Back to Volume
Paper: Spectral Principal Component Analysis of SDSS Quasars: Beyond Eigenvector 1
Volume: 419, Galaxy Evolution: Emerging Insights and Future Challenges
Page: 398
Authors: Ludwig, R. R.; Wills, B.; Greene, J. E.; Robinson, E. L.
Abstract: In 1992, Boroson & Green (hereafter BG92) discovered the so-called eigenvector 1; AGN exhibit an anti-correlation between [O III] λ5007 and optical Fe II emission, coupled with the width of broad-line Hβ, such that stronger [O III] corresponds to broader Hβ and weaker Fe II. They showed (see also Boroson 2002) that these relationships are related to black hole mass or Eddington accretion ratio, and thus to the central engine. Surprisingly, the physics at sub-parsec scales is correlated with narrow line emission arising up to hundreds of parsecs away. We revisit this fundamental relationship using the large numbers of quasar spectra provided by the Sloan Digital Sky Survey (SDSS). Our sample includes 9046 spectroscopically-identified QSOs from Data Release 5 (DR5) with a redshift range of 0.1 < z < 0.53, on which we perform spectral principal component analysis (SPCA) using the code from Francis et al. (1992). We find a subset of objects with extremely large [O III] EW that behave independent of Hβ linewidth. These objects drive nonlinear correlations among the data, and mask the correlation of FWHM of Hβ for the rest of the data set. However, these objects comprise less than a few percent of the entire data set, and thus are rare enough that they did not drive the relationships seen in previous, smaller samples such as BG92. These strong [O III] objects do not seem to differ from the data set significantly in other spectral properties, such as luminosity, redshift, line shapes, and continuum slope. Additionally, we find that SPCA is not the best way to analyze large quasar samples, given that the spectra are nonlinearly distributed, while one of the assumptions in SPCA is that the spectra can be related linearly. A solution to this problem is cutting large samples into subsets, which then behave much more normally. Our subset that includes the bulk of the original data then recovers traditional QSO behavior, such as eigenvector 1.
Back to Volume