B-Tagging and ttH, H bb Analysis on Fully Simulated Events in the - - PowerPoint PPT Presentation

September 27, 2005 FAKT 2005, Vienna Slide 1

B-Tagging and ttH, H → bb Analysis

• n Fully Simulated Events

in the ATLAS Experiment

A.H. Wildauer

Universität Innsbruck CERN ATLAS Computing Group

September 27, 2005 FAKT 2005, Vienna Slide 2

Overview

• Introduction
• The ttH, H→bb Channel

topology, cross section, backgrounds

• B-Tagging Algorithms

impact parameter, weight, performance on ttH

• Analysis of ttH

reconstruction, event selection, bkg rejection comparison with fast simulation

September 27, 2005 FAKT 2005, Vienna Slide 3

Introduction

• I am a PhD student in the

Austrian Doctoral Student Program at CERN. Main Working Areas

• start-up: work on e/gamma trigger efficiencies for the High Level Trigger TDR
• work on Atlas reconstruction software (Athena) with focus on Inner Detector
• development of vertex software and its Event Data Model
• development and performance of b-tagging software and its EDM
• analysis of the ttH, H→bb channel on AOD level

September 27, 2005 FAKT 2005, Vienna Slide 4

ttH→jjb lνb bb

• promising discovery channel for a light Standard Model Higgs Boson

complex final state

W W

• 6 jets where 4 are b-jets (εb

4!!)

• 1 W has to decay leptonicaly (trigger!)
• 1 neutrino: missing energy!

→ efficient b-tagging very important for signal reconstruction → channel has to be fully reconstructed to reduce combinatorics

September 27, 2005 FAKT 2005, Vienna Slide 5

Signal and Background

• fully simulated events with “initial” detector layout (2 pixel layers)

Signal:

• ttH(120) → lνb jjb bb (0.52 pb, H(120)→bb 70%)
• mH chosen to be 120 Gev/c2

60k events from private production on the Grid Background:

• ttjj background (474 pb) – 250k events
• ttbb (QCD) (gg: 8.1 pb, qq: 0.5 pb) – 50k events
• ttbb (EW) none produced

→ large background! good rejection needed: efficient b-tagging very important to reduce background

September 27, 2005 FAKT 2005, Vienna Slide 6

B-Tagging

• b-tagging: identify jets which come from a b-quark
• How? By using the properties of B-hadrons:
• longer lifetime
• reconstructable 2nd vertex
• semileptonic decay modes

“Dependencies”:

• Tracking
• Vertex reconstruction
• Jet finding

B B a0 < 0 a0 > 0 Secondary Vertex Primary Vertex

Jet-Axis Lepton

September 27, 2005 FAKT 2005, Vienna Slide 7

Impact Parameter Tagging

• most common way to tag b-jets: the signed IP significance distribution
• better than IP alone: give higher weight to well measured tracks!

) σ(a a ) S(a

0 =

z Signed Impact Significance rφ Signed Impact Significance

• knowledge of primary vertex important to calculate IP!

September 27, 2005 FAKT 2005, Vienna Slide 8

Likelihood/Weight

• Significance distributions are used as input pdfs to calculate a jet weight
• or a normalized b-tag likelihood.
• typical likelihood/weight plot for combined tagging in z and rphi:

∑

        =

tracks

Bkg(S) Sig(S) ln W

September 27, 2005 FAKT 2005, Vienna Slide 9

B-Tagging Performance

• B-Tagging performance is given in 2 connected quantities:
• light jet rejection Ru at a given b-jet selection efficiency εb:

u

ε 1 =

u

R

81 62 13 60% 34 5 1D 180 24 2D 220 31 CB 50% 70% Ru\εb

• numbers are without 2nd vertex tagger
• !performance depends heavily on truth matching and jet cleaning!
• more important: performance in an actual analysis (e.g.

)

B S

September 27, 2005 FAKT 2005, Vienna Slide 10

ttH Event Reconstruction

• 2 jets out of 4 b-jets out of 6 reco jets need to be assigned to the Higgs …

→ full reconstruction necessary to reconstruct Higgs Boson

b b b b ν ℓ j j H W W t

Event Selection

→ 1 µ(e) with pt > 20(25) GeV, |η|<2.5 → 6 jets with pt > 20 GeV, |η|<5. → 4 jets tagged as b-jets (cut defined at εb = 60%) → 2 reconstructed tops with |∆mtop|<20 GeV → this leaves 2 b-jets for the reco of the Higgs

t

September 27, 2005 FAKT 2005, Vienna Slide 11

Cut Flow Signal

• comparison of my analysis (AOD) with fully simulated events to 2

analyses based on fast simulation and older detector layout (3 pixel layers)

100 % 100 % 100 % All Events 0.8 (35) 2.3 (60) 3.8 (8) 46.2 TDR 1.5 3.7 3.8

• J. Cammin

(improved analysis)

4.15 (8) 4 bjets 2.0 (48) 2 tops reco 50.8 1l 6j 0.7 (35) Higgs reco AOD

ttH(120)

• numbers in () are relative to previous cut
• problem with top reconstruction? (might be at W→lν reco)

September 27, 2005 FAKT 2005, Vienna Slide 12

Cut Flow ttjj Background

• cut flow in the background with largest cross section: ttjj

100 % 100 % 100 % All Events 0.0001 (2.1) 0.0047 (47) 0.01 (0.1) 15.4 TDR 0.0013 (13) 0.01 (92.3) 0.01

• J. Cammin

(improved analysis)

0.035 (0.2) 4 bjets 0.013 (37) 2 tops reco 17.7 1l 6j 0.0007 (5.4) Higgs reco AOD

ttjj

• selection efficiency and background rejection comparable
• ttbb (QCD) background also comparable with earlier analyses

September 27, 2005 FAKT 2005, Vienna Slide 13

Reconstructed Masses in Signal

m =173.3 GeV σ = 9.3 GeV m =172.2 GeV σ = 10.1 GeV

GeV GeV GeV

t→jjb: TDR: 174 ± 11.7 GeV, J.Cammin: 174.7 ± 7.7 GeV t→lνb: TDR: 174 ± 8.8 GeV, J.Cammin: 174.6 ± 8.6 GeV

• tail in the Higgs mass spectrum due to mismatched b quarks

September 27, 2005 FAKT 2005, Vienna Slide 14

Number of expected Events at 30 fb-1

3166 3166 3166 All Events 25 70 117 1462 TDR 47 117 120

• J. Cammin

(improved analysis)

128 4 bjets 61 2 tops reco 1609 1l 6j 22 Higgs reco AOD

ttH(120)

• 30 fb-1 is the anticipated

integrated luminosity after 3 years of low lumi run

• tt is always forced to decay to

lνb ljj with BR ~ 29%

4.1M 4.1M 4.1M All Events 5 192 410 631k TDR 53 377 410

• J. Cammin

(improved analysis)

1435 4 bjets 533 2 tops reco 730k 1l 6j 29 Higgs reco AOD

ttjj

simulated events after cuts Signal: ~300 events left ttjj background: ~10 events left → no detailed analysis possible

September 27, 2005 FAKT 2005, Vienna Slide 15

Conclusion and Outlook

• first look at ttH channel with fully simulated events and initial

detector layout

• “realistic” b-tagging performance looks OK on ttH channel

(no SV tagger in use for this analysis so far)

• cut flow on sig and bkg in agreement with earlier studies
• small discrepancies in the W→lν reconstruction under study
• lack of simulated events

→ a lot more are needed (factor 10)

• might need to use fast simulation for more background …