WITHDRAWN Stream Members Only: Data-Driven Characterization of Stellar Streams with Mixture Density Networks

Oregon Convention Center 

We introduce a new method for constructing a smooth probability density model of stellar streams using all of the available astrometric and photometric data in a joint model. Modeling the stream and background with neural network driven mixture models, our method enables a flexible and statistically sound determination of stream membership probabilities. By using neural networks our models capture the variations in the stream's path and density in a model-free way not possible with traditional mixture models. The background is similarly model-free and is flexibly applicable to any observational field. Our method requires no assumptions about the gravitational potential. Moreover, it is capable of handling data with incomplete phase-space observations, making the method applicable to the growing census of Milky Way stellar streams. As demonstration the method is applied to the streams GD-1 and Palomar 5. When applied to a population of streams, the resulting homogeneous potential-model-free catalog of membership probabilities from our model form the required input to map the Milky Way's and even external galaxies' matter and dark matter distribution on scales large and small.