Transcript Detecting g-ray Sources - Pennsylvania State University

Detecting g-ray Sources
Brenda Dingus
[email protected]
23 January 2006
Outline:
I. Detection Techniques
II. Each g-ray is an Image
III. Source Detection
A.
B.
C.
Point Sources
Extended Sources
Variable Sources
g-rays Probe Nature’s Particle Accelerators
Black Hole producing
relativistic jet of particles
HST Image of M87 (1994)
Binary Neutron Star
Coalescing
Artist
Conception of
Short GRBs
Spinning Neutron Star
powering a relativistic
wind
Chandra x-ray image
TeV image of Vela Jr.
Supernova Remnant
HESS TeV
+ x-ray
Massive Star Collapsing
into a Black Hole
SuperComputer Calculation
Different Types of Detectors
for Gamma-Ray Astrophysics
Low Energy Threshold
EGRET/GLAST
Large Aperture/High Duty Cycle
Milagro, Tibet, ARGO, HAWC
Low Duty Cycle/Small Aperture
Large Effective Area
Excellent Background Rejection
Large Duty Cycle/Large Aperture
Space-based (Small Area)
“Background Free”
Large Duty Cycle/Large Aperture
Moderate Area
Good Background Rejection
Known Source Spectra
Known Source Lightcurves
Survey of Galactic Plane
Sky Survey > 100 MeV
Transient Sources
Extended Sources
Sky Survey >~ 1 TeV
Transient Sources
Extended Sources
High Sensitivity
HESS, MAGIC, CANGAROO, VERITAS
High Energy g-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
Satellites (30 MeV to 300 GeV g-rays)



g-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
g-ray direction is determined by
energy-weighted average of the
electron and positron tracks
g
Pair-Conversion Telescope
anticoincidence
shield

conversion foil
particle tracking
detectors
e+
e–
calorimeter
(energy measurement)

Angular Resolution is dominated by
multiple scattering (a 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
Energy Dependent Localization
EGRET’s
Galactic Center
Source
Galactic Latitude (deg)
 The number of g-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
Galactic Longitude (deg)
Diamonds show g-rays > 5 GeV.
95 % confidence intervals are
Black Circle for Previous Analysis
and Blue Area is New Analysis.
Point Source Survey
 Source confusion will be
a problem
Prediction for GLAST Detections
Of Active Galactic Nuclei
– ~ 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 g-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-2s-1
Time Variable Sources
 Small # of g-rays
limits minimum
variability time
scale
 At least 5 g-rays
are required to
detect a source
 Bayesian block
statistical
technique is
needed to
distinguish the
lightcurve
100 sec
1 orbit
- GRB940217 (100sec)
- PKS 1622-287 flare
- 3C279 flare
- Vela Pulsar
- Crab Pulsar
- 3EG 2020+40 (SNR g Cygni?)
1 day
- 3EG 1835+59
- 3C279 lowest 5 detection
- 3EG 1911-2000 (AGN)
- Mrk 421
- Weakest 5 EGRET source
Galactic Diffuse Emission
 g-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 g-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)
Galactic Diffuse Model
& EGRET Data
(Hunter et al. 1998)
-180
-140
-100
-60
-20
20
60
100
140
180
Water Cherenkov
Extensive Air Shower Detectors
• 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
e
m
g
8 meters
50 meters
80 meters
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 D even-odd = 1.0o
implies Gaussian  of 0.4o
for proton reconstruction
Red Monte Carlo
Black Data
D even-odd in degrees
Event Images in Milagro
Protons
Size of red dots indicate # of photoelectrons detected.
Proton MC
Gamma MC
Gammas
Data
Cut at C>2.5 to Retain 50% g and 9% protons.
Signal
Q  0.5
Background
0.09
 1.7

Efficiencygamma
Efficiencybackground
Q
MARS1 (Multivariate Adaptive Regression Splines)



Predicts the values of an outcome variable given a set of independent
predictor variables
Calculates probability of g vs background for all combinations of parameters
MARS Value is ln[P(signal)/P(background)]
– More positive means more g-like
Differential Distribution
1J.
Integral Distribution
Friedman, “Multivariate Adaptive Regression Splines”, Annals of Statistics 19 (1991).
Combining Data with Different Cuts: Weighted Analysis
Milagro’s Crab Signal
Std Cuts: NFIT>=20,C>2.5
Excess = 5410, Off = 1218288, S:B = 1:225
ehadron background =~ 0.1
Weight events by
Expected S:B
Hard Cuts: NFIT>=200,C>6.0
Excess = 60, Off = 140, S:B = 1:2.3
ehadron background =~ 1x10-5
Point Source Search - Weighted Analysis
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
= 2.1o
Cygnus Region
Vicinity of the Crab
=1.03
Mrk421
Crab
Milagro Background Estimation
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
Extended Source Search
Vary Bin Size from
~5O source)
2.1O
to
5.9O
(Optimal for
The Northern Sky above 100 MeV (EGRET)
Cygnus Region
=1.082
Crab
As bin size increases to > 6O background
estimation suffers
Milagro
FOV
Cygnus Region Significance: 9.1
Post-trials probability: >7
EGRET Data
Cygnus Region
Crab
A Closer Look at the Galactic Plane
 GP diffuse excess
clearly visible from
l=25O to l=90O.
 Cygnus Region shows
extended excess of
diameter ~5O-10O.
 FCygnus ~= 2x FCrab
Galactic Latitudinal Distribution
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 = 20O-100O
-2<b<2 gives 7.5
Exclude the Cygnus Region: l=20O-75O
-2<b<2 gives 5.8
Galactic longitude 20-75 excludes Cygnus region
=1.42 +/- .26
Galactic longitude 20-100 includes Cygnus region
Cygnus Region Morphology
Convolve Cygnus region excess with Milagro PSF(0.75O).
Region shows resolvable structure.
HEGRA detected TeV
Source: TEV J2032_4130.
EGRET
Model
Atomic Diffuse
Hydrogen
radio contours
PSF
EGRET Unidentified Sources in the Cygnus Region
psf
3rd EGRET Catalog sources
shown with 95% position
error circle.
7
1
2
3
4
5
6
7
3EG J2016+3657
3EG J2020+4017
3EG J2021+3716
3EG J2022+4317
3EG J2027+3429
3EG J2033+4118
3EG J2035+4441
F > 100 MeV/cm2s
(34.7 ± 5.7) x 10-8
(123. ± 6.7) x 10-8
(59.1 ± 6.2) x 10-8
(24.7 ± 5.2) x 10-8
(25.9 ± 4.7) x 10-8
(73.0 ± 6.7) x 10-8
(29.2 ± 5.5) x 10-8
g
2.09
2.08
1.86
2.31
2.28
1.96
2.08
Flux of maximum point: 500mCrab
(May be extended)
4
6
2
3
5
1
EGRET Data >1 GeV
Weight EGRET
>1 GeV g-rays by
EGRET’s energy
dependent psf
7
4
6
2
3
5
1
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?
2nd point
source
1st point
source
Galactic Diffuse
Max
error
bar
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