Rolling Search For a Cascade GRB Signal: 3 Year Results

Download Report

Transcript Rolling Search For a Cascade GRB Signal: 3 Year Results

Rolling Search For a Cascade
GRB Signal: 3 Year Results
Brennan Hughey
Baton Rouge Meeting
April 10th, 2006
1
Overview of Analysis
Motivation: Several AMANDA analyses have looked for neutrinos in
coincidence with GRBs. Many GRBs missed by gamma-ray
satellites, especially in era between BATSE and Swift. Additionally,
some transient sources may not have coincident photon emission
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
BATSE
Goal: to search for transients (specifically GRBs) in the period 20012003 without relying on satellite coincidence
2
Schematic of rolling search:
New time window starts with every event
3
Procedure
• Optimized for Prompt GRB neutrino emission: broken power
law spectrum with 105 and 107 GeV breaks
• Utilizes cascade rather than muon channel
• Searches in time bins of both 1 second and 100 seconds
after each surviving event
4
Data Reduction
•
•
High Energy Filter: cut on nhits and
fraction of hit modules with >2 hits
Flarechecking cuts
– First level cuts to restore exponential
distributions
– Extended flarechecking removes
“flariest” 1% from stable variables
•
•
Cut on number of direct hits
Six variable Support Vector Machine
Cut
– Combination of topological and
energy-related cuts
– Similar to neural network
- Many possible cut strengths
Black – surviving
Other colors –
Remove by
flarechecking
Black – real
Red – signal MC
Blue - Corsika
5
Optimization of Final Support
Vector Machine Cut
• Optimized for Discovery
• Observable is largest single cluster of events in any time window
during 3 year period – called Nlarge
• Assumes stochastic (Poissonian) background
• Short and long time windows optimized independently
6
Results
Results are very consistent with expectations based on background predictions
1 second search:
Nlarge = 2.
70.14% chance
100 second search:
Nlarge = 3.
75.25% chance
7
Doublets and Triplets:100 second
search
2001
328
2002
321
2003 | total
351 | 1000
2001
6
2002
8
2003 | total
6
| 20
8
Number of doublets: 1 second
search
2001
95
2002
102
2003 | total
114 | 311
No doublets or triplets in either time window found to be
in coincidence with a GRB from IPN lists.
Sample was unblinded again without flarechecking cuts,
no significant fluctuations found in this case either.
9
Limit/Sensitivity
Sensitivity determined using
- 90% CL Feldman Cousins likelihood ratio ordering method
- Primarily based on long time window, since long class has much greater fluence
- Poissonian Background with multiple periods per year
- Signal with relative per burst neutrino event rate distribution modeled by Guetta
et al. (astro-ph/0302524)
Pre-systematics
sensitivity/limit: 1.16 X 10-6 GeVcm-2s-1sr-1
(assumes 1:1:1 flavor ratio.
See, however, Kashti & Waxman astro-ph/0507599)
10
Systematic Uncertainty
A preliminary treatment of systematics has been done,
but included parameters will require adjustment
Included uncertainties are:
30% for signal efficiency (estimate from other analyses)
20% for uncertainty in distribution of events per burst
10% for uncertainty in signal retention by cuts
11
More about Systematic Uncertainty
Limit is modified according to standard AMANDA procedure from:
- Cousins and Highland
(Nuclear Instrumentation and Methods in Physics Research A320
(1992) 331-335)
- Conrad, Botner, Hallgren and Perez del Los Heros
- Hill
(hep-ex/0202013)
(physics/0302057)
L( n |  ,  ,  ' ) P( ' | ˆ ,   )d '

R
s
b
s
s
s
s
s
L( n | ˆ s , b , ˆs )
Numerator in Likelihood Ratio is integrated over PDF of possible signal
efficiencies (currently taken as flat distributions)
Adjusted limit/sensitivity: 1.26 X 10-6 GeVcm-2s-1sr-1 (from 1.16)
12
Choked Bursts
-
GRB jet may be unable to break through stellar envelope for very
extended or slowly rotating stars
(Meszaros and Waxman astro-ph/0103275)
-
-
These choked bursts still have neutrino signal produced during
propagation of jet within the star itself (equivalent to precursor
neutrino emission, but not a precursor to anything)
Lower energy spectrum than analysis is optimized for
Analysis not sensitive to these, how far off depends on the choked
burst event rate. It would be nice to have a detector that was a
couple of orders of magnitude bigger.
- 1X GRBs – factor of ~1200
- 10X GRBs – factor of ~(number)
- 100X GRBs – factor of ~68
13
Outlook/Conclusions
• Rolling search method provides a complementary procedure to
satellite-coincident GRB searches
– Sensitivity decreases due to lack of constraints, but competitive for
periods without dedicated GRB satellites
– Sampling from larger set of sources
• Can be extended to muon channel
– Erik Strahler has done preliminary work with Zeuthen point source cuts
– Spatial as well as temporal scan
– Cuts show less energy dependence, therefore less model dependence
• Cascade channel could be re-optimized for other transients, probably
not justified for AMANDA
• Next for me
– Write part of a paper
– Continue writing a thesis
14