IceCube Neutrino Observatory

Construction Completion Date: December 2010

End of Operations: No specific requirement

Data Archives:

• Dataverse
• Data Releases
• HEASARC

The IceCube Neutrino Observatory is a cubic-kilometer Cherenkov particle detector deployed in the Antarctic ice beneath the Amundsen-Scott South Pole Station. It consists of 86 strings of photo-detectors, extending to a depth of about 2,500 meters below the glacier's surface and instrumenting a cubic-kilometer of ice. The Digital Optical Module photo-detectors detect the light produced by relativistic charged particles produced by neutrino interactions in or near the instrumented volume of ice.

IceCube is sensitive to neutrinos from all directions. As neutrinos pass through the ice, their interactions can leave track signatures (order-km in length) in the IceCube detector array when they produce secondary muons or compact signatures (cascades of ~m in extent) when they produce secondary electrons or hadrons. Track events can be reconstructed with an uncertainty of less than 1 degree, while cascade events have higher signal purity. IceCube triggers on signals for neutrino energies greater than ~10 GeV (10¹⁰ eV) and can identify likely astrophysical neutrino events for neutrino energies greater than ~50 TeV (5x10¹³ eV) and issue alerts to the community.

The IceCube realtime alert point of contact can be reached at roc@icecube.wisc.edu.

IceCube has been operating a real-time alert system since 2016 (IceCube Collaboration 2017). The system includes multiple event types.

IceCube High-Energy Neutrino Track Alerts

This section focuses on high-energy (of order TeV-PeV) track-like alerts. These events originate from charged-current interactions between muon neutrinos and quarks in the ice. Owing to their excellent angular resolution, which improves with energy, track-like alerts are particularly valuable for neutrino astronomy.

The system allows rapid notification of the multimessenger community about likely astrophysical neutrino track-like events detected by IceCube, initiating follow-ups to detect source candidates. A major breakthrough occurred in September 2017, when a neutrino with an energy of approximately 300 TeV (IC-170922A) was found in spatial and temporal coincidence with the flaring blazar TXS 0506+056. This association, supported by gamma-ray data from Fermi-LAT, had a significance of ~3 (IceCube Collaboration 2018). This marked the first compelling multimessenger association between a neutrino and an astrophysical object, providing evidence that blazars can act as cosmic-ray accelerators. Archival analysis revealed a possible earlier neutrino flare from the same source between September 2014 and March 2015, with a significance of 3.5 independent of the 2017 alert (IceCube Collaboration 2018). These observations highlight the potential of real-time multimessenger astronomy.

In 2019, the real-time framework for track-like events was upgraded to improve event selection and alert message content (IceCube Collaboration 2023).

A recent enhancement in the IceCube real-time system was introduced in September 2024 (IceCube Collaboration 2024). IceCube now adopts a revised follow-up reconstruction strategy that applies different algorithms based on the event's energy deposition or topology. This hybrid approach improves angular resolution across a broad energy range and ensures robust statistical coverage of the reported localization uncertainties (IceCube Collaboration 2026).

Description

Events are classified into Gold and Bronze streams based on their likelihood of being astrophysical, quantified by a metric historically called signalness (with the switch to GCN over Kafka, this metric is now referred to as p_astro):

where and are the expected numbers of signal and background events (both atmospheric neutrinos and muons) with reconstructed event energy above the selected energy proxy at declination , derived from simulations.

The streams were optimized assuming an astrophysical neutrino power-law energy spectral index of .

The selection criteria for the alerts were tuned to achieve, on average, a probability of being astrophysical of 30% for Bronze alerts and 50% for Gold alerts.

Several neutrino candidate selections have been developed, each targeting different types of events. The resulting samples are combined to produce the Gold and Bronze alerts. For details about the selection filters, see (IceCube Collaboration 2023).

Each alert selection determines the values to report in the GCN Notice (probability of being astrophysical, false alarm rate, and likely neutrino energy) based on the parameters used in candidate event identification. When an event is identified by multiple selections, a single GCN Notice is issued, and a hierarchical rule is applied to decide which selection's information is included. This ordering is based on astrophysical signal purity and angular resolution studies.

The total rate of distributed alerts is approximately 26 per year (10 Gold and 16 Bronze).

The neutrino effective areas for these combined selections, for both through-going and starting events, are shown in Figure 1 of (IceCube Collaboration 2023). Astrophysical neutrino-generated tracks selected by these selections are expected to have very good angular resolution. The median angular resolution for these events as a function of energy is shown in Figure 2 of (IceCube Collaboration 2023).

Alert distribution

High-energy track alerts are openly circulated to the community via GCN as Notices and Circulars in a two-step process:

1. A prompt GCN Notice (Revision 0) is broadcast shortly after the identification of an astrophysical neutrino candidate at the South Pole.
2. A follow-up reconstruction, which is more computationally intensive, is then performed and circulated in a new revision of the alert via GCN Notice (Revision 1), with an updated position and uncertainty radius. This is accompanied by a GCN Circular, which includes a directional error box for the event as well as the position of nearby gamma-ray sources of interest, if any.
   Until recently, these Notices were circulated via the classic GCN system. A transition to broadcasting over Kafka is now underway, as described in a separate section below.

GCN Notice Types in GCN Classic and GCN Classic Over Kafka

Historically, IceCube alerts generated by real-time astrophysical neutrino searches(IceCube Collaboration 2017b)(IceCube Collaboration 2023) have been sent over the classic GCN system through the Astrophysical Multimessenger Observatory Network (AMON). Detailed descriptions and examples are available.

We are currently migrating these alerts to the new GCN Kafka system. During the initial months of this transition, alerts will be sent through both AMON and Kafka, before fully switching to Kafka.

This page documents alert content, schema, and probability maps.

GCN Notice Types in GCN Classic

Historically, IceCube alerts generated by real-time astrophysical neutrino searches(IceCube Collaboration 2017b)(IceCube Collaboration 2023) have been sent over the classic GCN system through the Astrophysical Multimessenger Observatory Network (AMON). Detailed descriptions and examples are available.

GCN Notice Types in GCN Over Kafka

New with GCN over Kafka:

• For each alert, multi-order probability maps are now released, which was never done with the previous system.
• Two GCN Notices are sent for each event - one corresponding to the preliminary reconstruction and one corresponding to the revised reconstruction - consistent with what is shown in the AMON table.

IceCube Gold/Bronze Track Alert Schema

topic = gcn.notices.icecube.gold_bronze_track_alerts

IceCube Gold/Bronze alerts now follow this GCN schema structure.

Key fields are described below:

• $schema - GCN schema structure according to the nasa-gcn/gcn-schema GitHub project.
• mission - Instrument issuing the alert (IceCube in our case).
• instrument - Detector configuration; e.g., IC86 refers to IceCube with 86 strings active (full detector).
• messenger - Neutrino (Type of astrophysical messenger).
• pipeline - Alert type, either Bronze Track Alert or Gold Track Alert. Bronze alerts have an average astrophysical probability of 30%, Gold alerts 50% (determined by the p_astro parameter).
• record_number - 0 for the first preliminary reconstruction, 1 for the revised reconstruction.
• event_name - Name of the event, e.g., IceCube-YYMMDDA (the letter increments if multiple alerts occur on the same day).
• id - Internal event ID, formatted as RUNID_EVENTID (currently published here).
• alert_datetime - Timestamp of the GCN notice (UTC time).
• alert_type - Can be "initial" for preliminary reconstruction (Revision 0), "update" for revised (Revision 1), "retraction" if the alert is retracted.
• alert_tense - "current" for real alerts, "test" for tests, "injections" for fake injections used as tests.
• alert_topology - Event topology: "Track".
• number_of_events - Number of events contributing to the alert (1, being a single neutrino alert).
• ra, dec - Best-fit right ascension and declination (J2000, degrees).
• ra_dec_error - Combined positional uncertainty, calculated from the RA and Dec errors at 90% containment. For "record_number"=1, it is calculated in the following way, taking into account the spherical geometry:

ra_err_mean = (abs(ra_err_plus) + abs(ra_err_minus)) / 2
dec_err_mean = (abs(dec_err_plus) + abs(dec_err_minus)) / 2
err_combined = np.sqrt( (ra_err_mean * np.cos(np.deg2rad(dec)))**2 + (dec_err_mean)**2 )

These ra_err_plus, ra_err_minus, dec_err_plus, and dec_err_minus values correspond to the 90% containment errors, which are published in the GCN circular associated with the event.

• containment_probability - Confidence level for the positional error (e.g., 0.9 for 90%).
• systematic_included - True if systematics are included in the positional uncertainty, otherwise False. For "record_number"=0 (preliminary reconstruction), this is set to False. For "record_number"=1 (updated reconstruction associated to the GCN circular), this is set to True. In fact, starting from September 2024, the reconstruction algorithms used for realtime alerts have been improved (IceCube Collaboration 2024). Systematic errors affecting the reconstructions, such as those from the ice model or detector geometry, have been thoroughly investigated to ensure proper statistical coverage (see details in (IceCube Collaboration 2026)).
• healpix_url - URL to the multi-order probability map of the event; null for preliminary reconstruction. For "record_number"=0 (preliminary reconstruction), healpix_url is expected to be null. For "record_number"=1 (updated reconstruction associated to the GCN circular), it contains a link from which the file can be downloaded. A description of the multi-order probability map of the event is provided in the section "Multi-order Probability Maps (MOC/HEALPix)" below. The probability maps have correct coverage including both statistical and systematic uncertainties.
• trigger_time - Time of the triggered neutrino event (UTC time).
• nu_energy - Estimated neutrino energy (TeV).
• p_astro - Estimated probability that the event is astrophysical (in the past, this was named signalness).
• far - False Alarm Rate (Hz), indicating the expected frequency of such events in IceCube.

NOTE: With the transition to GCN over Kafka, the reported values of p_astro and far are now updated in the revised reconstruction. In the previous system, the same quantities from the preliminary reconstruction were reported. The updated values are computed using the revised best-fit direction and the initial energy estimate of the event.

The GCN schema and example JSON message files are available to use.

Multi-order Probability Maps (MOC/HEALPix)

GCN strongly encourages the use of multi-order sky maps (MOC). See the documentation.

These maps use a variable angular resolution, with regions of higher probability represented at finer resolution and lower-probability regions encoded at coarser resolution. Compared to single-resolution HEALPix maps, multi-order sky maps are significantly more efficient in terms of storage footprint and I/O performance, while retaining full localization information.

For IceCube track alerts, the probability distribution of the true neutrino direction is provided in this multi-order format, with probability density expressed in units of sr^-1. This format retains the pixels originally scanned by IceCube with their original areas. Areas nearby the best-fit direction were scanned using a finer pixelization while further away a coarser pixelization is used. This ensures efficient storage while preserving the accuracy of the probability distribution. The value stored in each pixel corresponds to a probability density. When multiplied by the pixel solid angle and summed over all pixels, the total probability integrates to unity.

The header of each file contains:

• RUNID,EVENTID: RunID and EventID of the alert.
• SENDER: IceCube Collaboration.
• DATE-OBS: UTC date of the observation.
• MJD-OBS: Modified Julian date of the observation.
• EVENT-TYPE: Identification of event selection type: neutrino for gfu-gold or gfu-bronze, and HESE hese-gold, or hese-bronze types (see IceCat-1 paper publication for details about the different alert types).
• ALERT-STREAM: Identification of alert type: Gold or Bronze.
• RA,DEC [deg] (and _ERR_50, _ERR_90): Best fit direction in J2000 equatorial coordinates, with asymmetric 50% and 90% CL error rectangle boundaries.
• CONTOUR AREA (50%),CONTOUR AREA (90%) [sdeg]: Contour area of uncertainty contours at 50% and 90% confidence levels around the best-fit direction.

IceCube Alert Archive

The IceCube alert archive is publicly accessible at: https://roc-2.icecube.wisc.edu/public/alerts/

Each alert is provided as a compressed FITS file using the following naming convention: IceCube-YYMMDDX_skymap_probdensity_multiorder.fits.gz

The alert identifier follows the format: IceCube-YYMMDDX, where YYMMDD denotes the date of the event (e.g., IceCube-260125A corresponds to an alert issued on January 25, 2026), and X is a letter distinguishing multiple alerts occurring on the same day (A, B, C, …).

The download link for each file is also included in the corresponding GCN notice distributed via Kafka through the item healpix_url (see the schema explanation above).

Citing IceCube Track Alerts

For scientific work using IceCube track alerts or the associated probability maps, we kindly ask that you consider citing the following:

• The IceCube catalog (IceCat-1) (IceCube Collaboration 2023)
• The documentation describing the reconstruction algorithms adopted for alert tracks (IceCube Collaboration 2024)

If you refer to a specific alert and use a multi-order probability map, please cite also:

• the corresponding GCN Notice
• the associated GCN Circular

This helps ensure that particular events and datasets are clearly identified.

IceCube LVK Alert Neutrino Track Search

topic = gcn.notices.icecube.lvk_nu_track_search

IceCube performs realtime searches for coincident neutrino signals with all LIGO/Virgo/KAGRA gravitational-wave alerts, using a realtime muon neutrino track event selection(IceCube Collaboration 2016) and the sky maps from gravitational wave detectors. These low-latency joint searches identify neutrino signals within a +-500 second time window surrounding the LVK alert time, and aim to identify multi-messenger transient sources and seed electromagnetic followup up observations(IceCube Collaboration 2020) (IceCube Collaboration 2023a).

These searches are performed as quickly as possible, as soon as the realtime neutrino information for the 1000 second window are available, and are repeated for each update to the LVK alert information. The results from each repeated search will sent as a GCN notice, including identifying information to the corresponding LVK alert version. Results from all searches in LVK Runs 03 and 04 are available on the IceCube Realtime Gravitational Wave Followup page

For high-significance LVK alerts, two hypothesis tests are conducted. The first search is a maximum likelihood analysis which searches for a generic point-like neutrino source coincident with the given GW sky map. The second uses a Bayesian approach(Bartos et al. 2019) to quantify the joint GW + neutrino event significance, which assumes a binary merger scenario and accounts for known astrophysical priors, such as GW source distance, in the significance estimate. For low-significance LVK alerts, only the Bayesian search is performed. All searches are reported via GCN Notices over Kafka, and any coincident observation with a observed p-value less than 1% will also be sent as a GCN Circular.

The GCN Notice will highlight:

• Overall search results p-values for the generic and Bayesian searches (pval_generic and pval_bayesian)
• If the analysis p-value from either search is less than 10%, the number (n_events_coincident) and per-event information for coincident neutrino events (for events with a per-event p-value of less than 10%) are also provided:
  • event_dt - relative timing of neutrino to GW alert (sec)
  • id - unique identifier label for coincident neutrino event (string)
  • ra, dec - reconstructed neutrino event direction (deg., J200),
  • ra_uncertainty - circular direction error (deg.) at 90% confidence.
  • event_pval_generic, event_pval_bayesian - per-event p-values for each search. In the case of multiple coincidences, can be used to determine relative importance of each event to search results.
• If the analysis p-value from the generic transient search is less than 10%, the most likely direction from the generic neutrino source coincidence search is also given (most_probable_direction - RA/DEC, deg. J2000).
• neutrino_flux_sensitivity_range - Time integrated flux sensitivity range (min, max) [GeV cm^-2] and energy sensitivity range (lower, upper) [GeV], assuming an E^-2 neutrino spectrum (E² dN/dE) found within the points in the 90% region of GW map localization

The GCN schema and example JSON message files are available to use. See our Schema Browser for more information on the properties defined in the schema.

Please note: The reported p-values here do not account for any trials correction (multiple hypotheses testing). The false alarm rate of these coincidences can be obtained by multiplying the p-values with their corresponding GW trigger rates. All neutrino searches performed to date for Run O3 and O4 are cataloged on the IceCube Realtime Gravitational Wave Followup page. Rate estimates and more information on the LVK alerts can be found in the LIGO/Virgo/KAGRA Public Alerts User Guide.

Common GCN Circular Types

This section overviews the typical latency time for the release of GCN circular after IceCube real-time alerts.

Yearly Alert Rates

IceCube Acknowledges

The IceCube Neutrino Observatory is funded and operated primarily through an award from the National Science Foundation to the University of Wisconsin-Madison. The IceCube Collaboration, with over 350 scientists in 58 institutions from around the world, runs an extensive scientific program that has established the foundations of neutrino astronomy. See the full list of institutions.

IceCube's research efforts, including critical contributions to the detector operation, are funded by agencies in Australia, Belgium, Canada, Denmark, Germany, Italy, Japan, New Zealand, Republic of Korea, Sweden, Switzerland, Taiwan, the United Kingdom, and the United States including the NSF. IceCube construction was also funded with significant contributions from the National Fund for Scientific Research (FNRS & FWO) in Belgium; the Federal Ministry of Education and Research (BMBF) and the German Research Foundation (DFG) in Germany; the Knut and Alice Wallenberg Foundation, the Swedish Polar Research Secretariat, and the Swedish Research Council in Sweden; and the Wisconsin Alumni Research Foundation in the U.S.

Bibliography