Status of jet reconstruction and Higgs searches at LHCb

M. Musy - 26th May, 2009

p

p

beam jet 1

beam jet 2

q

q'

W

H

b

b

lept

on

neutrino

jet b

jet b

Motivations

Jet reconstruction is important for searches of particles producing b-jets and displaced vertices, LHCb has a very good b-quark trigger and identification, and will be well calibrated with the large number of B mesons.

Standard Model Higgs of light mass going to bbar

Precision EW fits give MH = 126 +73/-48 GeV MH < 219 GeV at 95% C.L.

~~30% SM Higgs events 30% SM Higgs events are in LHCb acceptanceare in LHCb acceptance

MH=120 GeV

This is a field of search which is outside the main LHCb scope, still the potentialities of the detector are worth being investigated

SM predicts ~3 pb of HZ+HW at MH=120 GeV

→ about 100 evts in one year LHCb data (2/fb)

Associated productionof SM Higgs

p

p

beam jet 1

beam jet 2

q

q'

W

H

b

b

lept

on

neutrino

jet b

jet b

Sig

ma

(pb)

H Mass (GeV)

Selection: isolated high Pt prompt lepton (Pt>15GeV, IP/<3) to tag W or Z

Reconstruction of 2 b-jets in the LHCb acceptance

Nearness in angle Cone Algorithm.

Jet algorithms

2-seed algo Kt algo

22+ΔΔη=R

Reconstruction

Matching tracks and electromagnetic clusters

Use track and calorimetric information to build

4-MOMENTA

B-seed finding Kt jet algorithm

Cone algorithm with B-seed

B-jet tagging

B-jetsB-jets

2 B-seeds from 2-tracks sec. vertex

2 b-jets/eventcreated

For R=0.5~15 jets/eventreconstructed

2 tagged jets/event

Nearness in relative transverse momentum => Kt algorithm.

Cone jetKT jet

Seed algorithm

1. Track selection: build a 2-track vertex using cuts on

P > 2 GeV

2 < 2.5 IP/ > 2.5, 2.8 (wrt Primary VertexPrimary Vertex) Pt >0.6, 0.8 KS candidates are excluded in the mass

window M=[0.490;0.505] GeV2. Seeds selection. Select seeds on the base of a

classifier using kinematic variables of tracks3. Double seeds selection. Pairs of seeds are

passed to a second classifier, pick up the best seeds pair candidate according to classifier output.

4. Jets construction. Other particles are added to form a jet if particle belongs to a cone of ΔR=0.7

Coming from bAll

Coming from bAll

seedB

seedA

HW, HZ Monte Carlo events generated with MH=120 GeV and a lepton of Pt > 10GeV within the acceptance

no biases in the reconstructed jet direction

Theta of reco vertices

Th

eta

of tru

e v

ertic

es

Resolution ~14 mrad

Pt/GeV

jet1

WZ evtsttbarHiggs120

Jet 1Jet 2Jet 3

Pt/GeV

seed algorithm, cont’d

Kt algorithm

• Need to identify the b-jet

A jet then is defined as a b-jet if it has at least 7 particles at least 2 tracks with impact parameter significance>2 at least 1 tracks with impact parameter significance>5 energy of charged particles > 3% of the total energy energy in a cone of R=0.4 around the jet axis>70% Ejet

A neural net also trained to discriminate against c-jets and light jets is also applied using kin. variables

cutSelection efficiency for b-jets 80% with good non-b jet rejection

• Standard algorithm used by many experiments

• in LHCb Kt algorithm is used to make jets in the event with R=0.75

b-jetsc-jetslight jets

Calibration of jet energy

On a event-per-event basis: using MC information, take a few reference points in the mass spectrum and give each jet a momentum correction as a function of the Pt and rapidity of the jet such as to reproduce the peak. Samples used: WZ (91 GeV), H(120 GeV) and H(140 GeV) events

-> Fit for free parameters minimizing = (Mjj – Mx)2/N

Need to use more than one known mass point to avoid biases in the mass peak position..

Function chosen for the fit:

f(Pt, = a + b/(Pt –c) + d/

This factor is used to correct the momentum assuming Mjet = 5 GeV

before calibration

after calibration (points)

Pt(

H)

/ P

t(Je

t1+

Jet2

)

Pt(Jet1+Jet2)

Pt/GeV

Calibration flattens out the reconstructed Pt distribution of H

Not yet perfect, especiallyat low Pt, but this regionis anyway less populated

f

Jet 1Jet 2Jet 3

Pt(Jet) /GeVPt(Jet) /GeV

Jet 1Jet 2Jet 3

BEFORE calibration AFTER calibration

- average jet’s pt increase from ~35 to ~50 GeV- third jet is unchanged

ttbar

beforeafter

beforeafter

GeV Mjj Mjj

91 65 23 94 23

120 83 31 122 31

140 94 37 139 33

raw corrected

Mjj /GeV

Mjj /GeV

Z, 91 GeV

Higgs, 120 GeV

Higgs, 140 GeVAlso tt background gets shifted

..calibration of jet energy (cont’d)

Other possibility to calibrate: use MC information to fit directly the dependence of the jet reconstructed energy from the b-quark energy or from the visible energy in the detector.

- This can be done by means of a simple function:

Eje

t/E

b

- Or by means of a Neural Net which takes into account…

…variables like:

- energies measured by the tracking and the calorimetry (ECAL, HCAL, PRSH, SPD, Muon chambers)- pseudorapidity of the jet- granularity and saturation of the calorimeter cells

Eje

t ca

lbra

ted

Ejet true generated Higgs mass

reco

n’ed

mas

seffect of saturation

raw mass

calibrated mass

90 GeV120 GeV140 GeV

90 GeV120 GeV140 GeV

Jet calibration from dataCalibration or jet energy correction can be based on pt balance between Z and reconstructed Jet in Compton-like scattering events:

2500 events per year with 2 muons of pt>10GeVand the b quark in LHCb acceptance [0;400mrad]

initial state radiation and final state radiation can be reduced by requirementson extra jet activity and

back-to-back jet-Z system

- Preliminary selection with simple cuts gives an efficiency of 2 muons of about 61%- b-jet identification efficiency is about ~80%.- trigger efficiency: 88%- Jet reconstruction, 75% ~25% of the events with pt>10GeV

and the b in acceptance~620 events for 2/fb (1year)

Main limitation might be the available statistic. Studies are ongoing!

Use the same type of correctionfunction of Pt and pseudorapidity:

Background studies

Sources of background [pb] handles to fight them:

Reducible bb 5.108 associated production tt 500 extra jet activity*/Z+jets 104

b jet identification W + jets 105

Irreducible ZW 6.3 di-jet mass resolution ZZ 2.5

Discrimination of a Higgs signal is a hard task due to various sourcesof background:

Search for sensitive variables and combine them into a NNet..

dominates

Backgrounds, ttbar vs Higgs

..feed a nr of variablesinto the TMVA machinery..

ttbarHiggs 120 GeV

Neural Net output

ttbarHiggs

At pure MC level:

tt

Higgs

Some performance figures

• Seed finding and jet reconstruction efficiencies

Higgs_120 ttbar

L0 trigger 79.5% 55.3%

Lepton 58.6% 38.2%

b-seeds 13.6% 7.1%

2 Jets 10% 4.4%

• Preliminary background estimation in the mass window 65 – 155 GeV gives a S/sqrt(B) ~ 0.2

• The expectation is O(30) reconstructed signal events in one year data taking (2/fb) for a 120 GeV mass Higgs

Room for improvements by using a NNet cut. Under study.

Study of di-jet events

Estimated sensitivity

The dijet analysis of Minimum Bias events in LHCb can be a very effective tool for the of QCD predictions in a previously inaccessible range, even in the early runs of LHC.

Jets in the forward region also allows for detection of possible narrow resonances in models like:

Model: Randall-Sundrum graviton G* with k=0.01

…see G. Auriemma talk!

Conclusion/Outlook

Very challenging attempt to reconstruct a SM Higgs at LHCb in the mass region up to ~135 GeV. At the moment the largely dominant background is tt. Work is focused in trying to optimize background rejection.

Different approaches for jet reconstruction and b-tagging have been developed. Jet calibration is necessary to discriminate against irreducible sources of backgrounds.

The high performance of LHCb in detecting secondary vertices and identify b-jets can be applied to any exotic particle decaying into 2 b quarks giving rise to high momentum jets in the high rapidity region. (e.g. Hidden Valley, MSSM with R-parity violation models)