A New Statistical Approach to the Optical Spectral Variability in Blazars

We present a spectral variability study of a sample of about 25 bright blazars, based on optical spectroscopy. Observations cover the period from the end of 2008 to mid 2015, with an approximately monthly cadence. Emission lines have been identified and measured in the spectra, which permits us to classify the sources into BL Lac-type or FSRQs, according to the commonly used EW limit. We have obtained synthetic photometry and produced colour-magnitude diagrams which show different trends associated with the object classes: generally, BL Lacs tend to become bluer when brighter and FSRQs become redder when brighter, although several objects exhibit both trends, depending on brightness. We have also applied a pattern recognition algorithm to obtain the minimum number of physical components which can explain the variability of the optical spectrum. We have used NMF (Non-Negative Matrix Factorization) instead of PCA (Principal Component Analysis) to avoid un-realistic negative components. For most targets we found that 2 or 3 meta-components are enough to explain the observed spectral variability