Development of a Particle Development of a Particle Flow Algorithms (PFA) at Flow Algorithms (PFA) at

ArgonneArgonne

Presented by

Lei Xia ANL - HEP

June 5-9, 2006 CALOR 2006

IntroductionIntroduction Measure jets in the PFA way…Measure jets in the PFA way…

Clear separation of the 3 parts is the key issue of PFAClear separation of the 3 parts is the key issue of PFA Charged particle, photon and neutral hadron: all deposit their energy in the calorimCharged particle, photon and neutral hadron: all deposit their energy in the calorim

eterseters Maximum segmentationMaximum segmentation of the calorimeters is needed to make the separation poss of the calorimeters is needed to make the separation poss

ibleible One Major R&D issue: development of PFAOne Major R&D issue: development of PFA

Show that the ILC goal for jet energy resolution (30%/sqrtE) can be achieved by PFAShow that the ILC goal for jet energy resolution (30%/sqrtE) can be achieved by PFA Develop a PFA that can be used for detector optimizationDevelop a PFA that can be used for detector optimization

Argonne has two parallel efforts on PFA development, this talk shows result frArgonne has two parallel efforts on PFA development, this talk shows result from one of themom one of them

Particles in JetsParticles in Jets Fraction of jet energyFraction of jet energy Measured withMeasured with

ChargedCharged 65% Tracker, negligible uncertainty

PhotonPhoton 25% ECal, 15%/ √ E

Neutral hadronNeutral hadron 10% ECal + HCal, ~50-60%/ √ E

June 5-9, 2006 CALOR 2006

Perfect PFA: NO algorithm effectPerfect PFA: NO algorithm effect

Take MC track momentum as the energy of charged particlesTake MC track momentum as the energy of charged particles Remove calorimeter hits associated with charged particles by looking at MC infomationRemove calorimeter hits associated with charged particles by looking at MC infomation Sum up everything else in the calorimeter as neutral energySum up everything else in the calorimeter as neutral energy

Apply appropriate sampling fractions for photon hits and neutral hadron hits Apply appropriate sampling fractions for photon hits and neutral hadron hits Use MC information to separate photon hits and neutral hadron hitsUse MC information to separate photon hits and neutral hadron hits

Z-pole events, just event energy sum, no jet algorithm appliedZ-pole events, just event energy sum, no jet algorithm applied

Example: SiD aug05_np

central peak ~2.3 GeV (~80%) (no event selection)

We have room for PFA development

June 5-9, 2006 CALOR 2006

PFA effort: overviewPFA effort: overviewCalorimeter Hits

Reconstructed TracksCalorimeter Clusters

ClusteringAlgorithm

PhotonIdentification

EM Clusters Hadron Clusters

‘Neutral’ Clusters Matched Clusters

Track-clustermatching

Charge fragmentidentification

Neutral Clusters Fragments

Ephoton Eneu-had 0 0 Ptrack

Hadron sampling fraction

EM sampling fraction

Total event

energy

Tracker Hits

Track findingAlgorithm

This is strangeWill explain

June 5-9, 2006 CALOR 2006

Clustering algorithm: hit densityClustering algorithm: hit density

I

J RijVf

Vb

V1V2

}{

)||/|(|)||/|(|)||/|(| )(2

332

222

11

ij

VRVVRVVRVi

ijijij eeeD

With V3 = Vf (if (Vf•Rij) > 0) or Vb (if (Vb•Rij) > 0)

• Hit density reflects the closeness from one hit i to a group of hits {j}• {j} = {all calorimeter hits} to decide if hit i should be a cluster seed• {j} = {all hits in a cluster} to decide if hit i should be attached to this cluster

• Consider cell density variation by normalizing distance to local cell separation • Density calculation takes care of the detector geometry• Clustering algorithm then treat all calorimeter hits in the same way

June 5-9, 2006 CALOR 2006

Clustering algorithm: grow a Clustering algorithm: grow a clustercluster

Find a cluster seed: hit with highest density among remaining hitsFind a cluster seed: hit with highest density among remaining hits Attach nearby hits to a seed to form a small clusterAttach nearby hits to a seed to form a small cluster Attach additional hits based on density calculationAttach additional hits based on density calculation

i = hit been considered, {j} = {existing hits in this cluster}i = hit been considered, {j} = {existing hits in this cluster} EM hits, DEM hits, Dii > 0.01 > 0.01 HAD hits, DHAD hits, Dii > 0.001 > 0.001 Grow the cluster until no hits can be attached to itGrow the cluster until no hits can be attached to it

Find next cluster seed, until run out of hitsFind next cluster seed, until run out of hits

Hit been considered

seed

Hits of a cluster

June 5-9, 2006 CALOR 2006

Density driven clusteringDensity driven clustering

ParticleParticle ECal hit efficienECal hit efficiencycy

HCal hit efficienHCal hit efficiencycy

Overall hit Overall hit efficiencyefficiency

Overall energy Overall energy efficiencyefficiency

Photon (1GeV)Photon (1GeV) 89%89% 43%43% 89%89% 91%91%

Photon (5GeV)Photon (5GeV) 92%92% 54%54% 92%92% 96%96%

Photon (10GeV)Photon (10GeV) 92%92% 61%61% 92%92% 97%97%

Photon (100GeV)Photon (100GeV) 95%95% 82%82% 95%95% >99%>99%

Pion (2 GeV)Pion (2 GeV) 78%78% 59%59% 75%75% 71%71%

Pion (5 GeV)Pion (5 GeV) 81%81% 70%70% 79%79% 80%80%

Pion (10GeV)Pion (10GeV) 84%84% 80%80% 83%83% 85%85%

Pion (20GeV)Pion (20GeV) 85%85% 87%87% 88%88% 91%91%

• Typical electron cluster energy resolution ~ 21%/sqrt(E)• Typical pion cluster energy resolution ~70%/sqrt(E)• All numbers are for one main cluster (no other fragments are included)

June 5-9, 2006 CALOR 2006

Cluster purity : Z pole (uds) eventsCluster purity : Z pole (uds) events

Most of the clusters (89.7%) are pure (only one particle contributes)Most of the clusters (89.7%) are pure (only one particle contributes) For the remaining 10.3% clustersFor the remaining 10.3% clusters

55% are almost pure (more than 90% hits are from one particle)55% are almost pure (more than 90% hits are from one particle) The remaining clusters contain merged showers, some of them are ‘trouble makers’The remaining clusters contain merged showers, some of them are ‘trouble makers’

On average, 1.2 merged shower clusters/Z pole eventOn average, 1.2 merged shower clusters/Z pole event This will result in double counting or underestimating of jet energy which leads to poor resolutionThis will result in double counting or underestimating of jet energy which leads to poor resolution Will re-visit clustering algorithm after other PFA components are more of less settledWill re-visit clustering algorithm after other PFA components are more of less settled

Number of contributingparticles in a cluster

Fraction from largest contributorfor clusters with multi-particles

June 5-9, 2006 CALOR 2006

Photon id – longitudinal H-matrixPhoton id – longitudinal H-matrix

EEphotonphoton 0.100.10 0.250.25 0.500.50 0.750.75 1.01.0 2.02.0 5.05.0 10.10. 20.20. 50.50. 100.100.

Eff(%)Eff(%) 1.131.13 43.943.9 89.789.7 95.795.7 96.896.8 98.498.4 98.398.3 96.696.6 93.993.9 83.683.6 52.552.5

If I use a single cut: log(Prob(chisqD)) > -10If I use a single cut: log(Prob(chisqD)) > -10log(Prob(chisqD)) is calculated from default H-matrixaccumulated from a wide energy range of photons

Trouble comes in at the two endsBut that’s not all yet…

Photon: 1GeV Photon: 5GeV

June 5-9, 2006 CALOR 2006

Photon id – longitudinal H-matrixPhoton id – longitudinal H-matrix

Efficiency of photons still need to improveEfficiency of photons still need to improve Try to accumulate longitudinal H-matrix(es) at smaller photon energy regionsTry to accumulate longitudinal H-matrix(es) at smaller photon energy regions

Efficiency of hadrons is way to highEfficiency of hadrons is way to high Take neutral hadrons as photons results in using wrong calibration constantTake neutral hadrons as photons results in using wrong calibration constant Take charged hadrons as photons results in double counting of energy directlyTake charged hadrons as photons results in double counting of energy directly

Current (temporary) solution: still subject ‘photons’ to track-cluster matching which will Current (temporary) solution: still subject ‘photons’ to track-cluster matching which will recover charged hadrons but as the same time will lose some real photonsrecover charged hadrons but as the same time will lose some real photons

Will use more variables to eliminate hadrons: first IL layer, shower depth/shape, etc.Will use more variables to eliminate hadrons: first IL layer, shower depth/shape, etc. A single cut on longitudinal H-matrix is NOT enough to identify photonsA single cut on longitudinal H-matrix is NOT enough to identify photons

EE 1.01.0 2.02.0 5.05.0 10.10. 20.20.

effeffK0K0

(%)(%) 26.126.1 30.830.8 33.233.2 36.136.1 35.235.2

effeffnn

(%)(%) 6.616.61 18.618.6 33.033.0 38.38. 37.737.7

effeffnbarnbar

(%)(%) 49.149.1 48.848.8 46.946.9 40.940.9 38.538.5

neutron: 5GeV

E(Pi-)E(Pi-) 2.02.0 5.05.0 10.10. 20.20.

Eff (%)Eff (%) 13.313.3 8.18.1 7.07.0 3.83.8

June 5-9, 2006 CALOR 2006

Photon id – try H-matrix(E)Photon id – try H-matrix(E)

Efficiency of photon looks much better, however, at the cost of Efficiency of photon looks much better, however, at the cost of accepting even more hadrons (not shown here)accepting even more hadrons (not shown here)

The energy range that one H-matrix can cover is still to be studied The energy range that one H-matrix can cover is still to be studied I am studying other variables to remove hadronsI am studying other variables to remove hadrons

First interaction layer, shower size/shape, etcFirst interaction layer, shower size/shape, etc

Eventually, after photon identification is done rather well, I will no Eventually, after photon identification is done rather well, I will no longer subject identified photons to track-cluster matchinglonger subject identified photons to track-cluster matching

EEphotonphoton 0.100.10 0.250.25 0.500.50 0.750.75 1.01.0 2.02.0 5.05.0 10.10. 20.20. 50.50. 100.100.

Eff(%)Eff(%) 95.695.6 95.995.9 96.996.9 97.497.4 97.697.6 97.497.4 98.598.5 98.398.3 97.997.9 97.197.1 97.097.0

Still use a single cut: log(Prob(chisqD)) > -10Still use a single cut: log(Prob(chisqD)) > -10But log(Prob(chisqD)) is calculated from individual H-matrixaccumulated at the same energies of the photons

June 5-9, 2006 CALOR 2006

Charge fragment Charge fragment identification/reductionidentification/reduction

Use geometrical parameters to diUse geometrical parameters to distinguish real neutral hadron cluststinguish real neutral hadron clusters and charge hadron fragmentsers and charge hadron fragments

Energy of matched clusters

From neutralparticles

From chargedparticles

Energy of clusters not matched toany track: neutral candidate

From neutralparticles

From chargedparticles(fragments)

From neutralparticles

From chargedparticles(fragments)

After charge fragmentidentification/reduction

1 : 1.24 0.88 : 0.35

June 5-9, 2006 CALOR 2006

PFA: Z-pole (uds) performancePFA: Z-pole (uds) performance

Barrel events: 60%

3.22 GeV @88.2GeV 59% 9.95 GeV 41%

All events:

3.41 GeV @87.9GeV 58.5% 10.4 GeV 41.5%

SiD aug05_np

Barrel: -45 deg < Theta (uds quark) < 45 deg

SiD aug05_np

The broad tails of the distribution need to be reduced significantly, otherwise, physics performance is not going to be good…

June 5-9, 2006 CALOR 2006

TimelineTimelineC

on

firm

/tun

e M

C sim

ula

tion

Test BeamPFA development

Working PFA

Z-pole, 2 jets

>= 500 GeV, multiple jets

Establish jetPFA E~

Optimize detector design

Done

In progressShould be done by a few months’effort

Hard to tell at this moment(doesn’t seem to be a trivial job!)

June 5-9, 2006 CALOR 2006

Summery Summery

Particle Flow Algorithms are being developed at Argonne Two ‘complete’ PFA’s are available to play with Performance of both PFA’s looks promising, and will be improved

Current performance of this PFA at Z-pole looks promising but not good enough yet Already identified several components that need to be worked on Will continue to work on it in the next few month Need to achieve ~30%/sqrtE with small tail component at this energy

Need to study PFA performance over the entire ILC interested jet energy range and with more complicate final states Need to show that ILC jet energy resolution goal can be achieved Get PFA ready for optimizing detector design

Test beam data need to come in time!