Fading into darkness: A weak mass ejection and low-efficiency fallback accompanying black hole formation in M31-2014-DS1

TL;DR

This study investigates the formation of stellar-mass black holes through a nearby failed supernova candidate, M31-2014-DS1, using JWST mid-infrared spectroscopy and imaging alongside deep Chandra X-ray observations. The data reveal a highly obscured remnant with a compact dust shell (~40–200 au) and a close-in molecular gas layer (~0.1 M⊙ at ~40–80 au), all while the central source continues to fade to about 7–8% of the progenitor luminosity. Modeling indicates a low-energy envelope ejection of ~10^46 erg and long-term fallback accretion onto a nascent BH that is highly inefficient or radiatively dim, consistent with the non-detection in X-rays due to substantial obscuration. The findings provide robust empirical support for a BH-formation pathway involving weak explosions and prolonged fallback, offering a valuable nearby benchmark for this elusive stellar death channel.

Abstract

Stellar-mass black holes (BHs) can form from the near-complete collapse of massive stars, causing them to abruptly disappear. The star M31-2014-DS1 in the Andromeda galaxy was reported to exhibit such a disappearance between 2014 and 2022, with properties consistent with the failed explosion of a $\approx 12 - 13$ M$_\odot$ yellow supergiant leading to the formation of a $\approx 5$ M$_\odot$ BH. We present mid-infrared (MIR) observations of the remnant obtained with the James Webb Space Telescope (JWST) and X-ray observations from the Chandra X-ray Observatory in 2024. The JWST MIRI/NIRSpec data reveal an extremely red source, showing strong blueshifted absorption from molecular gas (CO, CO$_2$, H$_2$O, SO$_2$) and deep silicate dust features. Modeling the dust continuum confirms continued bolometric fading of the central source to $\log(L/L_\odot)\approx3.88$ ($\approx7-8$% of the progenitor luminosity), surrounded by a dust shell spanning $\approx40-200$ au. Modeling of the molecular gas indicates $\sim 0.1$ M$_\odot$ of gas expanding at $\approx 100$ km s$^{-1}$ near the inner edge of the dust shell. No X-ray source is detected down to a luminosity limit of $L_X\lesssim1.5\times10^{35}$ erg s$^{-1}$. We show that the panchromatic observations are explained by (i) a low-energy ($\approx10^{46}$ erg) ejection of the outer H-rich progenitor envelope and (ii) a fading central BH powered by inefficient ($\sim0.1$% in mass) accretion of loosely bound fallback material. The analysis robustly establishes the bolometric fading of M31-2014-DS1 and provides the first cohesive insights into BH formation via low-energy explosions and long-term fallback.

Figures (6)

• Figure 1: (Left) JWST/MIRI RGB composite image of M31-2014-DS1 (color channels shown in the label). The spatial scale and orientation of the image are shown, and the source position is marked with a white crosshair. (Right) The evolution of the SED of M31-2014-DS1. The empty circles show the progenitor SED measured from 2005-2012 data, with the dashed line showing the best-fit DUSTY model. The filled circles show the SED of the source from 2022-2023, along with its DUSTY model as dot-dashed lines. The JWST data from NIRSpec, MIRI LRS and MIRI imaging from December 2024 are shown as colored lines (when detected at $>10\sigma$ significance; see legend), along with its best-fit DUSTY model as black lines. The source was not detected in NIRSpec/G140H observations; we show $10\sigma$ upper limits binned to $0.3\,\mu$m intervals. Some best-fit parameters derived from dust continuum modeling of the JWST data are shown (see also Table \ref{['tab:params']}).
• Figure 2: Comparison of M31-2014-DS1 to other dusty transients. ( Left) Comparison of the MIR JWST spectrum of M31-2014-DS1 to other dusty transients (indicated in the same color as the text along with the phase of observation) -- an optically discovered luminous red nova (M31-LRN-2015; Karambelkar2025Blagorodnova2020), a dusty Galactic stellar merger from an evolved primary star (OGLE-2002-BLG-360; Steinmetz2025Tylenda2013) and the failed SN candidate NGC 6946-BH1 Beasor2023Kochanek2023Adams2017. The flux scales have been arbitrarily shifted for visualization. ( Right) Comparison of the bolometric light curves of the sources relative to the estimated progenitor luminosity $L_{\rm prog}$ (the dashed line indicates a ratio of unity), with the same labeling scheme.
• Figure 3: Modeling of the molecular gas features in the NIRSpec and LRS data using a model involving a gas slab in front of the dust photosphere (see text). ( Left) Zoom-in of the best-fit model for the continuum-normalized NIRSpec/G395H data (no significant absorption is detected at shorter wavelengths). The dominant absorption features are highlighted. The inset shows two example velocity profiles (relative to M31 disk velocity $v = -200$ km s$^{-1}$ in the barycentric frame; dashed vertical line) of absorption features showing excess red emission above the continuum (dashed horizontal line), indicative of weak P-Cygni like profiles. ( Right) Corresponding fit in the MIRI/LRS data between $5.2 - 7.7\,\mu$m with the same color scheme.
• Figure 4: (Left) Comparison of the cumulative gas mass profile inferred from JWST observations to model predictions for mass ejection and fallback due to impulsive energy injection. The blue shaded region shows the estimated mass and radius of the gas shell surrounding the remnant and the red vertical dashed line shows the inferred radius of the inner photosphere. The solid, dashed and dotted lines show the predicted cumulative radial mass profile for three different energies ($E_{46} = 0.4, 1, 5$) with line styles indicating the adopted velocity profile as in the right panel. The bold sections indicate radii where the material is unbound (total energy $> 0$), while thin sections indicate bound material. The mass profiles at $\lesssim 1$ au (shown in the gray region) are not tracked in the simulation since they have fallen back. (Right) Comparison of the long-term bolometric decay of M31-2014-DS1 to our model for late-time mass accretion. The black and hollow circles indicate bolometric luminosity estimates from SED modeling and bolometric corrections to WISE data reported in De2026, respectively. The red point denotes the measurement from JWST reported in this paper. The gray shaded region shows the luminosity estimate for the progenitor star and the orange horizontal dashed line shows the Eddington luminosity ($L_{\rm Edd}$) for a $5$ M$_\odot$ BH. We adopt the same explosion time and radiative efficiency ($5$%) as in De2026. The line styles are as in the left panel (see legend); the case of $E_{46} = 0.4$ is not shown since there is nearly no unbound material. An accretion efficiency of $f_{\rm acc}/f_{\rm an} = 0.1$ is nominally adopted in the models, but we show an example of $f_{\rm acc}/f_{\rm an} = 1$ for the case of $E_{46} = 0.4, \alpha = 1$ as the black dashed line.
• Figure 5: Schematic model of mass ejection and fallback in the remnant of M31-2014-DS1. We show the inferred properties of the gas and dust shell surrounding the remnant, likely consisting of low kinetic energy gas ejected due to neutrino mass loss and inefficient accretion. The black arrows show the inferred direction of motion for the different components. The inferred properties of the inner photosphere and central accreting BH are shown (see text marked by gray arrows).