Predicting the properties of black holes merger remnants with Deep Neural Networks
This improves accuracy in gravitational wave astronomy for astrophysicists, though it is incremental as it builds on existing simulation data.
The paper tackled predicting the mass and spin of black hole merger remnants using a deep neural network, achieving errors as low as 0.04% for mass and 0.3% for spin in test datasets and reducing root mean square errors by half compared to existing methods.
We present the first estimation of the mass and spin magnitude of Kerr black holes resulting from the coalescence of binary black holes using a deep neural network. The network is trained on a dataset containing 80\% of the full publicly available catalog of numerical simulations of gravitational waves emission by binary black hole systems, including full precession effects for spinning binaries. The network predicts the remnant black holes mass and spin with an error less than 0.04\% and 0.3\% respectively for 90\% of the values in the non-precessing test dataset, it is 0.1\% and 0.3\% respectively in the precessing test dataset. When compared to existing fits in the LIGO algorithm software library, the network enables to reduce the remnant mass root mean square error to one half in the non-precessing case. In the precessing case, both remnant mass and spin mean square errors are decreased to one half, and the network corrects the bias observed in available fits.