Cosmic Ray Electron Spectrum with the

Fermi-LAT

6th Fermi Symposium

Raffaella BoninoUniversity and INFN Torino

rbonino@to.infn.it

on behalf of the Fermi-LAT Collaboration

1

]2Energy [GeV210 310

]2 G

eV-1

sr

-1 s

-2 In

tens

ity [m

× 3 E

0

50

100

150

200

250

300

350Fermi (2010)

HESS (2008)

HESS (2009) - only stat. errors

AMS-02 (2014)

MAGIC (2011)

VERITAS (2015)

Fermi• First high-statistics measurement of inclusive

spectrum between 7GeV and 1TeV• Measurements compatible with single power-

law over the entire energy range ∝ E-3.08±0.05

CRE(e+e-) experiments

2

H.E.S.S. (& Cherenkov)• break in the spectrum at ~ 1 TeV

Energy [GeV]210 310

]2 G

eV-1

sr

-1 s

-2 In

tens

ity [m

× 3 E

6080

100120140160180200220240 Fermi (2010)

Fermi (2012)

H.E.S.S. (2008)

H.E.S.S. - low energy (2009)

AMS-02 (2014)

AMS-02• Range: 0.5 GeV to 1 TeV.• No structures observed.• From 30 GeV to 1 TeV described by a single power

law with γ=-3.170±0.008(stat+syst)±0.008(E scale)

LAT as electron detectorNot only � rays

I Detector is designed for E. M.showers

I Naturally including electrons(e+ + e�)

I Triggering on (almost) everyparticle that crosses the LAT

I On-board filtering to remove manycharged particles

I Keeps all events with more than20 GeV in the CAL

I Prescaled (⇥250) unbiasedsample of all trigger types

I Event reconstruction assumes aE.M. shower

I Works fine for electrons

I Electron identificationI Dedicated event selection

I No charge separationCarmelo Sgro (INFN–Pisa) CRIS 2015, September XX 4 / 18

Detector is designed for E. M. showers:➡ naturally including electrons

➡ event reconstruction works also for electrons

Electron identification requires dedicated event selection

� rays detection principle

� ray

e+ e�

Tracker/converter

Calorimeter

Anti-coincidence shield

Tracking plane

Conversion plane

I Standard technique for high-energy �-ray astrophysicsI Dominant interaction mechanism for E >⇠ 20MeVI Used by past experiment like COS-B and EGRET

I �-ray converts in the middle of Tracker/Converter ! �-ray direction

I Calorimeter absorbs part of the e.m. shower ! �-ray energy

I No signal in the Anti-coincidence shield ! charged particle discrimination

Carmelo Sgro (INFN–Pisa) CRIS 2015, September XX 3 / 18

3 3

Analysis features

Objective: discriminate signal (e+/e-) from background (mainly p) and compute the CRE energy spectrum

Data set: ~ 7 years

Energy range: 50 GeV - 2.5 TeV

Event reconstruction: Pass 8

Event selection: PRECUTSClassification Tree: TMVA with the Boosted Decision Tree method- trained on MC data samples (signal=MC electron, bkg=MC proton)

- in 0.25 wide log energy bins ➞ training optimized for the whole energy range

CRE spectrum 4

Analysis steps

selection of variables

ClassificationTree

probability variablelog10(1-prob)

selection cut template fitting

CRESpectrum

bkg subtraction

5

The LAT uses shower topology information to separate the electron signal from the hadronic background

Shower transverse size10 20 30 40 50 60 70 80 90 100

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

EM shower Hadronic shower

1 TeV electron candidate 1 TeV proton candidate

Selection of variables

Example:

6

Separation efficiency improved after correcting some observables for geometry dependence

Selection of variables

7

Variables chosen according to: good MC-data agreement and high separation efficiency

MC-data agreement optimized after correcting for energy and incidence angle dependences

Electron Classifier Output-5 -4 -3 -2 -1 0

Rate

[Hz]

02468

1012141618

-610×

/ndof = 186.8/1862χ (x 1.17)±MC r/w e

MC r/w p (x 1.47)MC sumFlight data

= 0.50--1.00]θ[281.8--316.2 GeV, cos

Probability variable

PRELIMINARY

8

Fit the data with the MC electron and proton templatesignal rate = (rate from flight data) x (fraction of electrons from MC) use that rate for the spectrum directly!

MC proton template (i.e. parameterization of AMS-01) is renormalized and fit to the data in each energy bin

Electron Classifier Output-5 -4 -3 -2 -1 0

Rate

[Hz]

02468

1012141618

-610×

/ndof = 186.8/1862χ (x 1.17)±MC r/w e

MC r/w p (x 1.47)MC sumFlight data

= 0.50--1.00]θ[281.8--316.2 GeV, cos

a. Template fitting

PRELIMINARY

9

Cutting on the CT probability variable:“optimal cut” = point on the performance curve (ROC) in which the slope becomes greater than a defined threshold

fit the cut values as a function of energy → analytical function of Prob(E)

Bkg contamination estimated after applying “correction” from template fittingsr]2allHEE Acceptance [m

1.4 1.6 1.8 2 2.2 2.4 2.6M

c Bk

g C

onta

min

atio

n0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

562.34<E<630.96 GeVElectron Classifier Output

-5 -4 -3 -2 -1 0

Rate

[Hz]

02468

1012141618

-610×

/ndof = 186.8/1862χ (x 1.17)±MC r/w e

MC r/w p (x 1.47)MC sumFlight data

= 0.50--1.00]θ[281.8--316.2 GeV, cos

b. Bkg subtraction

PRELIMINARY

10

--- ROC curve★ optimal cut

Resulting CRE spectra

11Energy [GeV]

210 310

]2 G

eV-1

sr

-1 s

-2 In

tens

ity [m

× 3 E

6080

100120140160180200220240 Background subtraction spectrum PRELIMINARY

Template fitting spectrum PRELIMINARY

Energy [GeV]210 310

]2 G

eV-1

sr

-1 s

-2 In

tens

ity [m

× 3 E

6080

100120140160180200220240 Background subtraction spectrum PRELIMINARY

Energy scale syst.

Energy scale systematics

12

The uncertainty on the absolute energy scale is the largest source of systematicsPass 8 in-flight measurement of the absolute energy scale via geomagnetic cutoff study → 3.7% offset around 10 GeVWe have rescaled the whole spectrum by 3.7% and we have estimated the error on this scaling factor to be 2% at 10 GeV and increasing up to 5% at ~ 1 TeV

Energy [GeV]210 310

Rat

io

0.7

0.8

0.9

1

1.1

1.2

1.3Event Selection syst. PRELIMINARY

Energy [GeV]210 310

Rat

io

0.7

0.8

0.9

1

1.1

1.2

1.3Correction factor syst. PRELIMINARY

Energy [GeV]210 310

Rat

io

0.7

0.8

0.9

1

1.1

1.2

1.3Bkg simulation reliability syst. PRELIMINARY

Other systematicsA) Event selection:

main uncertainty at this level is due to the estimation of the effective area → we take it into account by varying the signal efficiency between 30% and 90%

B) Correction factor:the band is calculated by moving the correction magnitude by one sigma

C) Bkg simulation reliability:the uncertainties related to the MC simulation of hadronic interactions could produce a 30% uncertainty on the residual contamination

13

B

C

PRELIMINARY

A

PRELIMINARY

PRELIMINARY

Energy [GeV]210 310

]2 G

eV-1

sr

-1 s

-2 In

tens

ity [m

× 3 E

6080

100120140160180200220240

Fermi (2010)

HESS (2008)

HESS (2009) - only stat errors

AMS-02 (2014)

Fermi Pass 8 PRELIMINARY

CRE inclusive spectrum

14

Energy [GeV]210 310

]2 G

eV-1

sr

-1 s

-2 In

tens

ity [m

× 3 E

6080

100120140160180200220240

Fermi (2010)

HESS (2008)

HESS (2009) - only stat errors

AMS-02 (2014)

Fermi Pass 8 PRELIMINARY

CRE inclusive spectrum

15

Central points from the “optimal cut”, the shaded band has been obtained by summing in quadrature all the studied systematics (except for the energy scale)

Compatible with AMS up to ~ 100 GeV but different spectral indexDisagreement wrt published spectrum (Fermi 2010):

likely due to “ghost” signal not taken into account in the acceptance in our first analysis

Energy [GeV]210 310

sr]

2Ac

cept

ance

[m

0

0.5

1

1.5

2

2.5

3

3.5

Event Selection (scan on signal efficiency)Fermi Pass 8 PRELIMINARY

Instrument responseAcceptance resulting after applying the “optimal cut” +

acceptances resulting by varying the signal efficiency between 30% and 90% also shown

Contamination below 20% when applying the “optimal cut” +

contamination resulting by varying the signal efficiency between 30% and 90%contamination resulting by taking into account the bkg simulation reliability

16Energy [GeV]

210 310

Res

idua

l bkg

con

tam

inat

ion

0

0.1

0.2

0.3

0.4

0.5

0.6

Bkg simulation reliabilityEvent Selection (scan on signal efficiency)

Fermi Pass 8 PRELIMINARY

Conclusions

We performed a new measurement of CRE spectrum with Pass 8

Improvements in the new analysis:almost 7 times the PRD data setnew event reconstruction & selection (Pass 8)new multi-variate analysis toolnew selection of variables → variables are now “calibrated”new CTs trained in energy bins ➞ training optimized for the whole energy range

➡ new CRE spectrum and associated systematics

See also the posters:More technical details on the event selection and systematics (M.Negro et al - Diff.6)

Analysis extended down to 7 GeV by including the DGN filter effect (A. Manfreda et al - In/An.10) → in the overlapping energy region it’s in agreement with this one

Cross-check the energy scale at TeV energies by measuring the Earth Limb spectrum (F. Spada et al - In/An.7)Study CRE anisotropies to validate possible astrophys. interpretations (N. Mazziotta et al - Diff.5)Viable interpretations of our CRE spectrum (F. Donato et al - Diff.3) 17

BACKUP SLIDES

18

Systematics: corr. factor

19

Acd2PLCTkr1TileActDistEnergy1 10 210 310

-410

-310

-210

-110

Flight dataMC sum

±MC eαMC

MC Z > 2MC p

= 0.30--1.00]θ[31.6--3162.3 GeV, cos

Tkr1ToTAve0 2 4 6 8 10 12 14

-510

-410

-310

-210

-110

1 Flight dataMC sum

±MC eαMC

MC Z > 2MC p

= 0.30--1.00]θ[31.6--3162.3 GeV, cos

PRECUTS = TRIGGER FILTER + QUALITY CUT + ALPHA CUTTRIGGER FILTER: the event triggers the LAT and passes the on-board gamma filter '(GltGemSummary&0x20)==0 && (GltGemSummary&0x40)==0 && FswGamState == 0'

QUALITY CUT: the event has at least a reconstructed track, a minimal PSF quality and the path length in the Cal is larger than the Cal on-axis thickness 'EvtCalCsIRLn>8 && Cal1RawEnergySum>5000 && TkrNumTracks>0 && WP8CTPSFTail>0.05'

ALPHA CUT: MC doesn’t reproduce accurately interactions of α and heavy ions in the LAT → cut removing the majority of α and heavies

Pre-cuts

20