Impact Parameter b-jet tagging at Level-3 presented by Per Jonsson Imperial College London Thanks to Chris, Ray, Tim, Michele, Amber, Makoto, Andre, Michiel,

Impact Parameter b-jet tagging at Level-3

presented by

Per JonssonImperial College London

Thanks to Chris, Ray, Tim, Michele, Amber, Makoto,

Andre, Michiel, Jon, Stephen All DØ Meeting June 25th 2004

ADM 06/25/04 Per Jonsson - Imperial College London 2

Outline• Introduction

– What’s the aim?

• Level-3 tools– Primary Vertex

reconstruction• Beam spot

– Impact Parameter b-tagger: L3TBTagIP

• Performance– MC comparison– Real data offline/online

• Physics– Trigger list v13:

• hbb• Z->bb• ZH->nnbb


Introduction

• Why?– Many search signals are swamped by light quark QCD

background, e.g. Z->bb and Higgs– Light quark rejection needed already at trigger level to achieve

acceptable trigger rate at L > 50E30: alternative is pre-scaling!• Method:

– 2 dimensional Signed Impact Parameter based b-jet tag– Use reconstructed Level-3 Tracks, Jets and Primary Vertex from

other tools• Beam spot position and tilt from Vertex Examine

• Tool Results:– Track, jet and event probabilities– Continuous variables [0,1]:

• Easy to control rate/efficiency by simple cuts

Aim: provide b-tagging at Level-3 for high pT jet physics events,enable various topology dependent physics filters.

A set of track only based IP tools/filters for B-physics triggering also exist at L3, see talks by C. Barnes, R. Jesik


Jet based IP b-tagging an Overview

• Select well reconstructed 3D SMT+CFT tracks– Need tracker: L3TGlobalTracker; pT > 0.5 GeV

• Recalculate track parameters wrt Primary Vertex– Need Primary Vertex reconstruction: L3TCFTVertex and L3TXYVertex

• Match each track to closest jet– Need jet-tool: L3TJet; 0.5 cone jets, ET > 9 GeV

• Calculate the signed Impact Parameter of each track– Sign from crossing of track-jet axis:

• positive if downstream of PV• negative if upstream of PV, i.e. non-physical solution

• Construct track probability based on IP significance and fitted resolution function– Need well determined track errors, also true for PV rec.

• Combine track probabilities to jet and event tags

What do we want to do and what tools do we need?

Construct physics filters to reject background and achieve acceptable rate

Overview

Jet/Event tag


Primary Vertexing; the Z-coordinate

• Uses Pt weighted histogramming:histogram tracks according to z0weight bin by total Pt

choose highest 2 adjacent bins and average values

Bs-> Ds π + 0.8 min bias event

From Chris

Also important to sharpen jet turn-ons….and reject events outside SMT

Resolution < 2 mm

min bias hard scatter

Chris Barnes, Imperial CollegeFresno June 2004

DCA resolution in data• Pscat = P sin3/2 • Material causes multiple scattering• Large Pscat – low multiple scattering

•~ 15 m asymptotic value!

• Impact parameter minimization in 2 dimensions using linear tracks • Requires inversion of one 2×2 matrix - FAST

yxi

bi

numtracks

iipi

iii VVdVdr

,2

2

12

2

2

)()(

.

Minimise

Primary Vertexing; the XY-coordinates From Chris

Average number of PV tracks in data =6


Beam spot information

• Why do we need the beam spot information?– Starting/reference point for the Primary Vertex tool:

• Linearization of tracks• Constraint for Primary Vertex fit

– Enters equation with weight of ~3 GeV track, non removable

• Returned as PV if algorithm fails,• Beam tilt, sizeable correction.

• How does the information reach Level-3?– The beam position/tilt is measured by Vertex Examine

every few minutes:• DCA vs phi/Z fit to samples of L3 tracks

– COOR command: l3update transmits the information to Level-3

– ScriptRunner writes the names and values to NameValueTable

– L3TXYVertex reads the information at RunInit.

Beam spot stored both in database and in data: debugchunk


Beam spot dependence

• What if the beam position moves within a run?– Beam spot rather stable, < 30 m, within stores– Impact of larger beam shifts has been tested

with MC and real data– Mean Reconstructed PV will shift by about 50%

of beam shift– Leads to larger IPs on average and higher

trigger rates• No loss of signal efficiency, but more bkg. triggers

– Cure: stop and restart new run

Shifters: Vertex Examine must run at ALL times in the control room!

Beam spot position picked up by Vertex tool at beginning of each run


Impact Parameter b-tag;track selection

• Track pT> 0.5 GeV

• DCA error < 0.75 mm

• Matched to jet in 0.7 cone

– Jet ET > 9 GeV

– Jet rank =< pjets; pjets=3,6

• Distance to PVZ < 2 cm• 0 < Impact Parameter < 0.25 cm• Decay Length < 1 cm

Loose track selection:

Average of 3 tracks/event

Track quality

Min. bias rej.

Long lived bkg


Multiple Scattering correction to DCA error

• Track errors from L3TGlobalTracker simple function of number of SMT hits

• Use QCD MC to obtain MS correction:– No beam size or PV uncertainty– Expected dependence on the width of DCA

significance:

f(Pscat)=(c02 + (c1/Pscat)2)0.5

New error correction in L3TGlobalTracker, used by both PV and b-tagger

QCD MC

Data after corr.


From IP to b-tag

Fit resolution function, Gauss + 2 exp, to negative side of real data

Use to calculate track/jet/event probabilities

bbdataQCD

Data:Track prob

Data: event tag

“b-signal”


Performance MCHow well does it work compared to offline JLIP?

Z->bb hbb

Black = 2 jetsRed = 2 JLIP tags

Black = 3 jetsRed = 3 JLIP tags

Eff vs L3 btag-cut Eff vs L3 btag-cut

Inefficiency MCWhere do we lose efficiency wrt JLIP?


Z->bb MC, 2 JLIP tags, L3 btag < 0.1

Eff vs ET Eff vs

Eff vs nr. of tracks Eff vs PVZ

Less eff at low ET, ntrack, high PVZ


Performance: online• Special mu+jet L3 IP tag test run

A track only based version of L3TBtagIP has also successfully been running online since last August collecting B-physics events

From Tim

L3

L3

L3

Single JLIP tagged events


Trigger list v13• Triggering on b-jet events w/o leptons at lum above 50E30 is

challenging:– Rates driven up by non-b backgrounds– Aim to have un-prescaled triggers

• Alternatives to trigger level b-tagging are largely exhausted!• The Higgs group will use the L3 b-jet tag for the following triggers in

v13:– hbb:

JT2_3JT15L_IP_VX, with L3 btag < 0.05– Z->bb:

ZBB_TLM3_2JBID_V, with L3 btag < 0.1

MUJ1_2JT12_LMB_V and MUJ2_2JT12_LMB_V, with L3 btag < 0.05 – ZH->bb:

JT1_ACO_MHT_BDV, with L3 btag < 0.3

• Many very nice studies by Tim, Amber, Makoto et al.– See Agenda server since last fall:

Higgs, Trigger, Multijet and TB meetings for details

• The first data collected with these triggers are currently being analyzed:– A first look confirms enhanced b content with L3 IP based triggering!

Latest results from MakotoSpecial v13 test runs:194266-7ZH->bb trigger: JT1_ACO_MHT_BDV, L3 btag<0.3


Black: all events

JLIP prob

Red: JT1_ACO_MHT_BDV

Nr of tagged jets

Loose L3 tag, so still dominated by light quarks, but nice b enhancement


Summary & Outlook• L3TBTagIP is tagging b-jet events at Level-3• Will be used by the Higgs group for several

triggers in the imminent trigger list v13• The first data taken by this trigger list are

just being analyzed• D0 notes describing the L3 tools are coming

– Future Improvements (p17):• Tuning of vertex errors• Improved fit of resolution function• 3D version

– Longer term:• Soft lepton tag• 2ndary Vertex tool• Combination

Documents

Impact Parameter b-jet tagging at Level-3 presented by Per Jonsson Imperial College London Thanks to Chris, Ray, Tim, Michele, Amber, Makoto, Andre, Michiel,