Vanishing Acts: Quantifying Black Hole Formation with the DSNB Signal

TL;DR

This work analyzes the diffuse supernova neutrino background (DSNB) as a probe of invisible, black hole-forming supernovae by exploiting spectral differences between visible (neutron star-forming) and invisible (black hole-forming) events. Using a flux model that couples the cosmic core-collapse rate, a pinched-thermal neutrino spectrum, and cosmology, the authors employ five engine models to bound the black-hole fraction and generate NS/BH spectra. They implement two statistical tests—one to detect any invisible component and another to constrain the black-hole fraction—applied to JUNO, SK-Gd, and HK, with sidebands improving background control. The results indicate that, for a representative (median) model, a 3σ detection of the invisible component could be achieved within about a decade, and higher significance is possible for more extreme engine scenarios; near-future surveys and multi-messenger observations are crucial to break degeneracies and interpret the DSNB. Overall, the paper demonstrates that early DSNB observations, in concert with optical and gravitational-wave data, can constrain the elusive population of invisible stellar deaths and illuminate the end stages of massive stars, even with low event statistics.

Abstract

The diffuse supernova neutrino background (DSNB) created by stellar core-collapses throughout cosmic history is on the verge of discovery, with SK-Gd showing early deviations from the background expectation and JUNO starting to take data. However, the interpretation of early DSNB data will face significant challenges due to degeneracies between astrophysical parameters and uncertainties in supernova neutrino modeling. We explore how complementary astronomical observations can break these degeneracies and, in this context, we investigate whether early DSNB observations can constrain invisible supernovae, which have no optical emission but are powerful neutrino sources before being swallowed by a forming black hole. Leveraging the differences in the spectra between invisible and visible supernovae, we estimate the sensitivity of 1) detecting the existence of invisible supernovae, and 2) determining the fraction of invisible supernovae. Finally, we discuss how these conclusions depend on the spectral parameters of the black hole-forming component.

Figures (13)

• Figure 1: Plot of the DSNB flux spectrum for the median fiducial model (W18), with bands for extreme models (S19.8 - W20) for both neutron star and black hole forming core-collapse supernovae.
• Figure 2: Detector layouts of SK (Left) and JUNO (Right). The red parts shown are exclusion areas, to mitigate backgrounds such as fast neutrons from spallation processes in the surrounding rock. The areas excluded for SK are $\sim$2 m from the PMTs, leading to a cylindrical fiducial volume of $22.5$ kt with $r\sim15$ m and $h\sim 32$ m SuperKSRNS2012. For JUNO, the top part with a vertical position $z>16$ m and the sides with $r=\sqrt{Z^{2}+{r^2}_{XY}}$ with $r_{XY} > 16$ m are excluded Abusleme2022.
• Figure 3: Background spectra assumed for JUNO (Left) and SK-Gd (Right). JUNO's prediction is taken from Ref. Abusleme2022 and is restricted to the DSNB observation window, while the background spectra shown for SK-Gd are based on Ref. Harada2024DSNB and include as well large sideband areas. In both plots, we group similar backgrounds with similar colors.
• Figure 4: Neutrino flux integrated from 0 - 90 MeV as a function of black hole-fraction, $f_{\mathrm{BH}}$, for both neutron star and black hole-forming supernova. The solid lines indicate interpolation between the 'true' points (dots) for each model. Outside 0.178 and 0.417, we scale using the extreme models (dotted lines).
• Figure 5: The $\Delta \chi^2$-profiles in the dependence of $f_{\mathrm{inv}}$ for 10 years of data with JUNO (Left) and SK (Right). The energy window is 12 - 30 MeV for both, hence no sidebands are used for SK. The width of the bands is given by the difference of the integrated (low) and binned (high) log-likelihood. For JUNO this difference is not visible.