Determining the core-collapse supernova explosion mechanism with current and future gravitational-wave observatories

Gravitational waves are emitted from deep within a core-collapse supernova, which may enable us to determine the mechanism of the explosion from a gravitational-wave detection. Previous studies suggested that it is possible to determine if the explosion mechanism is neutrino-driven or magneto-rotationally powered from the gravitational-wave signal. However, long duration magneto-rotational waveforms, that cover the full explosion phase, were not available during the time of previous studies, and explosions were just assumed to be magneto-rotationally driven if the model was rapidly rotating. Therefore, we perform an updated study using new 3D long-duration magneto-rotational core-collapse supernova waveforms that cover the full explosion phase, injected into noise for the Advanced LIGO, Einstein Telescope and NEMO gravitational-wave detectors. We also include a category for failed explosions in our signal classification results. We then determine the explosion mechanism of the signals using three different methods: Bayesian model selection, dictionary learning, and convolutional neural networks. The three different methods are able to distinguish between neutrino-driven explosions and magneto-rotational explosions, even if the neutrino-driven explosion model is rapidly rotating. However they can only distinguish between the nonexploding and neutrino-driven explosions for signals with a high signal to noise ratio.

Figure 1

An example of each of the four types of CCSN waveforms used in this study, shown with time-frequency plots (or spectrograms). Top left is an example chirplet from the training set with a frequency of 600 Hz and quality factor $Q$ 300. Top right is the neutrino driven explosion mechanism waveform model y20 from [41]. Bottom left is the magneto-rotational mechanism waveform model A39 from [51]. Bottom right is the nonexploding waveform model mesa20_pert from [33].

Figure 2

The amplitude spectral density (ASD) curves for the Advanced LIGO design sensitivity [1], Einstein Telescope [13] and NEMO [12] gravitational-wave detectors. Einstein Telescope has a similar frequency band to LIGO but it is more sensitive. NEMO has good high frequency sensitivity, but it has poor sensitivity at low frequencies.

Figure 3

The classification results for the LIGO detector. From left to right are the results for Bayesian model selection, dictionary learning, and 2D CNN. The Bayesian model selection and 2D CNN methods were able to correctly classify every chirplet signal. The dictionary learning method produces the best results for the nonexploding models.

Figure 4

The classification results for the ET detector. From left to right are the results for Bayesian model selection, Dictionary Learning, and 2D CNN. The results for all three methods are similar to those obtained for Advanced LIGO.

Figure 5

The classification results for the NEMO detector. From left to right are the results for Bayesian model selection, dictionary learning, and 2D CNN. In the NEMO detector, the magneto-rotational waveforms and the chirplets are classified with a high accuracy. It is more difficult to distinguish between neutrino and nonexplosions as the SASI modes are not visible due to the detectors poor low frequency sensitivity.