Neutrino Atlas · Worksheet · Lesson 2.7
How to read a neutrino measurement
Read the lesson on neutrino-atlas.com and answer the questions. Tick exactly one answer each time.
https://neutrino-atlas.com/learn/reading-a-measurement/
Part A – Beginner
Which kind of uncertainty shrinks when an experiment collects more data?
- A Systematic uncertainty
- B Statistical uncertainty
- C Both shrink equally
- D Neither of them
What does KATRIN's result “less than 0.45 electronvolts” mean?
- A The neutrino mass is 0.45 eV
- B The neutrino mass is exactly zero
- C Data fit zero mass; larger values are excluded
- D The mass lies between 0.45 and 1 eV
What is the usual threshold for claiming a discovery?
- A 1σ
- B 2σ
- C 3σ
- D 5σ
What caused OPERA's apparent faster-than-light neutrinos in 2011?
- A A loose fibre-optic cable in the timing system
- B A sterile neutrino
- C Too few recorded events
- D Neutrinos really travelling faster than light
Part B – Intermediate
With 10,000 expected events, what is the relative statistical uncertainty?
- A 10 %
- B 1 %
- C 0.1 %
- D 100 %
Why can KATRIN's best-fit mβ² be negative?
- A Statistical fluctuations push the fitted value below zero
- B Neutrinos have imaginary mass
- C The Bayesian prior forces it
- D Tritium decays backwards
What combined significance did the LSND and MiniBooNE excesses reach?
- A 3.8σ
- B 4.2σ
- C 5.2σ
- D 6.0σ
What does the look-elsewhere effect require of an analysis?
- A Repeating the experiment at another site
- B Adding systematics in quadrature
- C Converting the local p-value into a global one
- D Using a Bayesian prior
Part C – Expert (optional)
For one parameter under Wilks' theorem, which Δχ² threshold gives 90 % CL?
- A 1.00
- B 2.71
- C 4.61
- D 5.99
Why does Wilks' theorem fail for the mass ordering?
- A The data are Poisson distributed
- B Nuisance parameters are profiled
- C Δm²21 is too small
- D The orderings are discrete, non-nested hypotheses
What one-sided p-value corresponds to Z = 5?
- A 2.87 × 10⁻⁷
- B 1.35 × 10⁻³
- C 5 × 10⁻²
- D 3.2 × 10⁻⁵
Why is KATRIN's 0.45 eV limit stronger than its median expected limit?
- A A Bayesian prior was used
- B Systematics were neglected
- C The best fit fluctuated downward
- D The confidence level was raised to 95 %
Extra task
In three or four sentences, explain to someone without a physics background what this lesson is about.
Answer key for teachers
- B – Statistical uncertainty comes from random fluctuations in counts; a miscalibrated instrument, a systematic effect, stays wrong however much data is taken.
- C – An upper limit means no effect was seen: the data are compatible with zero, and masses above the limit are excluded at the stated confidence level.
- D – A 5σ deviation arises by chance only about once in 3.5 million tries; 3σ, about once in 740, counts only as evidence.
- A – The roughly 6σ effect was a systematic error, which no amount of extra data could have revealed.
- B – Poisson fluctuations are about √N; √10,000 = 100, which is 1 % of 10,000.
- A – If the true mass is zero or small, fluctuations often push the fit below zero; −0.14 eV² is such a case, and boundary-aware methods handle it.
- D – LSND reported 3.8σ and MiniBooNE 4.8σ; combined they reach 6.0σ, yet MicroBooNE found no matching νe excess.
- C – Scanning many energies or sky positions makes a large fluctuation likely somewhere, so a trials factor converts the local into a global p-value.
- B – For k = 1, Δχ² = 1, 4, 9 give 1, 2, 3σ and 2.71 gives 90 %; 4.61 and 5.99 are the k = 2 values for 90 % and 95 %.
- D – The NO–IO Δχ² is roughly Gaussian with mean ±T₀ and width 2√T₀, so √Δχ² is only a rough guide; pseudo-experiments are needed.
- A – p = 1 − Φ(Z) = ½ erfc(Z/√2); for Z = 3 this gives 1.35 × 10⁻³, for Z = 5 it gives 2.87 × 10⁻⁷.
- C – The best fit mβ² = −0.14 eV² lies below zero, so the limit is tighter than the sensitivity; collaborations therefore quote both.