Normalizing Flows as an Avenue to Studying Overlapping Gravitational Wave Signals

Because of its speed after training, machine learning is often envisaged as a solution to a manifold of the issues faced in gravitational-wave astronomy. Demonstrations have been given for various applications in gravitational-wave data analysis. In this Letter, we focus on a challenging problem faced by third-generation detectors: parameter inference for overlapping signals. Because of the high detection rate and increased duration of the signals, they will start to overlap, possibly making traditional parameter inference techniques difficult to use. Here, we show a proof-of-concept application of normalizing flows to perform parameter estimation on overlapped binary black hole systems.

Figure 1

Representation of our analysis framework. It is made of a preprocessing part where we build a SVD basis to filter the data, followed by a normalizing-flow-based neural network.

Figure 2

$P\text{−}P$ plots for a subset of the recovered parameters for the two events in the data. The parameters shown are representative of all the BBH parameters for the two events. In both cases, the lines align along the diagonal, showing that our method can be trusted. The legend indicates which line corresponds to which parameters. The parameters for event 1 (resp. 2) are noted $P\_1$ (resp. $P\_2$), where $P$ are the usual parameter symbols as presented in Table 1. The values between the brackets are the KS test statistic.

Figure 3

Comparison between our approach and the one from [29] for two separate events and for the chirp mass, mass ratio, right ascension, and declination. The injected values are given by the black lines. For the left event the true value is encapsulated by the posteriors of both methods, for the right event this is only the case for our method. Our posteriors are generally broader but include the injected value within the 90% confidence interval. This could be corrected by applying importance sampling on the output samples.