✨Early-Warning Alerts

During O3, automated public alerts for CBC events have been sent within as little as 2 minutes after merger. However, some BNS mergers are detectable for up to tens of seconds before merger during the event’s inspiral phase during which the signal sweeps in frequency from tens of hertz to kilohertz across LIGO’s sensitive band.

Since it is generally assumed that detectable electromagnetic (or neutrino) emission starts shortly after merger, a pre-merger gravitational-wave detection would provide early warning of an impending electromagnetic transient and might make it possible for automated follow-up facilities to capture any prompt emission from the merger environment, the jet, and other unknown activity.

We have commissioned an experimental capability to produce and distribute early warning gravitational-wave alerts up to tens of seconds before merger. We have also reduced the latency of ordinary post-merger alerts to within tens of seconds after merger.

We are conducting a trial early warning public alert infrastructure. Since production analysis had to be suspended due to COVID-19 pandemic, we are replaying an 8-day period of saved LIGO data from O3 with a random (and undisclosed) time shift. Starting on 2020-06-09 and lasting for one week, early-warning alerts arising from the replayed data will be publicly distributed as test alerts at a rate of approximately once per day.

Subscribe to Early-Warning Alerts

Early warning alerts are publicly distributed as GCN Notices just like ordinary gravitational-wave alerts (see the Alert Contents section). They have exactly the same format and content as a Preliminary GCN Notice. In particular, early warning GCN Notices have significance estimates, sky localizations, and source classifications. Also, like Preliminary GCN Notices, Early Warning GCN Notices are sent prior to vetting by humans, and are not accompanied by a GCN Circular.

Early warning alerts are differentiated from ordinary alerts by the new LVC_EARLY_WARNING GCN Notice type. If you are receiving GCN Notices via an anonymous VOEvent connection and you are using the Python sample code from the Receiving GCNs section, then all you have to do is add the notice type to your GCN handler:

import gcn

    gcn.notice_types.LVC_EARLY_WARNING,  # <-- new notice type here
def process_gcn(payload, root):
    ...  # <-- put your code here


(If you are using a non-anonymous GCN connection or one of the many other notice formats provided by GCN, then you will also need to submit a change to your GCN Notice subscription settings.)


Since these early-warning events will arise from time-shifted and replayed data and not live observations, the GCN notices will be flagged as test events by setting the VOEvent role="test" attribute.

Detection Method

Three CBC search pipelines are participating in early-warning alerts: GstLAL, SPIIR, and MBTA. See the Online Pipelines section for details on these analyses. Localizations will be produced with BAYESTAR; see the Sky Localization and Parameter Estimation section for details.


The following material describes the detection method and simulated observing capabilities with GstLAL, but is also broadly applicable to SPIIR and MBTA.

BNS signals sweep up smoothly in frequency for a few minutes across the Advanced LIGO band. In that time, they may accumulate enough SNR to be detected before merger. A GW170817-like system with a total network SNR of 32 will already accumulate an SNR of 11 by the time the signal sweeps up to 30 Hz, about a minute before merger.

Time evolution of SNR for a GW170817-like system

The time evolution of the gravitational-wave frequency and the cumulative SNR for a GW170817-like BNS system.

The early warning search is a matched-filter search that uses templates that have been truncated at a selection of end frequencies—or equivalently, cut off at a selection of times before merger. The early warning template bank spans (source frame) component masses between 1 and 2 \(M_\odot\) and chirp masses between 0.9 and 1.7 \(M_\odot\). The end frequencies are 29 Hz, 32 Hz, 38 Hz, 49 Hz, and 56 Hz, corresponding to about 60 s, 45 s, 30 s, 15 s, and 10 s before merger.

Early warning events passing a FAR threshold of one per week are sent as alerts.

Source Classification

The automated source classification and properties have not been trained or tested extensively for early warning alerts. However, the early warning analysis is only sensitive to BNS-mass mergers. As a result, the favored source class in early warning GCN Notices will always be either BNS or Terrestrial, with a 0% chance of NSBH or BBH. The HasNS and HasRemnant fields will always show 100%.


Sky localizations for early warning alerts are typically very coarse because the early warning analysis inherently does not make use of the full duration and bandwidth of the gravitational-wave signal. The localization improves slowly up until the last second before merger, and then converges rapidly in the last second.

The animations below show the evolution of early-warning sky maps for three representative events with different SNR values.

Final SNR





250 Mpc

210 Mpc

160 Mpc

Sky map (animated GIF)

Animation of sky map for an event with SNR=11.0

Animation of sky map for an event with SNR=18.2

Animation of sky map for an event with SNR=25.2


Localization accuracy (90% credible area)

29 Hz

Not detected

Not detected

12000 deg2

32 Hz

10000 deg2

38 Hz

9200 deg2

8200 deg2

49 Hz

2300 deg2

1000 deg2

730 deg2

56 Hz

1000 deg2

700 deg2

250 deg2

1024 Hz

10 deg2

31 deg2

5.4 deg2

Detection Rate and Localization Accuracy

In the figure below, we show predicted detection rates, distances, and localization uncertainties for simulated BNS events. Of all BNS events detected by LIGO and Virgo, only 5-30% will be amenable for sending early warning alerts. About 5% of all BNS events will be localized to an area of ~400 deg2 by ~30 seconds before merger. At the time of merger, the sky localization will be reduced to about ~1 deg2 for these events. At 60 s before merger, one event per year is expected to be localized to within 400 deg2. At 30 seconds before merger, at least one event per year is expected to be localized to within 40 deg2 and ~4 events per year are expected to be localized to within 400 deg2. By 10 seconds before merger, ~10 events per year are expected to be localized to within 400 deg2.

Cumulative distribution of localization area for early warning events

Cumulative distribution of localization accuracy for early warning events. Assuming the median BNS merger rate, the right vertical axis shows the number of expected events to be recovered per year as a function of the 90% credible area. The detectors are considered to be operating at design sensitivity in this simulation. The volume reach of the detectors at design configuration is about 6-8 times larger than the current volume reach. This means that the number of events shown in the plot here are about 6-8 times more than the number of events we can currently detect.

The figure below shows the cumulative fraction of recovered injections as a function of distance. This figure shows the distance distribution of the events recovered at various early warning frequencies.

Cumulative distribution of distance for early warning events

Cumulative distribution of distance for early warning events.