Overview of Tracking for Low Pt Charged Hadronsbmangano/CMS_talks/ByBoris/lowPtTracking...Overview...

B.ManganoJoint Tracker/Physics Workshop - Low Pt tracking

Boris Mangano

(University of California, San Diego)

Overview of Tracking for Low PtCharged Hadrons


Outline

• One slide description of track reconstruction in CMS

• trajectory seed generation for low pt tracking

-“necessarily incomplete” history of previous algorithms

- currently available algorithms

• seed collection cleaning

• hit-locking and iterative reconstruction


Tracking modules in CMSSW

Trajectory seeding

Trajectory fitting

Trajectory building

Initial estimate of the trajectoryparameters from a small subset oftracker measurements

An iterative process which collects all themeasurements associated to the samecharged particle.

Final fit of the full set of measurementsassociated to the same charged particle.

Track collectionfiltering

Removal of “ghost” tracks (fakes)


a) seed generator from hit pairs + beam spot constraint:

- standard for tracking in pp collisions from ORCA till CMSSW_18X

- described in CMS Note 2006/026 (M.Konecki, T.Todorov, D.Kotlinski, S.Cucciarelli)

a’) extension of a) to use hit triplets

- same authors and CMS Note

- used for vertexing at HLT. Never used for offline tracking becauseof intrinsic inefficiency of triplets on 3 layers

- Optimized for tracks with pt > 1 GeV

b) use of triplets a’) to seed offline tracking in heavy-ions collision

- it guarantees high purity of the seed collection even for low pt tracking in environment with a very high multiplicity of charged particles.

- CMS RN 2003/003 (C.Roland)

Brief history of trajectory seed generation in CMS


c) seed generator from “hit pair finder a)” + “new third hit association”

- third hit found using exact helix propagation

- multiple-scattering parameterization from a) exploited to guaranteesefficiency at very low pt

- Optimized down to 0.1 GeV

- Very detailed analysis of performance for low pt tracking culminatedin CMS AN-2006/100 (F.Sikler)

d) small adjustments of implementation a’), triggered by results shown in c)

- dynamic “search window” for 3rd hit is reliable also for tracks with Ptbelow 1 GeV.

- http://indico.cern.ch/getFile.py/access?contribId=4&sessionId=2&resId=0&materialId=slides&confId=10929 (Marcin Konecki)

- Integrated into standard reconstruction sequence since CMSSW_1_3_X



e) combination of “seeder from triplets d)” + “seeder from hit pairs with vertexconstraint”

- hit pairs can recover part of the intrinsic triplets inefficiency

- seed collection with higher purity thanks to the vertex constraint onhit pairs

- Use ~99% of standard/previous seeding code for pp collision

- Based on original work of D.Kotlinski, implemented in CMSSW byG.Boudoul, T.Legrand, G.Petrucciani

http://indico.cern.ch/getFile.py/access?contribId=4&sessionId=1&resId=0&materialId=slides&confId=22542


Default seeder for tracking in pp collisions since CMSSW_200pre2


Performance of “original” seed generator fromhit triplets

algorithmic efficiency

global efficiency algorithmic efficiency

global efficiency

ORCA days and firstCMSSW releases

since CMSSW_1_3_X

configuration a’) was not optimized for lowpt tracking and high efficiency.

During the implementation, the mainconcern was speed

configuration d) was setup to copealso with low pt tracking

MarcinKonecki


Performance of “dedicated” seed generator fromhit triplets

algorithmic efficiency

pt [GeV]pt [GeV]

spectrum ofsimulated tracks

spectrum ofreconstructed tracks with“dedicated” algorithm

spectrum of reconstructedtracks with “standard”algorithm in ORCA

FerencSikler


Performance of “default” seed generator for ppcollision in CMSSW_20X

standard configuration till CMSSW_18X(min Pt = 0.9 GeV)

current standard configuration sinceCMSSW_20X (min Pt = 0.3 GeV)

new configuration with lower minPtthreshold (0.1 GeV)

pt [GeV]

global efficiency

CMSSW_2_0_0_pre2 release

MinBias RelVal sample


Performance of “default” seed generator for ppcollision in CMSSW_20X

standard configuration till CMSSW_18X(min Pt = 0.9 GeV)

current standard configuration sinceCMSSW_20X (min Pt = 0.3 GeV)

new configuration with lower minPtthreshold (0.1 GeV)

pt [GeV]

mul

tiplic

ity o

f rec

onst

ruct

ed tr

ack CMSSW_2_0_0_pre2 release

MinBias RelVal sample

with lower min pt threshold:

- we get about +30% tracks --> not negligible impact on Event size if we decideto store all of them.

-cpu time increases from 20% (MinBias) to 40% (Ttbar)


Comments on current seed generators

1) seeder by Sikler et al. (configuration c) has been fully studied and optimized forlow pt tracking, but:

- performance for medium/high pt tracks ?

- general timing performance ?

2) seeder by Konecki et al. (configuration a) , d) and e) ) was used since ever inCMS and is reliable for high pt tracks, but:

- Very detailed studies in the [0,0.9] GeV pt region are not available yet

A new comparison based on efficiency, fake rate and timing should be repeated

Can we converge on a single seed generator which is suitable for both low pt andhigh pt tracking ??

seed collection cleaning

Number of fake seeds can be greatly reduced requesting the “compatibility”between the seed direction and the cluster shape of measurements

higher purity of seed collectionimplies:

- faster overall reconstruction

- lower tracking fake rate

picture byF.Sikler


fake

rate

fake

rate

eta etaCMS AN-2006/100 (F.Sikler):

- optimized and fully studied on low pt tracking

- performance on high pt samples ??

seed collection cleaning (II)

F. Sikler


Similar approach based on cluster shape (pixel templates + stubs)has been developed by another group:

M.Swartz, D.Fehling, A. Domingues et.al

CMS NOTE 2007/033


- testing on general tracks (pt >1 GeV range) is proceeding well- Pixel cluster parameter estimator based on “template” is alreadythe default one used during the track fitting


- what are the performance of this seed collection cleaner in the lowpt range ??

seed collection cleaning (III)


seed collection cleaning (IV)

- Similarly to the seed generation, 2 different but similar approaches are availablefor the seed collection cleaning

-They have been developed and tested mainly in different pt regions

- It is not apparent whether one can or cannot work properly in the domain of theother

1) perform comparisons in the same pt domains

2) understand differences and analogies

Can we converge on a unique and general seed cleaner based on cluster shapecompatibility ??

hits removal (locking)

general idea:- reconstruct “easy” (high pt) tracks during a first iteration- “lock” the corresponding hits- perform the reconstruction of low pt tracks using only the remaining “un-locked”hits

because the hit multiplicity is reduced, we should benefit in:- lower fake rate- easier and faster pattern recognition

explored by- F.Sikler for V0 reconstruction- M.Pioppi and P.Janot for reconstruction of primary tracks in general pp collisions


hits removal (locking)the results obtained till now for primary tracks reconstruction are not thrilling:

- the fraction of hits removed on the innermost pixel layers is not enough tosensibly reduce the fake rate due to short “ghost” tracks

- removal of hits associated to fake tracks can compromise the reconstruction ofgood tracks

1) New tool which performs a “selective locking” of hits has been implementedrecently (CMSSW 20X) by G.Petrucciani:

hits are removed only if they are fully compatible with the reconstructed trackand they are not due to “merged” clusters


2) Hopefully the fraction of hits removed in the innermost layers can be

increased reconstructing completely the “looping” charged particles (see nextslide)

“looper” charged particles

tracker layers in r-phi view

looping chargedparticle Current trajectory builder is not able

to “follow” a track which crosses alayer twice and in opposite directions.

A looper is reconstructed “partially” oras different separate track segments

The necessary extension of the currenttrajectory builder is not too difficult

Small and interesting project for people whowant to start contributing to the tracking effort


Conclusion

- Special seed generator and cleaner have been developed specifically forlow pt tracking

- Similar tools are available for reconstruction of high pt tracks

New comparisons should be performed to understand if the different approaches can benefit from each other

- We should try to converge on common code avoiding duplications anddivergent development branches

- It is probably worth to further explore the “hit removal” approach and thecomplete reconstruction of looping particles


old standard tracking + newtrack collection filter

old standard tracking

QCD 80-120sample

old standard tracking+ new track collectionfilter

old standard tracking

CMSSW_16X

track collection filter based on vertexcompatibility in CMSSW_20X

Implementation thanks to:- P.Azzurri (SNS Pisa)Similar approaches by M.Pioppi,P.Janot and F.Ambroglini for ppcollisions and by heavy-ions group

|η|

|η|Fake rate is almost completely removedwithout spoiling the track reconstructionefficiency

selection is based on d0 and dz (andsignificances) respect to reconstructedprimary vertex

efficiency and fake forpt > 0.9 GeV

Overview of Tracking for Low Pt Charged Hadronsbmangano/CMS_talks/ByBoris/lowPtTracking...Overview...

Documents

Transcript of Overview of Tracking for Low Pt Charged Hadronsbmangano/CMS_talks/ByBoris/lowPtTracking...Overview...