75
Top Mass Reconstruction in Semi Electronic t¯ t Events Maryam Zeinali Overview Golden Channel Event Selection Software, Sample and Selection Cuts 25 GeV: Too Tight? Jet-Parton Matching Algorithm Performance Signal and Background Distributions Monitoring Variables Training Variables Testing the Method How MVA Methods work? Top Mass Reconstruction in Semi Electronic t ¯ t Events Maryam Zeinali IPM Weekly Meeting February 2 2010 1 / 32

Top Mass Reconstruction in Semi Electronic t Eventsparticles.ipm.ir/special seminar/materialDisplay-zeinali13bahman88... · Top Mass Reconstruction in Semi Electronic t t Events Maryam

  • Upload
    hahanh

  • View
    215

  • Download
    0

Embed Size (px)

Citation preview

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Top Mass Reconstructionin

Semi Electronic tt̄ Events

Maryam Zeinali

IPM Weekly Meeting

February 2 2010

1 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Outline

OverviewGolden Channel

Event SelectionSoftware, Sample and Selection Cuts25 GeV: Too Tight?

Jet-Parton MatchingAlgorithmPerformance

Signal and Background DistributionsMonitoring VariablesTraining VariablesTesting the MethodHow MVA Methods work?

2 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Outline

OverviewGolden Channel

Event SelectionSoftware, Sample and Selection Cuts25 GeV: Too Tight?

Jet-Parton MatchingAlgorithmPerformance

Signal and Background DistributionsMonitoring VariablesTraining VariablesTesting the MethodHow MVA Methods work?

2 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Outline

OverviewGolden Channel

Event SelectionSoftware, Sample and Selection Cuts25 GeV: Too Tight?

Jet-Parton MatchingAlgorithmPerformance

Signal and Background DistributionsMonitoring VariablesTraining VariablesTesting the MethodHow MVA Methods work?

2 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Outline

OverviewGolden Channel

Event SelectionSoftware, Sample and Selection Cuts25 GeV: Too Tight?

Jet-Parton MatchingAlgorithmPerformance

Signal and Background DistributionsMonitoring VariablesTraining VariablesTesting the MethodHow MVA Methods work?

2 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal?

To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How?

Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio?

19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature?

An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?

? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,

? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,

? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,

. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!

? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Overview

I Goal? To measure the top quark mass

I How? Using tt̄ events, semi electronic decay channel

I Branching ratio? 19 ∗

69 ∗ 2 = 4

27 ≈ 15%

I Event signature? An energetic electron, four jets thattwo of them are b jets and missing transverse energy

I Backgrounds to suppress?? W+jets,? QCD,? single top,. . . but non of them are really matters!? The main background to top mass measurementcomes from the signal itself!

3 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Golden Channel: Lepton + Jets Final State

I Leptonic side is used for triggering purposesI Four undistinguishable reconstructed jets at final state:

12 possible ways to reconstruct hadronic top 4-vector!4 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection Cuts

I CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is used

I TTbar PYTHIA is skimmed at GenLevel to select onlysemi electronic events

I At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic events

I At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829

Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925

>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587

== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?

5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Software, Sample and Selection CutsI CMSSW 3 1 4 is usedI TTbar PYTHIA is skimmed at GenLevel to select only

semi electronic eventsI At least 4 jets are required to pass ET > 25GeV and|η| < 2.4

I Exactly one electron is asked to have these criteria:? pt > 20GeV and |η| < 2.4 excluding the gap? identified as RobustTight? reliso which is defined as the sum of all isolationvalues in a cone of 0.3 over electron ET should be lessthan 0.1

I Cut flow table:

Initial tt̄ events 485829Initial semi ele events 70925>= 4 Jet 46587== 1 Electron 21822

⇒ 66% pass cuts on jets. Cuts are abnormal?5 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Distribution of Partons’pt in Semi ElectronicEvents

Leading jet pt Next leading jet pt

Next to next leading jet pt Next to next to next leading jet pt

6 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Algorithm

1. All events survived the selection cuts are fed to thejet-parton matching algorithm

2. Among four hard scattered partons, take the one withthe hardest pt .

3. Make a loop over all jets in the event. No cut is put onjets in the loop.

4. The jet with the minimum distance to the parton isflagged.

5. If ∆r < 0.3, jet is called as matched and removed fromthe list of jets.

6. Take the next leading parton and go to stp 2.

7 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background

? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,

? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.

? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.

? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Jet-Parton Matching Performance

I How the method works?

I Let’s define right and wrong jet combinations: Signaland Background? Take four leading jets in the event,? If all of them are matched to the partons, only oneamong 12 possible combinations contribute to signal.The other 11 possible ways take part in backgrounds.? If not all four leading jets matched to the partons,then all 12 combinations are considered as backgrounds.? For illustration: consider the number of initial eventsto be N and number of events with exactly 4 leadingjets have matched to the partons is S . Then thenumber of background contributions, B, can becalculated: B = S ∗ 11 + (N − S) ∗ 12.

I Now to see how good matching algorithm is, we cancompare Signal combination to the Gen Info.

8 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Mass of hadronic W− right jet combination

− Gen Info

9 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Mass of hadronic W: LOG scale− right jet combination

− Gen Info

10 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Pt of hadronic W− right jet combination

− Gen Info

11 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Mass of hadronic top− right jet combination

− Gen Info

12 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Mass of hadronic top: LOG scale− right jet combination

− Gen Info

13 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Pt of hadronic top− right jet combination

− Gen Info

14 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

∆φ(1st lightjet, 2nd lightjet)− right jet combination

− Gen Info

15 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

∑i p

lightjetit

− right jet combination

− Gen Info

16 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

∆φ(leptonicB , electron)− right jet combination

− Gen Info

17 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

∆φ(hadronicB , hadronicW )− right jet combination

− Gen Info

18 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

∆φ(leptonicB , hadronicW )− right jet combination

− Gen Info

19 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Monitoring variables that can discriminatebetween right and wrong jet combinations

20 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Monitoring variables that can discriminatebetween right and wrong jet combinations

21 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Monitoring variables that can discriminatebetween right and wrong jet combinations

22 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Correlation of variables for right jet combinations

23 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Correlation of variables for wrong jetcombinations

24 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Training Variables with TMVA: LikelihoodMethod

25 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Testing of the MVA Method: Likelihood

26 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Testing of the MVA Method: Likelihood

26 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Evaluation of the Results: Likelihood Method

27 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Evaluation of the Results: PEDRS Method

28 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Background Rejection Vs Signal Efficiency:Comparing Likelihood and PEDRS Methods

29 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Likelihood Method

Projective likelihood estimator (PDE approach)

I Builds a model out of probability density functions

I Reproduces the input variables for signal and background

I For a given event, the likelihood for being of signal type is

obtained by multiplying the signal probability densities of all

input variables, which are assumed to be independent, and

normalising this by the sum of the signal and background

likelihoods

I Because correlations among the variables are ignored, this

PDE approach is also called naive Bayes estimator

Classifier

I The likelihood ratio for event i , yL(i), is defined by:

yL(i) = LS (i)LS (i)+LB (i) where LS(B)(i) =

∏nvark=1 pS(B),k(xk(i))

? nvar is the number of input variables used

30 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Likelihood Method

Projective likelihood estimator (PDE approach)

I Builds a model out of probability density functions

I Reproduces the input variables for signal and background

I For a given event, the likelihood for being of signal type is

obtained by multiplying the signal probability densities of all

input variables, which are assumed to be independent, and

normalising this by the sum of the signal and background

likelihoods

I Because correlations among the variables are ignored, this

PDE approach is also called naive Bayes estimator

Classifier

I The likelihood ratio for event i , yL(i), is defined by:

yL(i) = LS (i)LS (i)+LB (i) where LS(B)(i) =

∏nvark=1 pS(B),k(xk(i))

? nvar is the number of input variables used

30 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

PDERS MethodMultidimensional likelihood estimator (PDE range-search approach)

I A generalization of the projective likelihood classifer to nvardimensions

I If the multidimensional PDF for signal and background were

known, this classifer would exploit the full information

contained in the input variables, and would hence be optimalI The PDE for a given test event (discriminant) is obtained by

counting the (normalised) number of training events that

occur in the ”vicinity” of the test eventI The classifcation of the test event may then be conducted on

the basis of the majority of the nearest training events

Classifier

I PDERS method takes the ratio yPDERS(i ,V ) =nS (i,V )

NSnS (i,V )

NS+

nB (i,V )

NB

as the estimator by defining a volume V around the test

event, i , and by counting the number of signal, nS(i ,V ) and

background events nB(i ,V ) obtained from the training

sample in that volume while NS(B) is the total number of

signal (background) events in the training sample

31 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

PDERS MethodMultidimensional likelihood estimator (PDE range-search approach)

I A generalization of the projective likelihood classifer to nvardimensions

I If the multidimensional PDF for signal and background were

known, this classifer would exploit the full information

contained in the input variables, and would hence be optimalI The PDE for a given test event (discriminant) is obtained by

counting the (normalised) number of training events that

occur in the ”vicinity” of the test eventI The classifcation of the test event may then be conducted on

the basis of the majority of the nearest training events

Classifier

I PDERS method takes the ratio yPDERS(i ,V ) =nS (i,V )

NSnS (i,V )

NS+

nB (i,V )

NB

as the estimator by defining a volume V around the test

event, i , and by counting the number of signal, nS(i ,V ) and

background events nB(i ,V ) obtained from the training

sample in that volume while NS(B) is the total number of

signal (background) events in the training sample

31 / 32

Top MassReconstruction

inSemi Electronic tt̄

Events

Maryam Zeinali

Overview

Golden Channel

Event Selection

Software, Sample andSelection Cuts

25 GeV: Too Tight?

Jet-PartonMatching

Algorithm

Performance

Signal andBackgroundDistributions

Monitoring Variables

Training Variables

Testing the Method

How MVA Methodswork?

Outlook and To Do List

I Although by applying event selection cuts thebackgrounds to the signal can be suppressed, top massreconstruction is suffering from a huge contribution ofcombinatorial backgrounds.

I The results reported are obtained by running TMVApackage based on ROOT. There is anotherimplimintaion of MVA methods based on CMSSWenvironment.

I The final analysis has to be moved to CMSSW framework sooner or later. Then it needs to synchronize thecode accordingly.

32 / 32