ML Tool collAIder Predicts 10% Stripped Stars from Collisions in Galactic Center
Preprint models stellar collisions via ML in the Galactic center, finding 10% stripped stars and substantial gas injection toward Sgr A*. Evidence from emulator trained on hydro simulations; key limit is grid coverage and isolated-encounter assumption. Strengthens case for collision-fed accretion in galactic nuclei.
The study bridges 3D hydrodynamic simulations with a dynamical model of the Galactic center using the new collAIder machine learning emulator. Researchers tracked collision outcomes across parameters of mass, age, impact parameter and velocity, finding that repeated high-speed encounters gradually strip stars before rare head-on disruptions. Gas release is concentrated within 0.1 pc, with 90% retained in the cluster.
This work extends earlier N-body and SPH studies of the S-star cluster and G-objects by quantifying cumulative mass loss rather than single events. It aligns with observed young star populations near Sgr A* and suggests collision-driven gas could contribute to the observed accretion flow, an effect previously underestimated in purely gravitational models.
Future multi-messenger observations with GRAVITY+ and JWST will test these rates. Refined collAIder training on additional metallicities and binary interactions could tighten predictions of black hole feeding and stellar remnant distributions.
Limitations include the emulator's dependence on the underlying hydrodynamic grid and the assumption of isolated two-body encounters without cluster-wide feedback.
Sand et al.: Within 3 years, JWST spectroscopy will detect collision-ejected gas signatures at >3 sigma above background within 0.05 pc of Sgr A*.
Sources (2)
- [1]Primary Source(https://arxiv.org/abs/2609.25318)
- [2]Supporting Source(https://arxiv.org/abs/2305.13345)