A novel approach to identify blazar emission states using clustering algorithms

Even after decades of multi-wavelength (MWL) observations, blazars still remain mysterious objects. Their extreme variability and variety of emission characteristics observed during different time periods make it hard to understand the fundamental processes behind their emission. Thus, a robust identification and characterization of the different emission states among blazars is vital to investigate the underlying processes causing the observed emission. In this contribution, we present a novel technique to determine emission states across MWL lightcurves (LCs) of blazars using a clustering algorithm. Using the Extreme Deconvolution algorithm, we apply a Gaussian Mixture model to the 12-year long-term LC of one of our archetypal blazars, Mrk 501. The two main advantages of the method are that, compared to more conventional methods, such as the Bayesian block algorithm, it considers multiple wavebands simultaneously and it is not dependent on the order in time of the data points. This allows to assign data points to the same emission state even though they are separated by other states in time. The well sampled gamma-ray, X-ray and radio LCs used as input allow to identify six clusters. The clustering is mainly driven by the X-ray flux, showing different levels of quiescent, intermediate and high flux states. However, the radio flux reveals a more complicated pattern, dividing some of the X-ray flux levels in low and high-radio flux states. This suggests that multiple emission regions maybe responsible for the radio to gamma-ray flux.