Multi-scale time- and frequency-domain likelihood analysis with photon weights

Abstract

We present an unbinned likelihood analysis formalism employing photon weights -- the probabilities that events are associated with a particular source. This approach is applicable to any photon-resolving instrument, and thus well suited to high-energy observations; we focus here on GeV $γ$-ray data from the Fermi Large Area Telescope. Weights connect individual photons to the outputs of a detailed, expensive likelihood analysis of a much larger data set. The weighted events can be aggregated into arbitrary time spans ranging from microseconds to years. Such retrospective grouping permits time- and frequency-domain analysis over a wide range of scales and enables characterization of disparate phenomena like blazar flares, $γ$-ray bursts, pulsar pulses, novae, $γ$-ray binaries, and other variable sources. To demonstrate the formalism, we incorporate photon weights into the Bayesian blocks algorithm and perform a hierarchical time scale analysis of 3C 279 activity. We analyze pulsar pulse profiles and estimate the unpulsed emission level and the optimal division of the data into on- and off-pulse intervals. We extend the formalism to Fourier analysis and derive estimators for power spectra, used to search for and characterize periodic sources. We show how the Fast Fourier transform can be used to probe orbital periods as short as a minute and we discuss mitigation of spurious signals. Our final example combines time- and frequency- domain analysis to jointly characterize the flares and orbital modulation of Cygnus X-3, yielding the strongest detection of the orbital signal ($>$13$σ$) to date. Finally, we discuss extensions of the work to other GeV sources and to X-ray and TeV observations.