1. Detecting  -ray Sources Brenda Dingus dingus@lanl.gov 23 January 2006 Outline: I.Detection Techniques II.Each  -ray is an Image III.Source Detection A.Point Sources B.Extended Sources C.Variable Sources

2.  -rays Probe Nature’s Particle Accelerators HST Image of M87 (1994) Black Hole producing relativistic jet of particles Binary Neutron Star Coalescing Artist Conception of Short GRBs Spinning Neutron Star powering a relativistic wind Massive Star Collapsing into a Black Hole SuperComputer Calculation HESS TeV + x-ray TeV image of Vela Jr. Supernova Remnant Chandra x-ray image

3. Different Types of Detectors for Gamma-Ray Astrophysics High Sensitivity HESS, MAGIC, CANGAROO, VERITAS Large Aperture/High Duty Cycle Milagro, Tibet, ARGO, HAWC Low Energy Threshold EGRET/GLAST Low Duty Cycle/Small Aperture Large Effective Area Excellent Background Rejection Known Source Spectra Known Source Lightcurves Survey of Galactic Plane Large Duty Cycle/Large Aperture Space-based (Small Area) “Background Free” Sky Survey > 100 MeV Transient Sources Extended Sources Large Duty Cycle/Large Aperture Moderate Area Good Background Rejection Sky Survey >~ 1 TeV Transient Sources Extended Sources

4. Atmospheric Cherenkov Telescopes (>~50 GeV) Atmospheric Cherenkov Telescope Arrays Collect light with large mirror (>10 m dia) and focus on array of pmts Cherenkov angle in air is ~1 o so mirror must point towards source Image of the atmospheric Cherenkov light is more concentrated & elliptical for  -ray vs proton initiated showers Arrays of 3-4 mirrors improve background rejection and improve angular resolution to a few arc minutes and energy resolution of 10-20% Larger arrays of 100s of mirrors are required to survey the overhead sky Dark, clear nights limit observing time to 5-10%

6. High Energy  -ray Observatories in Space  Pre Compton Observatory –SAS2 –CosB –10-20 sources  Compton Observatory –1991-2000 –EGRET (spark chamber) –~300 sources  GLAST –Launch September 2007 –>5000 sources

7. Satellites (30 MeV to 300 GeV  -rays)   -rays interact via pair production in dispersed foils  Cosmic-ray background (mostly protons) is rejected by anticoincidence shield AND inverted V-image of electron-positron pair   -ray direction is determined by energy-weighted average of the electron and positron tracks  e+e+ e–e– calorimeter (energy measurement) particle tracking detectors conversion foil anticoincidence shield Pair-Conversion Telescope  Angular Resolution is dominated by multiple scattering (  1/Energy) at low energies and by position resolution of tracker at high energies  Energy Resolution is ~10%, but lower energies are always more probable due to source spectra which is typically dN/dE  K E -2

8. Energy Dependent Localization  The number of  -rays is small with typically < 100 per source  Use spatially unbinned likelihood analysis (infinitesimally small bins with either 0 or 1 event)  Use Energy Dependent Point Spread Function to calculate the Model in small energy intervals  Require the normalization in each energy interval to fit a power law spectrum with free parameters for the overall normalization and spectral index Diamonds show  -rays > 5 GeV. 95 % confidence intervals are Black Circle for Previous Analysis and Blue Area is New Analysis. Galactic Longitude (deg) Galactic Latitude (deg) EGRET’s Galactic Center Source

9. Point Source Survey  Source confusion will be a problem –~ 1 source/ 4 sq deg –Point Spread Function at 0.1 (1) GeV is 3.5 (0.4) deg –Typical (weak) source will have < 100  -rays detected  Most sources vary with time as much as an order of magnitude  Different Spectra also help with harder (higher energy) spectra having better localization Integral Flux (E>100 MeV) cm -2 s -1 Prediction for GLAST Detections Of Active Galactic Nuclei

10. Time Variable Sources  Small # of  -rays limits minimum variability time scale  At least 5  -rays are required to detect a source  Bayesian block statistical technique is needed to distinguish the lightcurve - GRB940217 (100sec) - PKS 1622-287 flare - 3C279 flare - Vela Pulsar - Crab Pulsar - 3EG 2020+40 (SNR  Cygni?) - 3EG 1835+59 - 3C279 lowest 5  detection - 3EG 1911-2000 (AGN) - Mrk 421 - Weakest 5  EGRET source 100 sec 1 orbit 1 day

11. Galactic Diffuse Emission   -rays are produced by interaction of cosmic rays with matter and photons in the Galaxy  Structure (e.g. molecular clouds) is comparable to the size of the  -ray point spread function  The uncertainty in the model of diffuse emission is difficult to determine, but does effect point source detection  Use maximum likelihood test with diffuse model + point source vs only diffuse model to quantify significance of point sources (need Monte Carlo to derive probability from Test Statistic) -180 -140 -100 -60 -20 20 60 100 140 180 Galactic Diffuse Model & EGRET Data (Hunter et al. 1998)

12. Water Cherenkov Extensive Air Shower Detectors 8 meters e  80 meters 50 meters Detect Particles in Extensive Air Showers from Cherenkov light created in a covered pond containing filtered water. Reconstruct shower direction from the time different photomultiplier tubes are hit. 1700 Hz trigger rate (>50 billion events/yr) mostly due to Extensive Air Showers created by cosmic rays Field of view is ~2 sr and the average duty factor is nearly 100% Milagro Cross Section Schematic

13. Angular Reconstruction Use nsec timing from each PMT hit to fit direction of primary particle Monitor angular reconstruction with the space angle difference between reconstructions of individual events with the Even vs Odd # PMTs (delEO) delEO is ~ twice the angular resolution due to the error in each subset as well as the improvement when the # of points in the fit is doubled. Median  even-odd = 1.0 o implies Gaussian  of 0.4 o for proton reconstruction Red Monte Carlo Black Data  even-odd in degrees

14. Event Images in Milagro Protons Gammas Gamma MC Data Proton MC Cut at C>2.5 to Retain 50%  and 9% protons. Size of red dots indicate # of photoelectrons detected.

15. MARS 1 (Multivariate Adaptive Regression Splines)  Predicts the values of an outcome variable given a set of independent predictor variables  Calculates probability of  vs background for all combinations of parameters  MARS Value is ln[P(signal)/P(background)] –More positive means more  -like 1 J. Friedman, “Multivariate Adaptive Regression Splines”, Annals of Statistics 19 (1991). Differential Distribution Integral Distribution

16. Combining Data with Different Cuts: Weighted Analysis Hard Cuts: NFIT>=200,C>6.0 Std Cuts: NFIT>=20,C>2.5 Excess = 60, Off = 140, S:B = 1:2.3  hadron background =~ 1x10 -5 Excess = 5410, Off = 1218288, S:B = 1:225  hadron background =~ 0.1 Weight events by Expected S:B Milagro’s Crab Signal

17. Optimal Bin Size for Point Sources: If Guassian Point Spread Function, then Radius of Bin is 1.6 x  of the Gaussian Point Spread Function If Square Bin, then chose dimensions to give same area as square bin Milagro Opt Square Bin Size = 2.1 o Point Source Search - Weighted Analysis Cygnus Region Mrk421 Crab Vicinity of the Crab  =1.03

18. Milagro Background Estimation

19. Variable Source Search Search in spatial and time domain Examine >50 time intervals from < 1 msec to 2 hrs to days, weeks, months Shortest time intervals (< 1 sec) use starting times of the single events Longer time intervals are oversampled by factor of two Monte Carlo is used to access trials penalty of oversampling For this analysis, searching and oversampling worsens sensitivity by ~ factor of 2, because ~10  result is required to give a 5  chance probability

20. Extended Source Search Vary Bin Size from 2.1 O to 5.9 O (Optimal for ~5 O source) As bin size increases to > 6 O background estimation suffers Cygnus Region Significance: 9.1  Post-trials probability: >7  Cygnus Region Crab Milagro FOV The Northern Sky above 100 MeV (EGRET) Crab Cygnus Region EGRET Data  =1.082

21. A Closer Look at the Galactic Plane  GP diffuse excess clearly visible from l=25 O to l=90 O.  Cygnus Region shows extended excess of diameter ~5 O -10 O.  F Cygnus ~= 2x F Crab

22. Color Map does not show error bars Map is oversampled which smooths the data Make Slices in Latitude for different Longitude cuts Consider Region l = 20 O -100 O -2<b<2 gives 7.5  Exclude the Cygnus Region: l=20 O -75 O -2<b<2 gives 5.8  Galactic longitude 20-75 excludes Cygnus region Galactic longitude 20-100 includes Cygnus region  =1.42 +/-.26 Galactic Latitudinal Distribution

23. Atomic Hydrogen radio contours EGRET Diffuse Model  Convolve Cygnus region excess with Milagro PSF(0.75 O ).  Region shows resolvable structure. Cygnus Region Morphology HEGRA detected TeV Source: TEV J2032_4130. PSF

24. EGRET Unidentified Sources in the Cygnus Region  > 100 MeV/cm 2 s  13EG J2016+3657 (34.7 ± 5.7) x 10 -8 2.09 23EG J2020+4017 (123. ± 6.7) x 10 -8 2.08 33EG J2021+3716 (59.1 ± 6.2) x 10 -8 1.86 43EG J2022+4317 (24.7 ± 5.2) x 10 -8 2.31 53EG J2027+3429 (25.9 ± 4.7) x 10 -8 2.28 63EG J2033+4118 (73.0 ± 6.7) x 10 -8 1.96 73EG J2035+4441 (29.2 ± 5.5) x 10 -8 2.08 1 2 3 4 5 6 7 3 rd EGRET Catalog sources shown with 95% position error circle. Flux of maximum point: 500mCrab (May be extended) psf

25. EGRET Data >1 GeV 1 2 3 4 5 6 7 Weight EGRET >1 GeV  -rays by EGRET’s energy dependent psf

26. Slice of EGRET Data  Cut on the Dec. band around Milagro’s bright spot  2 point sources or 1 extended source?  EGRET catalog sources were fit as point sources ONLY  How close together can GLAST resolve 2 sources of this signal strength? 1st point source Galactic Diffuse 2 nd point source Max error bar

27. Summary of the Statistical Issues  Event Reconstruction requires advanced pattern recognition and analysis techniques  Background estimation and uncertainty effects detection significance  Signal events should be weighted by probability of being signal and angular resolution  Effective area of the detector is continuously changing and may vary over the size of the point spread function  Chance probabilities are effected by oversampling and must be simulated by Monte Carlo