Page 255 - Special Topic Session (STS) - Volume 1
P. 255

STS426 Guillaume B.

                         A statistically robust approach to the detection
                              of astrophysical transient and periodic
                                            phenomena
                                         Guillaume Belanger
                    European Space Agency (ESAC), Madrid, Spain – gbelanger@sciops.esa.int

            Abstract
            Transient  phenomena  are  interesting  and  potentially  highly  revealing  of
            details about the processes under study that could otherwise go unnoticed. It
            is therefore important to maximise the sensitivity of the detection methods
            used. We present a general procedure based on the likelihood function for
            identifying transients that is ideally suited for real-time applications because
            it requires no grouping or pre-processing of the data. The method use all the
            information available in the data throughout the statistical decision making
            process, and is suitable for a wide range of applications. Here we consider
            those  most  common  in  astrophysics  which  involve  searching  for  transient
            sources, events or features in images, time series, energy spectra, and power
            spectra, and demonstrate the use of the method in the case of a short-lived
            quasi-periodic  oscillation  in  a  power  spectrum.  We  present  two  new
                                         2
            periodogram statistics, the ℛ  and the Ƶ , and derive two fit statistics, the K
                                                    2
                                         
            and B statistics, relevant to model fitting in frequency space.

            Keywords
            Transient phenomena; periodic phenomena; likelihood function; astrophysics;
            X-rays.

            1.  Introduction
                The  way  in  which  the  measurements  are  distributed  defines  the
            appropriate statistical treatment. Each measurement considered individually,
            and the collection of measurements as a whole, carry statistical evidence that
            can be used to assess the agreement between a given hypothesis or model
            and the data. Treating data as evidence is a powerful means to detect changes,
            differences, deviations or variations. This is done using the likelihood function.
                The detection of an event localized in time, involves identifying something
            that was not there before. Whether it rises, dwells, and decays over weeks and
            months  like  a  supernova,  or  whether  it  just  appears  and  disappears  in  a
            fraction of a second like a -ray burst; whether it manifests as a complete
            change of shape of the energy spectrum during a state transition in a black
            hole, or as the short-lived emission line from an accretion event; whether it
            comes as a sudden change of spectral index in the power spectrum or as the
            appearance  of  an  ephemeral  quasi-periodic  oscillation  (QPO);  all  of  these


                                                               244 | I S I   W S C   2 0 1 9
   250   251   252   253   254   255   256   257   258   259   260