for the chosen statistic. It is not the probability of exactly five, and it is not the probability that the background hypothesis is true. A real analysis must also justify the assumptions behind its background expectation.

P(N\geq n\mid b)=1-e^{-b}\sum_{k=0}^{n-1}\frac{b^k}{k!}

N is the count, n the observed count (a nonnegative integer) and b the background mean, known exactly and dimensionless. When the observed count is zero, the sum is empty.

Sources: [1]

Measure what can imitate the signal

A useful background estimate comes from a defined control measurement or a validated physical model. In coincidence searches, shifting the times estimates the accidental pairs. In optical telescopes, control samples of muons entering from outside and predictions of atmospheric neutrinos constrain different parts of the background.

Carrying a control region over to the signal region needs a reason. A low-energy sample that is easy to measure does not automatically predict a rare high-energy tail. Differences in acceptance, detector period or physical origin can break a simple scaling.

Report both the estimate and how it was obtained. If a later calibration changes the control selection, recheck the transfer and its correlations. A background number without that chain of measurements cannot tell you whether a small excess is convincing.

Sources: [2] · [3] · [4]

Fix the test before looking at the answer

If cuts are tuned until an excess looks strongest, the final procedure has searched more possibilities than its last plot shows. Blind analysis eases that pressure: while the method is developed, a signal region is withheld, information is scrambled or a fitted parameter is hidden.

State what is hidden, which control information stays visible and what condition allows unblinding. Test reconstruction and selection on simulations or suitable control samples. A development log shows which choices were made before seeing the signal and which after.

Blinding cannot fix a wrong response model or catch an overlooked systematic effect. It is one safeguard against biased choices, alongside validation and review. Experiments implement it differently, so a paper should describe its actual procedure; the word “blind” alone guarantees nothing.

Sources: [5] · [6]

An uncertain background adds another layer of inference

The simple example treated the background mean as exact. A real control sample has limited statistics, and the detector efficiency or atmospheric production may be uncertain. These quantities enter the fit as nuisance parameters, constrained by other measurements.

A frequentist analysis can profile the likelihood over those parameters; a Bayesian analysis can integrate over a stated prior. The resulting tail probability or credible statement holds only within that complete model. Adding one arbitrary percentage is no substitute.

Correlations matter too. One optical calibration can shift signal acceptance, energy reconstruction and background prediction together. Checking alternative models and control distributions helps reveal an apparent excess caused by a shared response mismatch.

Sources: [1] · [4] · [7]

A sky scan always has a hottest spot

Testing one source chosen in advance is a different test from scanning the whole sky and reporting its most unusual patch. Even a sky with only background has a hottest spot. The global test must repeat the same search on background-only realizations of the sky and compare their hottest spots with the observed one.

With 100 independent tests, each with a one-percent false-positive probability, the chance of at least one false positive is about 63 percent. This is again a teaching example. Neighboring sky locations are correlated, so counting pixels does not give a valid trials correction.

Ask for the full package: selection, background controls, uncertainty model, search scope and replication. A small sample can be convincing when alternative explanations are tightly constrained. A large fitted excess can stay inconclusive when its background or search procedure is poorly defined.

Sources: [1] · [7] · [8]

Try it in the Lab

Supernova Neutrino Burst Observatory

Primary sources & revision

First published 2026-10-11; last revised 2026-10-11. 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.