Abstract
We describe in detail the online data analysis pipeline that was used in the multi-messenger search for common sources of gravitational waves (GWs) and high-energy neutrinos (HENs) during the second observing period (O2) of Advanced LIGO and Advanced Virgo. Beyond providing added scientific insight into source events, low-latency coincident HENs can offer better localization than GWs alone, allowing for faster electromagnetic follow-up. Transitioning GW+HEN analyses to low-latency, automated pipelines is therefore mission-critical for future multi-messenger efforts. The O2 Low-Latency Algorithm for Multi-messenger Astrophysics (LLAMA) also served as a proof-of-concept for future online GW+HEN searches and led to a codebase that can handle other messengers as well. During O2, the pipeline was used to take LIGO/Virgo GW candidates as triggers and search in realtime for temporally coincident HEN candidates provided by the IceCube Collaboration that fell within the 90% credible region of the reconstructed GW skymaps. The algorithm used NASA's Gamma-ray Coordinates Network to report coincident alerts to LIGO/Virgo's electromagnetic follow-up partners.
Introduction
In the era of multi-messenger astronomy, timing is critical. When gravitational wave detectors like LIGO and Virgo identify a binary neutron star merger, the resulting spatial uncertainty often covers hundreds or thousands of square degrees. For electromagnetic (EM) telescopes with narrow fields of view, searching this vast "error box" is a significant challenge. High-energy neutrinos (HENs) provide a powerful solution; unlike GWs, neutrinos can be localized to within approximately 0.5 degrees. A coincident GW and neutrino detection not only increases confidence in the astrophysical nature of the event but also provides a precise coordinate for rapid telescope pointing. LLAMA was developed to transition these joint searches from offline post-processing to a real-time online mode, aiming to deliver actionable alerts within minutes.
Methods and Implementation
LLAMA utilizes a modular, highly extensible architecture designed for stability and rapid recovery. The pipeline's workflow is structured as a Directed Acyclic Graph (DAG), where each node represents a specific task, such as downloading a skymap or executing a coincidence check. This ensures that if one component fails or data is delayed, the system can resume exactly where it left off.
Two Python-based daemons, gcnd and gwhend, manage the pipeline. gcnd listens for alerts from NASA’s Gamma-ray Coordinates Network (GCN), while gwhend monitors the file system to trigger analysis as soon as new GW skymaps or neutrino lists arrive. The core algorithm calculates a joint likelihood ratio based on spatial overlap and a temporal window of ±500 seconds, integrating the GW skymap's probability distribution with the neutrino's Point Spread Function (PSF).
Results and Discussion
The Localization Bonus
A primary achievement of the LLAMA pipeline is the dramatic reduction in search area for cosmic events. Even with a single-detector GW detection—which typically results in a large ring across the sky—the addition of a coincident neutrino can pinpoint the source location, making it a manageable target for traditional telescopes.
Operational Latency
During the O2 observing run, LLAMA successfully processed GW triggers and searched for neutrino coincidences in real-time, meeting nominal analysis targets of less than 30 minutes. However, analysis of the pipeline's performance revealed that a significant portion of total latency was due to policy rather than computation; LIGO/Virgo required human vetting of triggers before broadcasting, which added approximately 30 minutes of delay.
While no statistically significant joint GW+HEN events were found during O2, the pipeline proved that automated, multi-messenger searches are feasible and ready for future, more sensitive observing runs.
Future Work
Future developments for LLAMA include the implementation of "sub-threshold" searches—finding events that are too weak to be significant in a single channel but become undeniable when combined. Additionally, the team aims to integrate more complete galaxy catalogs to further constrain search areas and move toward a fully automated "lights-out" system that can trigger GCN Circulars without human intervention, further slashing the time between a cosmic collision and the first telescope observation.
Summary updated