Low latency detection of massive black hole binaries

The next decade is expected to see the launch of one or more space-based gravitational wave detectors: the European-led Laser Interferometer Space Antenna (LISA), and one or more Chinese mission concepts, Taiji and TianQin. One of the primary scientific targets for these missions is the merger of black holes with masses between ${10}^3{M}_{\odot }$ and ${10}^8{M}_{\odot }$. These systems may produce detectable electromagnetic signatures in addition to gravitational waves due to the presence of gas in minidisks around each black hole, and a circumbinary disk surrounding the system. The electromagnetic emission may occur before, during, and after the merger. In order to have the best chance of capturing all phases of the emission, it is imperative that the gravitational wave signals be detected at low latency and used to produce reliable estimates for the sky location and distance to help guide the search for counterparts. Low latency detection also provides a starting point for the “global fit” of the myriad signals that are simultaneously present in the data. Here, a low latency analysis pipeline is presented that is capable of analyzing months of data in just a few hours using a laptop from the last decade. The problem of performing a global fit is avoided by whitening out the bright foreground produced by nearby galactic binaries. The performance of the pipeline is illustrated using simulated data from the LISA Data Challenge.

Figure 1

Simulated X-channel TDI data produced for the Sangria round of the LISA Data Challenge. The simulated data covers a little over one year and includes colored instrument noise, signals from millions of galactic binaries, and 15 massive black hole binary mergers. Figure courtesy of Stanislav Babak.

Figure 2

Periodograms.

Figure 3

Wavelet denoising applied to the A-channel TDI data for month 2 of the Sangria training data.

Figure 4

PSD model.

Figure 5

Evolution of the amplitude and polarization phase for source 4 (source parameters given in Table 1).

Figure 6

Spectrograms of the whitened TDI A-channel data from month 4 (upper panel) and month 5 (lower panel) of the LDC Sangria training data. The black lines indicate the time-frequency tracks of signals found by the search in each segment. Note that two of the sources that merge in month 5 were also picked up prior to the merger in month 4.

Figure 7

Spectrograms of the whitened TDI A-channel data from month 5 of the LDC Sangria training data, showing the sequential detection and removal of the binary black hole signals.

Figure 8

The chirp masses and merger times for all the sources found by the search of the Sangria training data. The circles denote the true values, while the colored dots denote the recovered values. The dots are color-coded by the SNR. All the simulated sources were accurately recovered.

Figure 9

Posterior distributions for source 5 using the full one-year dataset (in red) and just the month 4 data (in blue) where the inspiral signal was first picked up by the search. The dashed lines indicate the true values of the parameters used to simulate the data. Here the masses are in the source frame, with most of the uncertainty coming from the distance estimate, which is then mapped to a redshift estimate that is used to convert between the detector frame and source frame masses.

Figure 10

Posterior distributions for the detector frame masses and dimensionless spins for source 4 using the full PSD model (in blue) and the smooth component of the PSD model (in red). The dashed lines indicate the true parameter values used to generate the data. The full PSD model whitens away the loud galactic binaries and returns an unbiased estimate for the masses and spins. The smooth component of the PSD allows the loud galactic binaries to overlap with the black hole signal and cause significant biases in the parameter estimates.