Transcript ppt

Assessment and Proposal
W. B. Atwood
UCSC, April 6, 2005
GLAST Phase Space
Science Requirements:
Instrumental:
Field-of-View (FoV) > 2.5 str.
(or -1. < cos(q) < -.2 )
Energy Range
20 MeV
300 GeV
Conversion location: 1-of-16 Tracker Layers (Thin & Thick)
Shower Location: Cal Module Edges
Other: High energy, side entering events...
GLAST PHASE SPACE IS VERY LARGE
Consequence: unlikely that one energy correction algorithm will
cover all aspects optimally
Several techniques being pursued: Parametric, Last Layer, Profile, ...
Current Situation
Source: 1 GeV g's at near vertical incident at the ~ center of a CAL module
Parametric
Last Layer
Similar resolutions – how to select which to use?
Proposal: Use Classification Trees
LL:Para – 50:50
GEnergy = CT-predicted-best-of Parametric , Last-Layer
The CT selected energy is significantly better then either
Parametric or Last-Layer in isolation.
The CT mechanism allows us to mutually benefit in that region of
phase space where one Algorithm works better then the others
Can easily be extend to an arbitrary number of algorithms
Phase Space Breakdown
Overall All Energy resolution
at 1 GeV
4 Bins in CalX0
4 Bins in cos(q)
Implementation
The mechanism of Classification Trees can allow is to take the
"best" solution across GLAST Phase-Space
Structural Implications
- CalRecon needs to calculate (or attempt to calculate) all
solutions for each event (This is needed so that all the
algorithms can be tested and monitored).
- Top (or near top) Level Algorithm then uses a CT to select
"optimal solution" which is presents to the user