EMPCA and Cluster Analysis of Quasar Spectra: Application to SDSS Spectra
Abstract
Accurate modeling of the quasar continuum is necessary to measure and analyze absorption lines. But quasar continua, in particular the emission lines, vary from object to object. Patterns in the variations allow a spectral principal component analysis (SPCA) approach using large samples of quasar spectra, e.g., from the SDSS. Then, a small number of the derived principal component spectra can be used to reconstruct an arbitrary quasar's continuum.A problem with this approach is that the number of principal components required to model an arbitrary quasar, usually 8 to 20 in the literature, is large. One reason why so many components are required is that SPCA implicitly assumes that spectra bins are independent. Quasar emission lines are spread over a range of spectral bins, and more importantly, can sometimes be blueshifted. So while the intrinsic variability may only be a function of a few physical parameters, the nonlinearity inherent in the variations from object to object requires a large number of prinicipal components to accurately model a quasar continuum.We present a modified approach. We perform a SPCA analysis, using an expectation-maximization algorithm by Bailey et al. 2012, which takes into account uncertainties and missing data. We project the sample spectra on the resulting eignevectors to obtain the projection coefficients. Reasoning that intriniscally similar spectra will have similar projection coefficients, we perform a cluster analysis on the projection coefficients. The results are used to divide the sample into groups of similar spectra. A second PCA analysis is then performed on each group. We find that many fewer eigenspectra are required to accurately model the spectra in each group. We apply this approach to several samples of quasars from the SDSS.
- Publication:
-
American Astronomical Society Meeting Abstracts #229
- Pub Date:
- January 2017
- Bibcode:
- 2017AAS...22925016L