Dissecting stellar chemical abundance space with t-SNE

In the era of large-scale Galactic astronomy and multi-object spectroscopic stellar surveys, the sample sizes and the number of available stellar chemical abundances have reached dimensions in which it has become difficult to process all the available information in an effective manner. In this paper we demonstrate the use of a dimensionality-reduction technique (t-distributed stochastic neighbour embedding; t-SNE) for analysing the stellar abundance-space distribution. While the non-parametric non-linear behaviour of this technique makes it difficult to estimate the significance of any abundance-space substructure found, we show that our results depend little on parameter choices and are robust to abundance errors. By reanalysing the high-resolution high-signal-to-noise solar-neighbourhood HARPS-GTO sample with t-SNE, we find clearer chemical separations of the high- and low-[α/Fe] disc sequences, hints for multiple populations in the high-[α/Fe] population, and indications that the chemical evolution of the high-[α/Fe] metal-rich stars is connected with the super-metal-rich stars. We also identify a number of chemically peculiar stars, among them a high-confidence s-process-enhanced abundance-ratio pair (HD 91345/HD 126681) with very similar ages and v X and v Y velocities, which we suggest have a common birth origin, possibly a dwarf galaxy. Our results demonstrate the potential of abundance-space t-SNE and similar methods for chemical-tagging studies with large spectroscopic surveys.