Inferring the core-collapse supernova explosion mechanism with three-dimensional gravitational-wave simulations

A detection of a core-collapse supernova signal with an advanced LIGO and Virgo gravitational-wave detector network will allow us to measure astrophysical parameters of the source. In real advanced gravitational-wave detector data, there are transient noise artifacts that may mimic a true gravitational-wave signal. In this paper, we outline a procedure implemented in the supernova model evidence extractor (SMEE) that determines if a core-collapse supernova signal candidate is a noise artefact, a rapidly rotating core-collapse supernova signal, or a neutrino explosion mechanism core-collapse supernova signal. Further, we use the latest available three-dimensional gravitational-wave core-collapse supernova simulations, and we outline a new procedure for the rejection of background noise transients when only one detector is operational. We find the minimum signal-to-noise ratio needed to detect all waveforms is reduced when using three-dimensional waveforms as signal models.

Figure 1

Example gravitational-wave signals from CCSNe simulations. (Top left) Gravitational waves from a neutrino mechanism $15{\mathrm{M}}_{\odot }$ progenitor star simulated by Kuroda et al [32]. (Top right) Müller et al. [38] neutrino mechanism waveform with a $15{\mathrm{M}}_{\odot }$ progenitor star. (Bottom left) Representative magnetorotational waveform from Scheidegger et al. [39] with a $15{\mathrm{M}}_{\odot }$ progenitor star. (Bottom right) Neutrino mechanism simulation with a $20{\mathrm{M}}_{\odot }$ progenitor star simulated by Andresen et al. [37].

Figure 2

First four PCs used to represent each explosion mechanism. For the magnetorotational signal model, 15 waveforms from Scheidegger et al. [39] are used. The main feature of the PCs is the spike at core bounce. The neutrino mechanism signal model is made from a mixture of waveforms simulated by Andresen et al. [37], Kuroda et al. [32], and Müller et al. [38]. Turbulent features, such as convection and SASI, are the main features of the waveforms. Each PC is multiplied by an amplitude scale factor during the analysis.

Figure 3

Increase in log Bayes factors and cumulative variance as the number of PCs is increased. (a) Signal vs noise log Bayes factors for an increasing number of PCs, for five representative magnetorotational waveforms. An ideal number of PCs is reached when the log Bayes factors are no longer sharply increasing. (b) Same result for five representative three-dimensional neutrino mechanism signals. All signals are injected with an SNR of 17. (c) Variance curve for the magnetorotational waveforms and (d) variance curve for the neutrino mechanism waveforms. We choose five PCs for the magnetorotational mechanism as this corresponds to the “knee” of the variance curve. We choose eight PCs for the neutrino mechanism as this value gives a high Bayes factor for all neutrino mechanism waveforms.

Figure 4

Distributions of Bayes factors for 1000 instances of simulated aLIGO and AdV design sensitivity noise and real recolored noise. In Gaussian noise, when no signal is present, we expect SMEE to produce a signal vs noise log Bayes factor of $\sim -1$ or $\sim -2$ using the magnetorotational and neutrino signal models, respectively. In real noise, when no signal is present, we expect SMEE to produce a signal vs noise log Bayes factor of $\sim 0$ using the magnetorotational or neutrino signal models.

Figure 5

Minimum network SNR needed to detect different CCSN signals. Five representative waveforms from each explosion mechanism are injected at increasing network SNR values, and a Bayes factor is determined. In the top two figures, we use the signal models created from the two-dimensional waveforms in [9]. In the bottom two figures, we use the new signal models created from three-dimensional waveforms. The dashed line shows the threshold for a detection. We find the minimum SNR needed to detect all models is improved by using the three-dimensional waveforms.

Figure 6

(Left) Network SNR of the magnetorotational waveforms injected at 10 kpc. (Middle) SNR of the neutrino mechanism waveforms injected at 2 kpc. (Right) SNR of the background cWB triggers used in this study. The SNR depends on the antenna pattern at the time of the injection and the quality of the data.

Figure 7

Bayes factors used to determine the explosion mechanism of the signal candidates. (Left) Results for a three detector network. (Right) Results when only the Livingston detector is considered. A signal is considered as being magnetorotational if ${\text{logB}}_{\text{MagRot}-\text{Neu}}>10$ and is considered to be neutrino mechanism if ${\text{logB}}_{\text{MagRot}-\text{Neu}}<-10$.

Figure 8

Bayes factors used to determine the explosion mechanism when two signals are left out of the signal models to test robustness. Left is the results for a three detector network. Right is the results for one LIGO detector only. A signal is considered as being magnetorotational if ${\text{logB}}_{\text{MagRot}-\text{Neu}}>10$ and is considered to be neutrino mechanism if ${\text{logB}}_{\text{MagRot}-\text{Neu}}<-10$.