How many times should we matched-filter gravitational wave data? A performance comparison of the GstLAL online and offline analyses

Gravitational Wave (GW) search pipelines for compact binary coalescences utilize matched-filtering to identify GW candidates in the strain data. Traditionally, two types of analysis are performed on every stretch of data, a low-latency (online) analysis, and a high-latency (offline) analysis. The online mode is crucial for multi-messenger follow up, whereas the offline mode uses acausal methods to produce more robust results. Matched-filtering represents the vast majority of the computational burden in both analysis modes. We describe a novel method that applies acasual techniques to the data products created by an online analysis to effectively deliver an offline analysis without repeating the process of matched filtering. This method enables a dramatic reduction in the computational cost of an offline analysis. To estimate the effectiveness of our method, we perform an online analysis to which we apply our method, as well as a traditional offline analysis using the GstLAL pipeline over a 40-day stretch of data from the third observing run (O3) of the LVK collaboration. We find that our method is around 90-100% as sensitive as a traditional offline analysis.