A Multiscale Method (``Sparse Bayes Blocks'') for Revealing Sharply-Peaked High Energy Pulsars

Multiscale models are the best basis for elucidating both very fine and broad faint structures, even for X-ray and γ --ray Poison count data, esp. in 2D imaging (Starck, J. L., Pantin, E., Murtagh, F., 2002, PASP, 114,1051). But one sees at once they are also well-suited to describe 1D structures such as pulsar light-curves, with their extreme sharp peaks and occasional broad bridges between them (Thompson 2001/astro-ph/0101039). Hence, we have derived a statistic for detecting pulsars at X- or gamma-ray energies with the `spikiness' explictly built in. The best current methods, being Fourier-transform based, may not be optimal for detecting this class of `spiky' light-curve (Z_n²; e.g. de Jager et al./astro-ph/0010179 and references therein). Instead, we use one or a very few `Bayes Blocks' (Scargle/astro-ph/9711233) of arbitrary height and width to represent the light-curve; then derive an optimum statistic (likelihood ratio) for testing against flatness via Bayesian Inference. Preliminary Monte Carlo tests show that it works as well or better than a standard Z_6² statistic for a range of standard sharply peaked light-curves, especially at low signal-to-noise (Connors and Carramiñana, 2002, SCMAIII Proceedings). Here, we demonstrate it on known pulsars using CGRO/EGRET and COMPTEL data. Many of the bright unidentified sources in the gamma-ray sky may also turn out to be relatively nearby (Gould-belt: Gehrels et al2000, Nature, 404, 363) radio-quiet, gamma-ray loud pulsars, such as Geminga (Halpern and Holt 1992, Nature, 357, 222; Bertsch et al 1992, Nature, 357, 306). Can a new high-resolution algorithm help illuminate the identities of some of these? Funded in part by AISR ``AS-DATA'', with J. Scargle and V. Kashyap; and AISR ``Astrostatistics Recipes in Python'', with T. Loredo and T. Oliphant. AC thanks the hospitaltiy of Wellesley College, and the Harvard Astrostatistics Working Group.