Stellar collisions in the vicinity of supermassive black holes

Abstract

Galactic nuclei are unique environments in which dense populations of stars orbit supermassive black holes (SMBHs) at velocities that can reach a few per cent of the speed of light due to the SMBH's gravitational influence. While tidal disruption events (TDEs) have become a common explanation for many luminous flares which originate from this region, direct collisions between such high-speed stars must also occur. Whether those collisions generate transients bright enough to potentially be confused as supernovae (SNe) or TDEs--and, if so, how they might be uncovered in the vast amounts of data generated by modern surveys--forms the central question of this dissertation. To address it, we combine semi-analytic rate calculations, simulations of debris fallback and accretion, and a machine-learning pipeline that is scalable to next-generation surveys of transient phenomena. This study begins by deriving differential and total collision rates in galactic nuclei. We consider the stellar mass function, velocity distributions, and galactic density profiles to calculate collision frequencies as a function of SMBH mass and collision ejecta energy. For galaxies hosting SMBHs with masses of $M_{\bullet}=10^8,10^9,10^{10},\Msun$, the resulting rates are $\Gamma=2.2\times10^{-3},2.2\times10^{-4},4.7\times10^{-5}$ yr$^{-1}$ collisions, respectively. We additionally borrow from well-established supernovae models to calculate basic light curves from the collisions; we find that these events can yield luminosities roughly equal to or greater than that from SNe, but the light curves are expected to decay much faster, making serendipitous detection unlikely. The subsequent fate of the post-collision debris is studied via numerical simulation. Each collision ejecta is represented by millions of freely moving test particles under Schwarzschild geometry; an additional viscous time delay is added to each particle's trajectory assuming an $\alpha$-disk model. Summing over all particles yields mass-accretion rates which are converted to luminosity to generate light curves corresponding to the accretion of the post-collision debris. The light curves that emerge display diverse morphologies--single peaks, double peaks, plateaus, and more--with behavior that depends on physical parameters such as the SMBH mass and initial distance from the SMBH. We find that some of our simulated light curves exhibit behavior that has been previously associated with unexplained TDE phenomena, such as unusual late-time evolution and repeated flares. We move on to conducting a systematic search for observed nuclear transients that resemble accretion flares from stellar collision debris. We generate a bank containing tens of thousands of simulated light curves with varied SMBH mass, initial distance from the SMBH, and relative velocity orientations. In addition, we have two scaling parameters on the $x$- and $y$- axes, which correspond to a time offset and varied ejecta mass. Observational candidates are obtained from the Lasair alert broker’s Zwicky Transient Facility (ZTF) stream, filtered for nuclear, blue, well-sampled light curves with high signal-to-noise ratio. We implement an approximate nearest neighbor (ANN) algorithm using the \texttt{annoy} (Approximate Nearest Neighbors Oh Yeah) library, which quickly and efficiently matches each observational candidate to a simulated light curve. Using this pipeline, 26 ZTF transients are found to align well with stellar collision accretion flare models. Host-galaxy photometry and spectroscopy are used to infer SMBH masses whenever possible. Some candidates are found to have $M_{\bullet}\gtrsim10^8,M_{\odot}$, supporting a collision interpretation. Stellar collisions in galactic nuclei therefore represent a plausible and previously overlooked source of nuclear transients. A collision releases $10^{49}-10^{51},\mathrm{erg}$ promptly and drives an accretion flare whose properties depend on SMBH mass and galactocentric radius at the time of the collision. The combined analytical, numerical, and data-driven methods presented here provide both predictive observables and a practical identification tool that scales to next-generation survey volumes. Future work should include magnetohydrodynamic simulations to refine debris structure and accretion physics, the integration of anomaly-detection networks or deep-learning classifiers to improve recall of rare events, and systematic spectroscopic follow-up to separate collision flares from partial TDEs. With the forthcoming data volume from the Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST), the tools developed in this dissertation offer a timely approach for recognizing and studying stellar collision transients in the center of galaxies.