Stellar twins in RAVE with Gaia

To search for twins in this large data set, we apply the t-student stochastic neighbour embedding projection that distributes the data according to their spectral morphology on a two-dimensional map. From this map, we choose the twin candidates for which we calculate a {chi}^2^ to select the best sets of twins.

Our results show a competitive performance when compared to other model-dependent methods relying on stellar parameters and isochrones. The power of the method is shown by finding that the accuracy of our results is not significantly affected if the stars are normal or peculiar since the method is model free. We find twins for 60 per cent of the RAVE sample that are not contained in Tycho-Gaia Astrometric Solution (TGAS) or that have TGAS uncertainties that are larger than 20 per cent.