status and prospects of the h → γγ analysis

Status and Prospects of the H→γγ Analysis

Jim Branson - Marco Pieri - Sean Simon

UCSD Meeting

March 11th 2008

Updated for March 18th

11-Mar-08 Marco Pieri 2

Introduction

H→ γγ analysis will start to be more important for Int L >~ 1 fb-1

UCSD has played a major role in the PTDR studies and is expected to play a major role in the next years

Other people/groups contributing are: Caltech, Lyon, Notre Dame, Rome, Saclay, UC Riverside, UCSD

For 2008 not much to be expected in H→ γγ channel In addition the ECAL calibration will not be optimal Related analyses: γ+jet, γγ from SM (except Higgs) – Should

collaborate more with people working on them Since about 1 month started revisiting the analysis framework to have it

more flexible and common with other analyses For now we ran over small MC samples: ~100k GamJet + ~100k Higgs +

~ 100k QCD + photonsJets + ~50k Dy All what shown here very preliminary News: In CMSSW 2_0_0 photons a 5 GeV Et cut an H/E cut at 0.2 is

proposed to be applied for reconstructing photons


forward jets

Photons from Higgs decay

qqH → qqγγ MH = 120 GeV

H→ γγ Signal

SIGNAL: two isolated photons with large Et

Gluon-gluon fusion WW and ZZ fusion (Weak Boson Fusion) WH, ZH, ttH (additional leptons and MET) Total σ x BR ~95 fb for MH = 110-130 GeV Very good mass resolution

H → γγ MH = 115 GeV Jets from qq are at

high rapidity and large Δη


BACKGROUND ‘irreducible’ backgrounds, two real photons

gg→ γγ (box diagram) qq→ γγ (born diagram) pp→ γ+jets (2 prompt γ)

‘reducible’ backgrounds, at least one fake photons or electrons pp→ γ+jets (1 prompt γ + 1 fake γ) pp→ jets (2 fake γ) pp→ ee (Drell Yan) when electrons are mis-identified as photons

Handles for Irredicible BG – Kinematics Handles for Reducible BG – Until now only Isolation

Should add photon identification (converted) and π0 rejection

Background to H→ γγ

Process Pthat (GeV) Cross section (pb) Events/1 fb-1

pp→γγ (born) >25 82 82K

pp→γγ (box) >25 82 82K

pp→ γ+jets >30 90x104 90M

pp→jets >25 1x108 1x1011

Drell Yan ee - 4x103 4M


Cross section and K-factors

Signal cross sections and BR used for the PTDR (NLO M. Spira)

K-factors for the background used for the PTDR (to be re-evaluated if needed)

pp→γγ (born) 1.5

pp→γγ (box) 1.2

pp→ γ+jets (2 prompt) 1.72

pp→γ+ jets (1 prompt+ 1 fake) 1

pp→jets 1

M=115 GeV M=120 GeV M=130 GeV M=140 GeV M=150 GeV

σ (gg fusion)(pb) 39.2 36.4 31.6 27.7 24.5

σ (IVB fusion) (pb) 4.7 4.5 4.1 3.8 3.6

σ (HW, HZ, Hqq) (pb) 3.8 3.3 2.6 2.1 1.7

Total (pb) 47.6 44.2 38.3 33.6 29.7

BR (H→ γγ) 2.08x10-3 2.21x10-3 2.24x10-3 1.95x10-3 1.40x10-3

Inclusive σ x BR (fb) 99.3 97.5 86.0 65.5 41.5


PTDR Mass Spectrum of Selected Events

All plots are normalized to an integrated luminosity of 1 fb-1 and the signal is scaled by a factor 10

Fraction of signal is very small (signal/background ~0.1) Use of background MC can be avoided when we will have data Data + signal MC can be used for optimizing cuts, training NN and

precise BG estimation


CSA07 MC Samples

Requests at: https://twiki.cern.ch/twiki//bin/view/CMS/HiggsWGMCRequestsForHiggsToGamGam

Higgs Signal (Pythia) masses between 60 and 160 GeV (at Fnal, Cern, Lyon) gluon-gluon fusion, IVB fusion , WH, ZH, ttH

Background (and even Signal) started to came very late in 2007 at it is not yet complete + Two samples were forgotten and resubmitted at the end of January

GamJet, Twophoton_Box, DY - OK Twophoton_Born 450 K events Lyon - 1/2 of requested Jets_Pt50up 1.4 M events Cern - 1/6 of requested It would probably be good if the production was finished

HiggsTo2Gamma Skims of the soups available, we should start running on them

process pythia lev cuts gen level cuts gen sigma

sim sigma

gen level cuts reduction factor

# of gen evts

# of sim evts

Int L (fb-1)

gg->gamgam

(box) pthat>25 GeV none 36 pb 36 pb 1 1M 1M ~28

qq->gamgam (born)

pthat>25 GeV none 45 pb 45 pb 1 1M 1M ~22

pp->gam +jet pthat>25 GeV Special cuts (~sel B' in CMS IN 2005/018)

90 nb 0.6 nb ~150 300M 2M ~3.3

pp->jets pthat>50 GeV Special cuts (~sel C' CMS IN 2005/018)

24 ub 4.8 nb ~5000 50G 10M ~2.1


Important Points – Reconstruction Level

Trigger and Skims L1 Trigger HLT Skims

Photon isolation

Primary Vertex estimation

Energy Measurement Ecal crystal calibration SuperCluster calibration Photon energy scale Energy Resolution and Error (maybe optional, was done

before)

Photon conversion identification and π0 rejection


Electromagnetic trigger towers are classified in two categories depending on the energy deposition in the calorimeter trigger towers: non-isolated, isolated.

Nominal Low Lumi (2x1033 cm-2s-1) Single isolated

Et>23 GeV Double isolated

Et>12 GeV Double non-isolated

Et>19 GeV At startup thresholds lower

Total electron+photon Level-1 trigger rate ~ 4 kHz Level-1 trigger efficiency for H→ γγ larger than 99% Perhaps could still optimize the threshold at which all Isolation L1 cuts

are removed

Level-1 Trigger


H → γγ signal has two isolated photons Dominant background from di-jets and γ+jet has at least one candidate

from jet fragmentation that is not well isolated

We keep early conversions in the double stream HLT trigger efficiency 88% - almost 100% for events selected in the

analysis Trigger is relatively easy for H→ γγ because of high Et photons Total rate for photons after HLT ~5 Hz Need to make some improvements, particularly for pre-scaled triggers,

try to add the double from single L1 HTL paths (also for electrons?)

HLT for Photons

PTDR HLT photon selectionNominal Low Lumi (2x1033 cm-2s-1)


Skim for H→ γγ

I made a very simple skim selection last summer For now very simple:

Double Photon HLT .OR. Single Photon HLT with an additional SC – to easily study trigger efficiency

Will hopefully keep it simple forever Skimmed datasets not too large ~1-3 Hz for photons

RECO format planned to be used for now PDPhoton Skim higgsTo2Gamma files are at UCSD now We should run on them

No veto for electrons – Stream can also be useful to study electrons


Reducible backrounds (π0’s and mis-identified jets) have other particles near at least one photon candidate

We are in process of repeating and improving the study we carried out for the PTDR

Most of discriminating variables are built by summing up the Et or Pt of calorimeter deposits or tracks within a cone

ΔR = (Δη2+ Δφ2)

To study the performance of isolation variables we use individual photon candidates match or not within ΔR < 0.2 to a prompt generator level photon

Signal is: 120 GeV H→γγ gg-fusion reconstructed photon with Et>30 GeV matched with a generated photon within ΔR<0.2, background is: a super-cluster with Et>30 GeV NOT matched with a generated photon

Low statistics for now, cannot really look at correlations Trigger (L1 and HLT) not included

Photon Isolation

ΔR


Photon Isolation – Barrel – QCD pthat 80 – 120 GeV

Two possible views, first better for high purity, second better for high efficiency

Trigger not included


Photon Isolation – Barrel – QCD pthat 50 – 80 GeV



Photon Isolation – Endcaps – QCD pthat 80 – 120 GeV



Photon Isolation – Endcaps – QCD pthat 50 – 80 GeV



Photon Isolation II

For low pthat, isolation much less effective Should study it better – need more statistics at low pthat Note that pre-selected QCD events below 50 GeV pthat not simulated

Run on Gumbo skims – already at UCSD

Some more checks must still be carried out Study the correlation between isolation variables and specify

benchmark selections for photons

For the PTDR analysis we used a Neural Network with 2, 3 or 5 of following inputs: ΔR of the 1st track with Pt>1.5 GeV/c Sum ECAL Et within ΔR<0.3 The shower shape variable R9

Sum HCAL Et within ΔR<0.35 Sum tracks Et within ΔR<0.2

We did not use kinematical information, easy to combine these variables with reconstructed mass and photons Et in an optimized H→γγ analysis

Repeat the study in the near future


Primary Vertex Determination

New longitudinal interaction spread σ~7.5 cm (was 5 cm) Vertex estimated from the underlying event and recoiling jet In PTDR analysis the efficiency of determining the right vertex

was ~83% for H→ γγ events after selection Efficiency for the different types of background is similar and

basically irrelevant

First check of usage of identified converted photons – very preliminary

Currently we have datasets with no pileup Efficiency of reconstructing the right primary vertex ~98% on

all generated H→ γγ events

Must be compared with minimum bias events


Primary Vertex Determination II

Process Eff (%)

H→γγ (gg fusion) 82

H→γγ (IVB fusion) 89

pp→γγ (born) 71

pp→γγ (box) 72

pp→γ+jet (2 prompt) 78

pp→γ+jet (1 prompt + 1fake) 86

pp→jets 90

PTDR low luminosity Efficiency of determining the primary vertex within 5 mm from the true one

PTDR analysis

Use old z beam spot 100 pb-1 calibration

CMSSW_1_6_7CSA07 MC


Primary Vertex Determination III

Generator level plots for different track pt cuts are provided in the Extra slides


Primary Vertex From Photon Conversions

At least 1 convpho identified

1 or 2 tracksAt least 1 selected convpho identified

Nearest convpho (or track) used

(Cheat)

All 51.1% 17.9% 51.1%

Vtx within 1 cm 20.1% 13.7% 24.3%

Vtx within 2 mm 12.3% 8.7% 15.7%

Selected converted photons: use only thosewith Mass <2 GeV, |z1-z2|<2cm

Choose Converted Photon with best e/p H→ γγ events passing PTDR selection

All reconstructed converted photons, 1 or 2 tracksBest e/p

Selected reconstructed converted photons, with 2 tracksBest e/p

CMSSW_1_6_7CSA07 MC


Primary Vertex Studies

Wider longitudinal beam spot will: Worsen the Mass resolution for events with the wrong primary

vertex or no vertex Make easier the discrimination between different vertices using

tracks from converted photons Even with no pileup can already superimpose Higgs events and

minimum bias events and carry out all studies When we want to optimize primary vertex finding we can also

use the direction of the total tracks transverse momentum that should be opposite to the Higgs pt


PTDR Selection for Cut-Based Inclusive Analysis

Photon selection: photon candidates are reconstructed using the hybrid clustering algorithm in the barrel and the island clustering algorithm in the endcaps ET1, ET2 > 40, 35 GeV |η|<2.5 Both photon candidates should match L1 isolated triggers with

ET > 12 GeV within ΔR < 0.5 Track isolation

No tracks with pt>1.5 GeV present within ΔR<0.3 around the direction of the photon candidate

Calorimeter isolation Sum of Et of the ECAL basic clusters within 0.06<ΔR<0.35 around

the direction of the photon candidate <6 GeV in barrel, <3 GeV in endcaps

Sum of Et of the HCAL towers within ΔR<0.3 around the direction of the photon candidate<6 GeV(5 GeV) in barrel (endcaps)

If one of the candidate has |eta|>1.4442 the other has to satisfy also: Sum of Et of the ECAL<3, Sum of Et of the HCAL<6 GeV

L1 + HLT inefficiency negligible after selection


Higgs Mass Resolution

ECAL calibration for 100 pb-1

Peak resolution all selected events σfit 1.45 GeV, σfit 1.75 GeV Much worse than with ideal calibration, especially in endcaps

Barrel Endcaps

R9>0.93 R9<0.93

CMSSW_1_6_7CSA07 MC


Higgs Photons Efficiency Plots

Top plots photon finding efficiency Bottom plots photon isolation efficiency (PTDR cuts)


Higgs Mass – Primary Vertex Effect

Barrel

Endcaps

R9>0.93

R9<0.93


γ+jet Background

Plots are normalized to an integrated luminosity of 1 fb-1 and the signal is scaled by a factor 10

BG seems similar to PTDR

Barrel Endcaps


Fake Photons from Jets

We ran on very low BG statistics, did not yet estimate the two photon BG

Start studying the single photon efficiency and fake rate Will compare between QCD and γ + jets

Should evaluate the needs in terms of BG rejection and consequently optimize isolation


QCD Fake Photon Rate – 1 pb-1


Fake Photon Rate Fake Photon Rate after isolation


Photon+jet Fake Photon Rate – 1 pb-1

Trigger not included??? Should check

Fake Photon Rate Fake Photon Rate after isolation


Fake Photon Isolation Efficiency



One Photon Rate – 1 pb-1 Trigger not included


ECAL Calibration and Photon Energy Scale

Crystal Intercalibration Electrons from W→eν decays will be used Also π0 and/or η will be used

In CMSSW 2_0_0 there will only be SC corrections, no photon nor electron corrections anymore

Photon energy scale being studied from μμγ by Lyon, Florida State University and Kansas State University

μμγ events can also be used for efficiency studies


ECAL Calibration and Photon Energy Scale

Crystal Intercalibration Electrons from W→eν decays Also π0 (and perhaps η) will be used See for example presentation by V. Litvin at:

http://indico.cern.ch/conferenceDisplay.py?confId=29156 In CMSSW 2_0_0 there should only be new SC corrections, no

photon nor electron corrections anymore unless it will be shown that they are needed See for example presentation by Y. Maravin in:

http://indico.cern.ch/conferenceDisplay.py?confId=27059 Basically ready for Barrel, in progress for endcaps

Photon energy scale being studied from Z->μμγ (and Z->eeγ) See for example talk by S. Gascon at:

http://indico.cern.ch/conferenceDisplay.py?confId=27555


Photon Conversions

Most of the work carried out by Nancy Marinelli and Notre Dame University

They are currently trying to choose the best candidate Some changes Photon Objects in CMSSW 2_0_0 In my opinion much more word needed in order to use them for

photon identification


Recovery of early conversions currently removed by track isolation Probably difficult

Converted Photons and π0 rejection

Barrel Endcaps


π0 Rejection

Converted photons can also be used for π0 rejection Start looking at the performance of the π0 rejection variables

that are provided in CMSSW since version 1_6_7 See for example presentation by A. Kyriakis in:

http://indico.cern.ch/conferenceDisplay.py?confId=20797

Start looking at the π0 rejection NN variables provided in CMSSW


Important points – Analysis Level

Simulation – Signal an Background

Real analysis on data and related channels

Optimization of the Analysis


Simulation

Background simulation Generator level preselection for fake photons has been studied and

used for CSA07 MC production The Lyon group is working with DiPhox authors to have a full NLO

irreducible BG simulation Anyway, be ready to carry out the analysis using the BG from data,

enough events from sidebands

Signal Simulation (common with other Higgs channels) We should get NLO/NNLO calculations in order to exploit at best the

signal topology: HNNLO for gluon fusion, M. Grazzini et al. VBFNLO for IVB fusion, D. Zeppenfeld et al.

Think about the requests for the next MC production with CMSSW Version 2


Real analysis

Real analysis – take as much as possible from data Efficiency from data (Z->ee , Z->eeγ, Z->μμγ) Fake rate from data (important even if not crucial for H→ γγ) Use data (sidebands) to optimize the selection and to estimate the

BG properties Study of systematic errors

Only sources of tagged high Et photons Z->μμγ Z->eeγ

Related Analyses (to be studied since the beginning) γ+jet (Fake rate needed) γγ (Fake rate needed)


Z + Z + , , ZZA clean source of photons, A clean source of photons,

can determine, with real data: can determine, with real data:

• Efficiency of photon triggers Efficiency of photon triggers

• Determination of photon energy scale Determination of photon energy scale

• Determination of photon id efficiencyDetermination of photon id efficiency

• Determination of photon energy corrections Determination of photon energy corrections

Z->μμγ

See for example talk by S. Gascon at:http://indico.cern.ch/conferenceDisplay.py?confId=27555

ALPGEN


Z->eeγ

See presentation by Marat Gataullin at:http://indico.cern.ch/conferenceDisplay.py?confId=29791

Efficiency of the Photon ID cuts is 88%, but the background isalmost gone, 96% purity in the window 85 GeV < M(eeγ) < 95 GeV.Total yield: 4.6K events per 1fb-1


Optimized Analysis

Coherently exploit the different production modes (signatures 1l, 2l, MET, VBF)

See if possible avoid using MC background also for these Add additional variables that were not used in the PTDR

because of the poor description of the LO generators that were used

Carry out optimized multivariate/multicategorized analysis


Effect of Systematic Errors - PTDR

Input for CL calculation is: Background expectation from fit to the data (sidebands) Signal expectation from MCOrigin of systematic errors Error on the BG estimation (statistical from fit of sidebands +

uncertainty of the form of the fitted function) Error on the signal (theoretical σxBR, integrated luminosity,

detector + selection efficiency)Effect of systematic errors Systematic errors on the signal do not change the expected

discovery CL Systematic error on the signal makes exclusion more difficult Systematic error on the BG makes exclusion and discovery

more difficult


Main Systematic Errors - PTDR

SIGNAL Theoretical error on cross

section times BR (~15%) Integrated luminosity (~5%) Higgs Qt distribution – effect

to be evaluated Selection efficiency (~10%)

Can assume a total of 20% (anyway not important in case of discovery)

Nevertheless systematic errors on the signal may cause the analysis to be less optimized

BACKGROUND Statistical error on the fit of

the sidebands (~0.3% for ~20 fb-1)

Systematic error on the shape of the fitted function (~0.3%)

No other errors when data available


Outlook

We started revising the H→ γγ analysis framework so that it can also be used for all other analyses

We only ran over small samples for now We can now run on larger samples

We are also trying to organize the CMS-wide effort in order not to be alone in the analysis as it was for the PTDR

Getting other groups to contribute to the H→ γγ analysis

NEXT STEPS Continue the studies presented here Include HLT (and re-optimize it) in our analysis Need to re-optimize the basic selection for the cut-based analysis Study more converted photons and π0 rejection to see if they can be

used in the analysis Get NLO/NNLO description of the signal and rescale Pythia – Also check

ALPGEN, MC@NLO Look at all issues of the real analysis on data Look again at the optimization of the analysis


End of the talk

End of the talk


EXTRA

EXTRA


Barrel – pthat 80 – 120 GeVTrigger not included


Track Isolation BarrelTrigger not included


Track Isolation EndcapsTrigger not included


Ecal Isolation BarrelTrigger not included


Ecal Isolation EndcapsTrigger not included


Hcal Isolation BarrelTrigger not included


Hcal Isolation EndcapsTrigger not included


Generator Level, charged pt>1.5 GeV |eta|<2.5


Generator Level, charged pt>0.3 GeV |eta|<2.5


Just to remember – IVB fusion

Sasha Nikitenko:http://indico.cern.ch/getFile.py/access?contribId=24&sessionId=4&resId=0&materialId=slides&confId=30337http://indico.cern.ch/getFile.py/access?contribId=24&sessionId=4&resId=1&materialId=slides&confId=30337

status and prospects of the h → γγ analysis

Documents

fake pp jets

datadata signal mc

diagrampp jets

electronspp jets

pp box1

fake photons

fraction of signal

small mc samples