Inferring the core-collapse supernova explosion mechanism with gravitational waves

A detection of a core-collapse supernova (CCSN) gravitational-wave (GW) signal with an Advanced LIGO and Virgo detector network may allow us to measure astrophysical parameters of the dying massive star. GWs are emitted from deep inside the core, and, as such, they are direct probes of the CCSN explosion mechanism. In this study, we show how we can determine the CCSN explosion mechanism from a GW supernova detection using a combination of principal component analysis and Bayesian model selection. We use simulations of GW signals from CCSN exploding via neutrino-driven convection and rapidly rotating core collapse. Previous studies have shown that the explosion mechanism can be determined using one LIGO detector and simulated Gaussian noise. As real GW detector noise is both nonstationary and non-Gaussian, we use real detector noise from a network of detectors with a sensitivity altered to match the advanced detectors design sensitivity. For the first time, we carry out a careful selection of the number of principal components to enhance our model selection capabilities. We show that with an advanced detector network we can determine if the CCSN explosion mechanism is driven by neutrino convection for sources in our Galaxy and rapidly-rotating core collapse for sources out to the Large Magellanic Cloud.

Figure 1

Time domain GW strain for representative models of rotating core collapse (top panel) and neutrino-driven (bottom panel), as seen by an equatorial observer at 10 kpc, drawn from the RotCC and C&S waveform catalogs, respectively [51, 58]. We note that the typical GW strain from rotating core collapse is roughly an order of magnitude larger than the typical GW strain from neutrino-driven explosions. In addition, the typical GW signal duration is roughly an order of magnitude longer for neutrino-driven explosions than for rotating core collapse.

Figure 2

Time domain $h_{+}$ GW strain for representative 3D models of rotating core collapse (top panel) and neutrino-driven convection (bottom panel), as seen by an equatorial observer at 10 kpc, drawn from the Scheidegger et al. [59] and Ott et al. [35] waveform catalogs.

Figure 3

(Left) The first four PCs for the RotCC model. (Right) As for the left, but for the C&S model. The first few PCs represent the most common features of the waveforms used in the analysis. A larger number of PCs is needed to represent the broad set of features in waveforms from the C&S model. The main feature of the RotCC model PCs is the spike at core bounce.

Figure 4

The log $B_{S,N}$ values for an increasing number of PCs for the RotCC (top panel) and C&S (bottom panel) models using five representative waveforms from each mechanism. Log $B_{S,N}$ increases as more PCs are added and more information is gained about the injected signal. Log $B_{S,N}$ will decrease if the model becomes too complex.

Figure 5

Response of SMEE to 1000 instances of simulated and recolored aLIGO and AdVirgo design sensitivity noise for RotCC with six PCs (top panel) and C&S with nine PCs (bottom panel). Transient noise artifacts and lines in the real data can increase log $B_{S,N}$ and the standard deviation of the noise response.

Figure 6

Log $B_{\mathtt{RotCC}-\mathtt{C}\&\mathtt{S}}$ for waveforms injected from the RottCC and C&S catalogs. (a) At 2 kpc and the sky position of the Galactic center, the explosion mechanism is correctly determined for all $1437/1440$ detected waveforms. (b) At 10 kpc $1198/1440$, waveforms are detected, and their explosion mechanism is correctly determined. (c) Almost all the C&S waveforms have a SNR too small for them to be detected at 20 kpc. (d) Distance of 50 kpc and sky position of the Large Magellanic Cloud.

Figure 7

Log $B_{\mathtt{RotCC}-\mathtt{C}\&\mathtt{S}}$ for five extra waveforms representing each explosion mechanism. (a) At 2 kpc, all extra magnetorotational mechanism waveforms can be distinguished from noise, and their explosion mechanism is correctly determined. For the extra neutrino mechanism waveforms, only the explosion mechanism of the yak waveform is correctly determined. (b) At 10 kpc, all extra neutrino mechanism waveforms cannot be distinguished from noise. (c) At 20 kpc, $45/100$ injected extra magnetorotational waveforms can be distinguished from noise, and their explosion mechanism is correctly determined. (d) The correct explosion mechanism is determined for all extra magnetorotational waveforms distinguishable from noise ($27/100$) at 50 kpc.