Data release for event GW170817

This page has been prepared by the LIGO Scientific Collaboration (LSC) and the Virgo Collaboration.

The event occurred at GPS time 1187008882.43 == August 17 2017, 12:41:04.43 UTC.

The event was observed by data from the LIGO Hanford (H1), LIGO Livingston (L1) and Virgo (V1) detectors.

Because the Binary Neutron Star signal spends much more than 32 seconds in the detector's frequency band, data products of 32 seconds in duration are not being distributed with GW170817.

This page serves as a supplement to the paper GW170817: Observation of Gravitational Waves from a Binary Neutron Star Inspiral, available from LIGO DCC and doi:10.1103/PhysRevLett.119.161101.
Companion papers are also available.
This dataset has DOI doi:10.7935/K5B8566F.
The GW170817 science summary gives an overview of the event.
The one page factsheet lists statistics.
Parameter estimation samples - see results from two publications: P1800115 | P1800061
The technical details page describes details of the data set.
Contact us with any questions.

Gravitational-Wave Strain Data

Data usage notes Please Read This First!
The data are provided in three formats. HDF5 , Frame (.gwf), and gzipped ascii text. Many data analysis environments can read in data from HDF5 files, including Python (see the h5py package), MATLAB, C/C++, and IDL.
Virgo data are noisier at frequencies above 2 kHz (See note on technical details page)
Tutorials provide an introduction to strain data.
Software libraries are available for data analysis.

Strain data after noise subtraction

After data collection, several independently-measured terrestrial contributions to the detector noise were subtracted from the LIGO data using Wiener filtering. This subtraction removed calibration lines and 60 Hz AC power mains harmonics from both LIGO data streams. At times near GW170817, the sensitivity of LIGO-Hanford was particularly improved by the subtraction of laser pointing noise; several broad peaks in the 150 - 800 Hz region were effectively removed, increasing the Binary Neutron Star horizon distance of that detector by 26%.

In addition, a short instrumental noise transient appeared in the LIGO-Livingston detector 1.1 s before the coalescence time of GW170817. This transient noise, or glitch, produced a very brief (less than 5ms) saturation in a digital-to-analog converter. This glitch has been removed from the noise subtracted data. For reference:

Notes:

Below are links to strain h(t) time series centered at GPS 1187008064.
All cleaned data files have been shortened to a total duration of 2048 seconds.
The H1 and L1 noise subtraction procedure introduces excess noise at frequencies below 20 Hz. Users should restrict analysis to frequencies above 20 Hz.
The H1 cleaned data include times corrupted by a window function used as part of the cleaning procedure. This impacts roughly the last 150 seconds of the H1 cleaned data.
Due to V1's smaller coupling factor to gravitational waves from the sky direction of GW170817, and the lower detector sensitivity, the signal is below the noise level and is thus not visible in the Virgo data.
The files and some technical notes may all be found in LIGO P1700349
md5 checksums may be used to ensure files are correctly downloaded.

After noise subtraction, 4096 Hz	H1	L1	V1
2048 seconds (event signal reaches peak amplitude 1842.43 seconds ± 30 msec from start†)	hdf5 gwf txt.gz	hdf5 gwf txt.gz	hdf5 gwf txt.gz

After noise subtraction, 16384 Hz	H1	L1	V1
2048 seconds (event signal reaches peak amplitude 1842.43 seconds ± 30 msec from start†)	hdf5 gwf txt.gz	hdf5 gwf txt.gz	hdf5 gwf txt.gz

† The ± 30 msec window for peak amplitude is due to different arrival time in each of the three detectors.

Strain data before noise subtraction

These files contain data before noise subtraction. These data were produced in low-latency, and used by the search pipelines to evaluate event significance.

A short instrumental noise transient appeared in the LIGO-Livingston detector 1.1 s before the coalescence time of GW170817. This transient noise, or glitch, produced a very brief (less than 5ms) saturation in a digital-to-analog converter. This glitch is present in the strain data before noise subtraction.

To evaluate the significance of GW170817, the CBC search analyses applied a window function to zero out the data around the glitch. This Tukey window uses the following parameters: zero the data +/- 0.1 seconds (0.2 seconds total in length) around the glitch time 1187008881.389 and the roll-off time is 0.5 seconds either side of the zeroed data. This inverse Tukey window is presented in Figure 2 of the GW170817 discovery paper.

Some Burst searches may use a CAT2 data quality flag to remove this glitch from search results. There is an active Burst CAT2 data quality flag at LIGO-Livingston starting at 1187008880 and extending to 1187008884.

Notes:

Below are links to strain h(t) time series centered at GPS 1187008882.
The files and some technical notes may all be found in LIGO P1700337
md5 checksums may be used to ensure files are correctly downloaded.

Before noise subtraction, 4096 Hz	H1	L1	V1
4096 seconds (event signal reaches peak amplitude 2048.43 seconds ± 30 msec from start†)	hdf5 gwf txt.gz	hdf5 gwf txt.gz	hdf5 gwf txt.gz

Before noise subtraction, 16384 Hz	H1	L1	V1
4096 seconds (event signal reaches peak amplitude 2048.43 seconds ± 30 msec from start†)	hdf5 gwf txt.gz	hdf5 gwf txt.gz	hdf5 gwf txt.gz

† The ± 30 msec window for peak amplitude is due to different arrival time in each of the three detectors.

Sky Localization

Source localization data for GW170817 are available as LIGO G1701985

Rapid triggers from LIGO data are available as GCN notices for GW170817

A python library for reading the FITS DATA is healpy.
See example code to work with source localization data
Collections of simulated skymaps are available here and here.

See the About the Detectors Page to learn about the instruments and collaborations.

The technical details provides detailed usage notes.

Feel free to contact us with any questions.

Revision History

October 19, 2017: Added note regarding glitch in L1 data.
June 1, 2018: Added note about window function visible in last 150 seconds of H1 cleaned data.
Aug 28, 2018: Added link to PE samples