Event reconstruction & identification
Turning raw detector hits into physical quantities: vertex, direction, energy, flavour and event type.
Reconstruction algorithms fit light-arrival times and charges, or tracks and clusters, to infer what happened in an interaction. Particle identification and event classification, increasingly done with machine learning, separate signal channels from backgrounds and from each other.
In plain words
A detector does not record a neutrino. It records a pattern: a few thousand PMT pulses with times and charges, or a set of wire signals and tracks. Event reconstruction is the detective work that turns this pattern back into a story: where the interaction happened, which particles came out, in which direction they moved, how much energy they carried, and therefore what kind of neutrino it most likely was.
The steps
- Recording. A trigger decides in real time which activity is worth keeping, and the energy threshold sets the lowest events that survive.
- Locating. Vertex and direction reconstruction fits light arrival times or track geometry to find where the interaction occurred and where its products went, with an angular resolution that depends on event type.
- Identifying. Particle identification separates muons from electrons, nuclear recoils from electron recoils, and single from multiple particles, using ring shapes, pulse shapes and track properties.
- Classifying. Events are sorted into topologies such as tracks and cascades, and tagged with time-correlated signatures using the coincidence technique, for example a delayed neutron capture or a Michel electron.
- Measuring. The visible light or charge is converted into deposited energy and, with a model of the interaction, into an estimate of the neutrino energy, which is the subject of energy reconstruction.
All of these steps depend on calibration with known sources, which ties the detector's response to physical units.
In detail
Classical reconstruction uses maximum-likelihood fits: for each hypothesis, such as a single electron starting at a given point with a given energy, the expected charge and time at every sensor is predicted and compared with what was seen. The hypothesis with the highest likelihood wins. Super-Kamiokande's fiTQun algorithm is an example. Since the mid-2010s machine learning has become a second pillar. NOvA introduced a convolutional neural network classifier in 2016 that treats detector views like images, and similar networks are used in liquid argon experiments such as MicroBooNE and DUNE.
Why it matters for neutrino physics
Oscillation, solar and astrophysical results are only as good as the reconstruction behind them. The atmospheric oscillation discovery in Super-Kamiokande depended on telling muon-like from electron-like rings and on measuring their direction. Reconstruction biases, for example in the energy scale or the position, are among the leading systematic uncertainties of modern experiments.
Frequently asked questions
Is the neutrino's energy measured directly?
No. Only the visible products are measured. The neutrino energy is inferred from them, which requires models of the interaction and of particles that escape unseen, such as neutrons or the outgoing neutrino in neutral-current events.
Do machine-learning methods replace physics-based fits?
Not entirely. They are often better at classification, while likelihood fits remain common for precise vertex and energy estimates. Both are validated with calibration data and simulation.
Go deeper
- Energy thresholdThe lowest energy at which a detector can register or reliably select neutrino events.
- Trigger & data acquisitionThe real-time decision system that selects which detector activity is recorded.
- Coincidence / Delayed-signal techniqueRequiring two correlated signals, close in space and time, to identify an event and reject random backgrounds.
- Vertex & direction reconstructionDetermining where an interaction happened and in which direction its products moved.
- Particle identification (PID)Determining which type of particle produced a signal.
- Event topologies in neutrino telescopesThe characteristic light patterns that different neutrino flavours and interactions produce in large detectors.
- CalibrationDetermining a detector's response using known signals.
Mentioned in 3 pages
Last updated: · Editorial standard
Cite this page
Text
Neutrino Atlas (2026). “Event reconstruction & identification”. Retrieved 22 November 2026, from https://neutrino-atlas.com/terms/event-reconstruction/BibTeX
@misc{neutrinoatlas2026termseventreconstruction,
title = {Event reconstruction & identification},
author = {{Neutrino Atlas}},
year = {2026},
url = {https://neutrino-atlas.com/terms/event-reconstruction/},
note = {Accessed 2026-11-22},
language = {english}
}