The IceCube Neutrino Observatory at the South Pole searches for weakly interacting particles called neutrinos that are able to travel undisturbed through the cosmos. When a neutrino interacts with a molecule of ice, it produces secondary particles that emit blue light, which is then detected by IceCube. From the shape of that light pattern, the neutrino type can be determined, with muon neutrinos leaving longer “tracks” and electron neutrinos leaving a spherical pattern or “cascade.” However, the light patterns become harder to discern at lower energies where less light is seen.
In a study submitted to the Journal of Instrumentation, the IceCube Collaboration presents WavePID, a classifier that exploits nanosecond-scale timing on individual light sensors. WavePID was shown to improve both cascade purity and classification performance, providing useful early timing information that existing IceCube classifiers do not fully capture.
“Even when the overall shape of an event is hard to see, the arrival times of the first photons on each sensor may still carry a fingerprint of the particle that produced them,” explains Steven Eulig, a research scholar at Harvard University and study lead. “We wanted to test whether this timing fingerprint is strong enough to improve classification in events where existing methods struggle.”
The collaborators were inspired by the FNAL-1267 beam test at Fermilab, which explored whether a single optical module in a water tank could identify the type of particle just from the shape of a light pulse.
“The answer was yes; the pulse shape alone could statistically separate the two particle types with just one sensor,” says Carlos Argüelles-Delgado, a professor at Harvard University. “We wondered if the same timing idea would work in IceCube, where photons travel through the deep Antarctic ice before reaching the sensors.”
Using a two-step process, they constructed WavePID templates from pulse-level timing information in simulated IceCube events and then used those templates to classify the detector light patterns. For each individual sensor, WavePID used three pieces of information: the sensor’s distance from the neutrino interaction point, the fraction of light arriving in the first 14 nanoseconds, and how that timing compared across sensors.
“We chose the 14-nanosecond window because it works well for the detector, contains useful type information, and is larger than the detector’s timing uncertainty,” adds Nicholas Kamp, a postdoctoral researcher at Harvard University and study colead. “The added signal remains stable across detector-systematic variations.”
For each event, WavePID compared the observed timing pattern with each template and returned a track-ness score. While DynEdge, IceCube’s leading graph-neural-network classifier, learns from the overall pattern of detected light, WavePID adds a different perspective by focusing on the first nanoseconds of photon arrivals. These earliest photons carry a genuinely new discriminating signal, which WavePID captures using only three physically motivated observables. Because this signal complements rather than duplicates DynEdge’s representation, combining the two methods recovers more of the available separation than either can do on its own.
“WavePID is useful for IceCube because it gives us an additional handle for classifying low-energy events, where the detector often sees too little light to rely on shape alone,” says Eulig. “The next step is to bring WavePID into real IceCube analyses and assess its impact on physics results.”
Another future application of WavePID would be for neutrino oscillation measurements where low-energy neutrinos are key in identifying the different types of neutrinos. With the addition of more sensors and improved timing information in the proposed extension of IceCube, IceCube-Gen2, the timing-based particle identification will become an even more powerful tool for neutrino physics.
+ info “WavePID: Low-energy flavor identification using single-PMT time series in IceCube,” IceCube Collaboration: R. Abbasi et al. Submitted to the Journal of Instrumentation. https://arxiv.org/abs/2607.02078