Abstract:
|
We present a novel approach to studying the diversity of galaxies by empirically determining the natural variations in observed spectra. Using nonlinear dimensionality reduction techniques inspired by ideas from spectral graph theory, we apply a method that simultaneously takes into account the measurements of spectral lines as well as the continuum shape. Unlike Principal Component Analysis, this approach does not assume that the Euclidean distance is a good global measure of similarity, but rather is based completely on local differences. The power of the method is demonstrated on the Sloan Digital Sky Survey's Main Galaxy Sample, where the derived embeddings of spectra carry an unprecedented amount of information. At first sight, the main features strongly correlate with star formation rate and clearly separate active galactic nuclei. Second order parameters describe line strengths and their interdependencies. A locally-biased semi-supervised enhancement of the technique empowers one to focus on the typical variations around specific objects of interest, enabling new discoveries and a detailed understanding of the different mechanisms at play that otherwise would be overwhelmed
|