Embed Size (px)
Citation preview
Fermi-LATmeasurement of
cosmic-ray electronspectrum with Pass 8
Carmelo SgroINFN–Pisa
on behalf of the Fermi LATcollaboration
October 20, 2014
The LAT as electron telescope
I The LAT is designed for E. M. showers
INaturally including electrons
(e+ + e�)I
No direct charge separation
I Triggering on (almost) every particlethat crosses the LAT
I Sending to ground all events depositingmore than 20 GeV in the CAL
I Electron identificationI
Dedicated event selection
I CRE spectrum and limits on anisotropyalready published
I The goal of Pass 8 electron analysis is:I
Update the results with a superior
event level analysis
IExtend the energy range Minimum Energy (GeV)
1 10 210 310 410
Dip
ole
Anis
otro
py
−310
−210
−110
LAT 95% ULVela
Carmelo Sgro (INFN–Pisa) October 20, 2014 2 / 14
The new event reconstruction package: Pass 8
x
z
I Pass 8 is a complete rework of the entire event level analysisI
See Philippe Bruel talk (in session 10B) for details
I E↵ectively a new instrument, with superior performance
I Calorimeter clustering tohandle “ghost” events
I Tree-basedtracking patternrecognition
I Improved shower profile fit forenergy reconstruction
Carmelo Sgro (INFN–Pisa) October 20, 2014 3 / 14
Introduction to the new analysis
Energy [GeV]210 310
sr]
2Ac
cept
ance
[m
0
1
2
3
4
5
PRELIMINARY
Trigger & FilterTrack found & energy
Minimal PSF quality
in the CAL0> 8 X
Alpha removal
I Basic quality cuts:I
At least a reconstructed track and 5 GeV of energy deposition in the CAL
IA loose selection on the PSF quality (using the same handle as in �-rayanalysis)
IAt least 8 radiation length in the CAL
I Alpha and heavier particles are removed using simple selections (next slide)I Field of view is limited to 60�
I Using all runs in survey mode available up to now: ⇠3.9 year of livetimeCarmelo Sgro (INFN–Pisa) October 20, 2014 4 / 14
Alpha particle removal
ACD tile energy (MeV)1 10 210 310
Cou
nts/
bin
0
500
1000
1500
2000
2500
3000
3500
4000Proton
Alpha
Ions
PRELIMINARY
I Alpha and ions are relatively easy to separate using, e.g., the pulse heightinformation in the ACD and the tracker
I Their hadronic interactions are comparatively hard to simulate
I We have a series of simple cuts to bring down the alpha/ion contaminationto a negligible level
IIn the following stages of the event selection we are essentially dealing with
only two classes of events (electrons and protons)
Carmelo Sgro (INFN–Pisa) October 20, 2014 5 / 14
Electron identification: event classification
I Take advantage of the experience with� rays
ISelect �/e±/etc. candidate at
di↵erent background level
IAssess the quality of the
reconstructed direction and energy
I Extensive use of multivariateclassification technique
IExploiting the TMVA package
(http://tmva.sourceforge.net)
IBoosted Decision Trees (BDT)
provide the best performance
I Several combination of trainingsetting tested
IBDT parameters (tree depth,
boosting, etc...)
IInput variables
I Decision Tree:I
Sequential application of cuts
splits the data into nodes, where
the final nodes classify an event
as Signal or Background
IWell known in “data mining”,
becoming popular in Physics
Carmelo Sgro (INFN–Pisa) October 20, 2014 6 / 14
Relevant quantities in the subsystems
2χShower profile fit 0 50 100 150 200
Rat
e [H
z]
0
0.001
0.002
0.003
0.004 = 0.44--1.00]θ[112.2--144.5 GeV, cos
±MC r/w eMC r/w pMC sum
PRELIMINARY I Shower transverse size
I Shower profile fit �2
I ...
TKR average time over threshold (MIPs)0 2 4 6 8 10
Rat
e [H
z]
0
0.0002
0.0004
0.0006
0.0008
0.001 = 0.44--1.00]θ[112.2--144.5 GeV, cos
±MC r/w eMC r/w pMC sum
PRELIMINARYI Time Over Threshold: energy
deposition in the TKR
I Extra hits around the main track
I ...
Carmelo Sgro (INFN–Pisa) October 20, 2014 7 / 14
Output of the classifier
Electron Classifier Output-3 -2.5 -2 -1.5 -1 -0.5 0
Rat
e [H
z]
0
5
10
15
20
25-610×
/ndof = 115.5/542χ (x 0.87)±MC r/w e
MC r/w p (x 1.58)MC sumFlight data
= 0.50--1.00]θ[563.3--648.0 GeV, cos
90%50%
20%
PRELIMINARY
I Example of one of the BDT weare using
I Others classifier configurationsunder evaluation, with sameprocedure
I Compared with flight dataI
Testing the data-MC agreement
IUsing MC with a realistic energy
spectrum
I Fitting only normalization
IEstimating signal directly from the fit
IEstimating the residual background
correction
I Event selection done with a cuton this quantity
IScanning several e�ciency
level
ITesting stability of the
spectrum
IResidual contamination
corrected with fit result
Carmelo Sgro (INFN–Pisa) October 20, 2014 8 / 14
Instrument response
Energy [GeV]210 310
sr]
2Ac
cept
ance
[m
0
0.5
1
1.5
2
2.5
3 Efficiency scanBest selection PRELIMINARY
Energy [GeV]210 310
Con
tam
inat
ion
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Efficiency scanBest selection
PRELIMINARY
I Testing the stability of thespectrum in this very wide range
ISpectrum variation likely relate
to data-MC disagreement
I Form 90% to 20%, (almost)energy-independent
IMaybe a too wide...
I Average acceptance (after cuts) forthis scan shown on the left
I “Best” cut can be evaluated usingthe MC-based ROC, as the pointin which the slope goes above adefined threshold
I Bottom plot shows thecorresponding residualcontamination
ICan be very large at high energy
Carmelo Sgro (INFN–Pisa) October 20, 2014 9 / 14
Preliminary e+ + e� spectrum
Energy [GeV]210 310
]-1
sr
-2 m
-1 s
-1 J
(E) [
GeV
× 3 E
0
50
100
150
200
250
300
350Fermi (2010)AMS-02 preliminaryH.E.S.S. (2010)Fermi (2012)Fermi Pass 8 PRELIMINARY
I Shaded region includesthe maximum variationof the spectrum in thee�ciency scan
I Central points fromthe “best” cut
I Result from templatefit within uncertainties
I E↵ect of absolute energy scale uncertainty not included in the plotI We have evidence that at least a significant part of the di↵erence with
our 2010 result is due to “ghost” signalI
This was not taken into account in the acceptance in our first analysis
ISubsequent studies (e.g. the control region in the positron analysis) suggest
an overestimation of acceptance by 10–15% at ⇠ 10 GeV
IPass 8 is designed to be insensitive to “ghost”
Carmelo Sgro (INFN–Pisa) October 20, 2014 10 / 14
Conclusions
I A new measurement of the e+ + e� spectrum with ⇠ 6 years of data andextending to 1.2 TeV is presented
I The new Pass 8 event reconstruction and selection performs very wellI
Better rejection power than the previous analysis
INo sign of dependence from “ghost” signal
I Pass 8 is still young also for electron analysesI
Results are preliminary
IWorking on improvements for energy extension, angle-resolved analyses,
reduced systematics
Carmelo Sgro (INFN–Pisa) October 20, 2014 11 / 14
