FTK for the Trigger Phillip Urquijo Soshi Tsuno Atlas Trigger - - PowerPoint PPT Presentation

FTK for the τ Trigger

Phillip Urquijo Soshi Tsuno Atlas Trigger Workshop Athens December 3, 2012

FTK in τ triggers Phillip URQUIJO

Introduction

• The FTK will improve τ triggering.
• Primary vertex information allows
• τ-jet vertex association
• More accurate perigee for Δz0 and

pileup robustness

• Unified selection at L2
• tau triggering sequence is in 3 steps,

limiting effectiveness of MV selection

• τ selection efficiency
• Improved track cluster shape

resolution.

• PV reweighting of calo. based selection.

2

0.4 0.2

pile-up tau underlying event calculate REM, Rtrack in cone count # tracks in cone ∆R

PV

1. Vertex association 2. IP resolution

Ryan Reece (Penn)

• PV pT(track)
• all pT(track)

Fraction of track p from the p vertex.

!"#$%"&'()"*)+ !#,)-.!'()"*)+ !#,)-.!'()"*)+

“Jet V

b e a m l i n e

FTK in τ triggers Phillip URQUIJO

2012 Pileup dependence & Tracking

• Most pileup dependence in

2011 was due to Energy- weighted calorimeter radius variables using the EM cal.

• Remedied in 2012 in 3 ways:
• Smaller cone sizes.
• Removed most sensitive calo.

criteria: variables in τ recon. now based on track info.

• Track shapes based on

association/distance to highest pT track (Δz0) - pseudo PV reference.

4

FTK in τ triggers Phillip URQUIJO

Remaining Pile-up Dependence

5

• Δz0 is effective but imperfect. In 2012

some loss w.r.t. offline τ, partly due to tracking/PV:

• RoI limitations on track finding
• PV location (with TJVA), for IP & flight
• nVtx reweighting of calo. thresholds
• Δz0 criterion reduces separation

power for τ ID.

raction of track pT from the primary vertex.

10

JVF(jet, vertex)

• tracks matched

to jet pT(track)

• tracks matched

to jet and vertex

• pT(track)

= “Jet Vertex Fraction”

( ) ( )

FTK in τ triggers Phillip URQUIJO

Tau-jet vertex association: Offline

6

Number of tracks

1 2 3 4

τ Number of selected 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 =0 µ =20 µ =20 with TJVA µ Simulation Internal ATLAS

(a) True 1-prong taus

Number of tracks

1 2 3 4

τ Number of selected 0.1 0.2 0.3 0.4 0.5 0.6 =0 µ =20 µ =20 with TJVA µ Simulation Internal ATLAS

(b) True 3-prong taus

[GeV]

p Tau charged pion 5 10 15 20 25 30 35 40

E 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 =0 µ =20 µ =20 with TJVA µ Simulation Internal ATLAS

(a) True 1-prong taus

[GeV]

p Tau charged pion 5 10 15 20 25 30 35 40

E 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 =0 µ =20 µ =20 with TJVA µ Simulation Internal ATLAS

(b) True 3-prong taus

• TJVA is used offline:
• Choose vertex with highest

JVF for that τ candidate.

• Recovers Ntrack bin

migration - important since cuts are Ntrack dependent.

• Largest effects at low

track pT at high pileup.

• Need good IP resolution

for low pT τ tracks.

Parameters fJVF ∆Z(trk, vtx) 2mm ∆Z(trk, vtx)/σ(Ztrk) 1000 d0 2mm d0/σ(d0) 1000

1P 1P MP MP

FTK in τ triggers Phillip URQUIJO

Improved perigee for Δz0: Trigger

7

(wrt true PV) [mm]

z

• 3
• 2
• 1

1 2 3 s

• Number of

0.05 0.1 0.15 0.2 0.25 0.3 0.35

(wrt lead track) [mm] IPz

• 3
• 2
• 1

1 2 3 Number of tracks 2000 4000 6000 8000 10000 12000

µ 5 10 15 20 25 30 35 40 Track selection efficiency 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

• |Default

• |ex00

• µ

5 10 15 20 25 30 35 40 Track selection purity 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

• |Default

• |ex00

• d0&z0 depend on choice
• f PV:
• 4 techniques examined

for trigger (K.G. Tan): 1.Default 2.Recalculate perigee at (0,0) 3.beam spot BS 4.TJVA

• TJVA provides well

resolved dz0.

• BS will provide similar

performance of Efficiency/Purity if accurate (offline BS used for this study).

FTK in τ triggers Phillip URQUIJO

Improved perigee for Δz0: Trigger

7

(wrt true PV) [mm]

z

• 3
• 2
• 1

1 2 3 s

• Number of

0.05 0.1 0.15 0.2 0.25 0.3 0.35

(wrt lead track) [mm] IPz

• 3
• 2
• 1

1 2 3 Number of tracks 2000 4000 6000 8000 10000 12000

µ 5 10 15 20 25 30 35 40 Track selection efficiency 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

• |Default

• |ex00

• µ

5 10 15 20 25 30 35 40 Track selection purity 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

• |Default

• |ex00

• d0&z0 depend on choice
• f PV:
• 4 techniques examined

for trigger (K.G. Tan): 1.Default 2.Recalculate perigee at (0,0) 3.beam spot BS 4.TJVA

• TJVA provides well

resolved dz0.

• BS will provide similar

performance of Efficiency/Purity if accurate (offline BS used for this study).

0.7 0.8 0.9 1

• µ

5 10 15 20 25 30 35 40

EF/Offline BDT comparison

1. Track variable resolution 2. PV weighting

FTK in τ triggers Phillip URQUIJO

Selection at EF and Offline

9 ⚈*: Pileup corrected variables

• 2012 selection largely track based, calorimeter quantities are sensitive

to pileup. (dZ0 requirement). FTK could help with 1.nVtx dependent calo threshold reweighting, and 2.track cluster shape & IP resolution

N(vertex)

ATLAS Preliminary

2 4 6 8 10 12 14 〉

EM

R 〈 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22

jets true τhad p i l e

• u

p

Variables ariables 1P 1P MP MP Variables ariables code name EF Off. EF Off. Rtrack

trkAvgDist

⚈ ⚈ ⚈ ⚈

mtracks

massTrkSys

⚈ ⚈

Track STflight

trFlightPathSig

⚈ ⚈

Track Slead track

ipSigLeadTrk

⚈ ⚈

Nwidetrack

nWideTrk

⚈ ⚈

ΔRmax

dRMax

⚈ ⚈

Calo. fcore

centFrac

⚈ ⚈* ⚈ ⚈*

Calo. +Track ftrack

etOverPtLeadTrk

⚈ ⚈* ⚈ ⚈*

Nvtx

fcore = P∆Ri<0.1

i∈all

ET,i P∆Rj<0.2

j∈all

ET,j

∆Rj<conesize

X

j∈all

ET,j

f 0

core = fcore + 0.006 · Nvtx

FTK in τ triggers Phillip URQUIJO

Pile up corrected variables (Offline)

10

linear fit: slope: -0.006 slope: -0.000

nVtx y

• n

2011 Ztautau MC

fcore

Nvtx

f’core fcore

linear fit: slope: 296 slope: 74

nVtx r ]

2011 Ztautau MC

Nvtx

ΔR<0.4 ΔR<0.2

NVtx pileup corrected

fcore

FTK in τ triggers Phillip URQUIJO

Offline - EF comparison of Track shapes

11

• Track variables

with poor EF resolution

• µ=30
• IP dependent

variables.

• Room for improvement in EF resolution of track observables, if

PV, and RoI limitations can be avoided (A. Tanasijczuk)

Nwidetrack Rtrack Slead track STflight

FTK in τ triggers Phillip URQUIJO

Sensitive variables (Not currently used)

• Many variables not used in selection in 2012
• ffline and online, due to pile up sensitivity.
• Although offline could deal with it, offline

BDT simplified for harmony with trigger. Rejection power not optimal but stable efficiency.

• With PVs (multiplicity&positions) can apply

nVtx dependent cuts to e.g. EM calo. radius. to remedy pileup dependence.

• MVA training can be performed to be flat in

pT & nVtx

• Then we can consider more calo. variables.
• Preselection reduces rejection power: Smaller

∆R for REM 0.4 → 0.2, ∆z0 at 2mm. So it could also be revisited.

12

• e.g.Variables not (directly)

used ...

• Calo-radius (EM&Had)
• Topocluster-invariant mass
• Ratio of Isolation to Core

Scalar PT Sum

• Electron veto (TRT HT)
• Stripwidth

EM

R 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Normalised events 0.02 0.04 0.06 0.08 0.1 0.12 ATLAS Preliminary 1-prong

Unified L2 selection

Use FTK to remove calo preselection stage from L2

FTK in τ triggers Phillip URQUIJO

Selection at L2 & Multivariate feasibility

14 Step Variable

• 1. L2Calo

EtRawMin

• 1. L2Calo

CoreFractionMin

• 2. L2Track

NtrkMax

• 2. L2Track

SumPtRatioMax

• 3. L2Tau

EtOverPtLeadTrkMax

• 3. L2Tau

TrkAvgDistMax

• 2012: Largely track based, calorimeter quantities are sensitive

to pileup. (dZ0 requirement).

• Rejection ~ 10, c.f. >100 at EF.
• Unlike EF & Offline, selection is 3 steps ε(FEX+HYPO)3.

L2 MV algorithms severely limited: rejection factors kept low to minimise efficiency loss (L2 Calo. energy resolution is poor, and not ideal for a first step

• f selection)
• Limiting factor for a 1 step MV is tracking timing.

Algorithm CPU time (ms) 10 20 30 40 50 60 70 80 90 100 Algorithm calls 1 10

2

10

3

10 ATLAS data 2012 L2StarB

tableChnL2 L2_tau29_medium_L2StarB calls_cpu_time

2stTest

FTK in τ triggers Phillip URQUIJO

L2 BDT: 2 Step approach

• New study: BDT with 2 step

selection

1.BDT Calo (acceptance=0.5) 2.BDT Tau (track+tau)

• Rejection: 10 (current)→~15,

Efficiency: +~10%

• Test requires 2x more input

variables (more pileup dependence) than current L2.

• Needs nVtx to work reliably.
• Feasibility for unified BDT

expected to be much better (in preparation, A. Karamaoun)

15

HADtoEMenergy := HADenergy/EMenergy HADRadius IsoFrac stripWidth CoreFrac CaloRadius EMFrac EnergyTonCells scalarPtSumIsoToCore nCoreTracks chargeTrans := ||charge| − 1| trkAvgDist etOverPtLeadTrk

CoreFrac EnergyTonCells := (EMenergy + HADenergy)/numTotCells background roi rejection 0.88 0.9 0.92 0.94 0.96 0.98 1 signal roi acceptance 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

00001001 τ , τ → Z 00001001 τ , τ → Z

BDT Calo

BDT τ

⚈ ⚈ ⚈ ⚈ ⚈ ⚈ ⚈ ⚈

Current Cut BDT

FTK in τ triggers Phillip URQUIJO

Selection at L2 & Multivariate feasibility

16

• Soshi will discuss this tomorrow.

Step$1$ Step$2$ CALO$ reconstruc3on$ TRACK$ reconstruc3on$ BDT?$Selec3on$1$ Input$ BDT?$Selec3on$2$ Trigger$accept$ BDTF1$ training$ BDTF2$ training$ CUT?$ CUT?$

Under$$ development$

Input$ CALO$ reconstruc3on$ TRACK$ reconstruc3on$ BDTF1$ training$ Trigger$accept$ BDT$Selec3on$ If$FTK$is$not$fully$func3onal,$ If$FTK$is$fully$func3onal,$ 1Astep&CALO&TRACK&reconstruc(on.& Much&bejer&for&MVAtrigger&at&L2&

December.03.2012$ trigger$workshop$ 22$

FTK in τ triggers Phillip URQUIJO

Conclusion

• PV information will allow for better track selection, and EF-offline

convergence.

• At L2, it will enable MVA selection, impossible with the current 2/3 step

procedure.

• Plans for 2013: Quantify full impact of an FTK in 4 key areas:
• PV location information for track selection: IP(Δz0), flight
• L2 unified selection for an MVA.
• nVtx dependent or corrected selection.
• Extensions to current set of selection variables & potential for wider track

RoIs.

• FTK simulation appears to be working for τ’s already, so we can get on with

testing its full potential!

https://indico.cern.ch/getFile.py/access? contribId=173&sessionId=25&resId=0&materialId=slides&confId=158040

17

Glossary

FTK in τ triggers Phillip URQUIJO

EF BDT Variable definitions

19

Variables ariables code name Description Rtrack

trkAvgDist Track radius (pT weighted)

mtracks

massTrkSys Mass of the track system

Track STflight

trFlightPathSig Transverse flight path significance

Track Slead track

ipSigLeadTrk Leading track impact parameter significance

Nwidetrack nWideTrk

Number of isolation region tracks

ΔRmax

dRMax Maximum dT between jet-axis and core tracks

Calo. fcore

centFrac Core energy fraction

Calo. +Track ftrack

etOverPtLeadTrk Leading track momentum fraction

FTK in τ triggers Phillip URQUIJO

L2 Variable definitions

20 Stage Variable Description

• 1. L2Calo

EtRawMin L2 Et

• 1. L2Calo

CoreFractionMin Core energy fraction

• 2. L2Track

NtrkMax Track multiplicity

• 2. L2Track

SumPtRatioMax Scalar sum of track pT in core w/r/t isolation region

• 3. L2Tau

EtOverPtLeadTrkMax Leading track momentum fraction

• 3. L2Tau

TrkAvgDistMax Track radius (pT weighted)