Low Mass Higgs to gamma gamma Run-II HLT

During this period of time I developed the trigger algorithm implmented in the HLT for the exploration of the low mass region in the gamma gamma channel.

L1 Studies NO PRESELECTION

The DoubleEG seed studied are, L1_DoubleEG2010Iso, L1_DoubleEG2210_v1 and L1_DoubleEG1510_v1 and the full seeding.
The turn on was computed with a "cocktail of signal samples" of 5000 events 60GeV+80GeV+125GeV each one

50ns L1_SingleIsoEG30er OR L1_SingleEG40 OR L1_DoubleEG_15_10 OR L1_DoubleEG_22_10 OR L1_DoubleEG_20_10_1LegIso

Gen Lead Gen Trail 2D

25ns L1_SingleIsoEG30er OR L1_SingleEG40 OR L1_DoubleEG_22_10 OR L1_DoubleEG_20_10_1LegIso

L1_DoubleEG2010Iso_v1

L1_DoubleEG2210_v1

More efficiency plots can be found here 60GeV+80GeV+125GeV,

L1 Studies ETA PRESELECTION

The DoubleEG seed studied are, L1_DoubleEG2010Iso, L1_DoubleEG2210_v1 and L1_DoubleEG1510_v1 and the full seeding.
The turn on was computed with a "cocktail of signal samples" of 5000 events 60GeV+80GeV+125GeV each one

50ns L1_SingleIsoEG30er OR L1_SingleEG40 OR L1_DoubleEG_15_10 OR L1_DoubleEG_22_10 OR L1_DoubleEG_20_10_1LegIso

Gen Lead Gen Trail 2D

25ns L1_SingleIsoEG30er OR L1_SingleEG40 OR L1_DoubleEG_22_10 OR L1_DoubleEG_20_10_1LegIso

L1_DoubleEG2210_v1

L1_DoubleEG2010Iso_v1

More efficiency plots can be found here 60GeV+80GeV+125GeV,

L1 Studies, BARREL ONLY

The DoubleEG seed studied are, L1_DoubleEG2010Iso, L1_DoubleEG2210_v1 and L1_DoubleEG1510_v1 and the full seeding.
The turn on was computed with a "cocktail of signal samples" of 5000 events 60GeV+80GeV+125GeV each one

50ns L1_SingleIsoEG30er OR L1_SingleEG40 OR L1_DoubleEG_15_10 OR L1_DoubleEG_22_10 OR L1_DoubleEG_20_10_1LegIso

Gen Lead Gen Trail 2D

25ns L1_SingleIsoEG30er OR L1_SingleEG40 OR L1_DoubleEG_22_10 OR L1_DoubleEG_20_10_1LegIso

L1_DoubleEG2210_v1

L1_DoubleEG2010Iso_v1

More efficiency plots can be found here 60GeV+80GeV+125GeV,

The paths used for the 125GeV Higgs discovery during run-I (2011-2012) were the following:

bit_194_HLT_Photon26_R9Id85_OR_CaloId10_Iso50_Photon18_R9Id85_OR_CaloId10_Iso50_Mass60_v5 OR
bit_195_HLT_Photon26_R9Id85_OR_CaloId10_Iso50_Photon18_R9Id85_OR_CaloId10_Iso50_Mass70_v1 OR
bit_205_HLT_Photon36_R9Id85_OR_CaloId10_Iso50_Photon22_R9Id85_OR_CaloId10_Iso50_v5

The paths that we have ready right now for the 125GeV re-discovery in the run-II (most probably to be updated):

Resonant path
HLT_Diphoton30_18_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_Mass95_v1

Control path (study the Z peak, for callibration studies)
HLT_Diphoton30_18_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_PixelSeed_Mass70_v1

a short explanation on how these paths work in this presentation , or here (by M. Planner):

Efficiency Study for the low mass HLT with 8TeV dataset.... Or what to expect

Due to the full mass-range coverage with the datasets available for the run-I analysis. An exhaustive study for the low mass higgs efficiency was performed. An emulation/reconstruction of the run-I paths was done offline and this is the validation of the method:

The Trigger was the reason for the low-mass boundary at m_gg=80GeV for the H->gg low mass analysis during run-I. We are planing to extend this down-to 50GeV. During run-II the situation could be even worse, due to the increase of luminosity the thresholds for Double-Photon Triggers are higher and a new path for the low mass is necessary.
Below you can see the efficiency for the low mass region with the FIRST proposed HLT paths for the Hgg analysis (Sept-2014) (meant to confirm the 125GeV boson observed in 2012)

The efficiency in the low mass region for run-II would be even lower than the efficiency that we have got for run-I.

Strategies for the low mass higgs->gg analysis

Due to the similarities in the High Level Trigger for the 125GeV rediscovery paths and the low-mass paths we are planning to follow as much as possible to the "by default" fully optimized rediscovery paths. This would bring several advantages also when the analysis would be performed. Several changes have to be done:

Enhancing the efficiency in the low mass region

1. The natural changes to accomplish this goal is to reduce the p_T thresholds, as low as the L1-Trigger thresholds Permits it.
2. The invariant mass of the diphoton state should be also reduced (initial goal is down to 50GeV)

Controlling the rate

To control the rate is necessary to sacrifice some signal-efficiency, trying to keep the sensitivity of the analysis. We have proposed the following strategies:

Changing the logic OR -> AND

The new rediscovery paths select super-clusters with high-R9 (>0.85) with higher priority than the ones with lower R9 (0.65Adding and Eta Restriction(just the barrel) It is well known that the Higgs to gamma gamma analysis was more sensitive in the barrel region than in the endcap region. Another possibility to control the rate is to ignore all the photons in the endcap and to keep the good photons in the barrel instead.

Isolation: Pixel Veto (Pixel Detector), Solid Cone(Tracker)

Requiring isolation at trigger level will reduce the rate. This isolation could be done by:
1. Implementing a Pixel Veto (following the Z-control path).
2. Changing the Tracker isolation for a full solid cone

Results

/users/carrillo/hgg_740/V14

Global information about these results
- release CMSSW_7_4_0
- global-tag run2_mc_GRun
- hltIntegration_command: hltIntegrationTests /users/carrillo/hgg_740/V14 -s /dev/CMSSW_7_4_0/HLT -i /store/user/carrillo/13TeV_60GeV/8C589E9C-694A-E411-99B6-00266CF9B59C.root --mc -x "--l1-emulator stage1,gt"

Following the Available version from the rediscovery paths under /users/mplaner/2015April7_hgg_all_paths/V12 :
HLT_Diphoton30_18_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_Mass95_v1
HLT_Diphoton30_18_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_DoublePixelSeedMatch_Mass70_v1

We have implemented the follwing 4 paths following our strategies (/users/carrillo/hgg_740/V14):
HLT_Diphoton28EB_14EB_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_DoublePixelSeedVeto_Mass50_v1
HLT_Diphoton30_18_R9Id85_AND_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_DoublePV_v1
HLT_Diphoton30_18_R9Id85_AND_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_PV_v1
HLT_Diphoton30_18_R9Id85_AND_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_Solid_Mass30_v1

Extra control paths have been included but just for control of our developments.
HLT_Diphoton28EB_14EB_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_Mass50_v1
HLT_Diphoton28_14_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_Mass50_Eta_1p5_v1
HLT_Diphoton28_14_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_DoublePixelSeedVeto_Mass50_v1
HLT_Diphoton30_18_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_DoublePixelSeedVeto_Mass50_v1

Signal Efficiency

The preliminary hltintegration test for /users/carrillo/hgg_740/V14 with a 60GeV signal sample, gave as result:


And this is the efficiency as a function of higgs-gen-eta, there is not normalization for this efficiency is just the amount of events passing the Trigger selection ussing a 60GeV sample.
HLT_Diphoton30_18_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_Mass95_v1

HLT_Diphoton28EB_14EB_R9Id85_OR_Iso50T80LCaloId24b40e_AND_HE10P0_R9Id50b80e_DoublePixelSeedVeto_Mass50_v1

Rates

The analyzed datasets and the multiplicative constants to convert passing events ratio to rate in Hz are listed below:

EM Enr:
/QCD_Pt-20to30_EMEnriched_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (2341.724112)
/QCD_Pt-30to80_EMEnriched_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (158003.846)
/QCD_Pt-80to170_EMEnriched_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (7629.76858)
QCD
/QCD_Pt-30to50_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (2143654.1)
/QCD_Pt-50to80_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (273014.28)
/QCD_Pt-80to120_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (35516.55)
/QCD_Pt-120to170_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (5768.9604)
/QCD_Pt-170to300_Tune4C_13TeV_pythia8/Fall13dr-castor_tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (1684.2)
DYtoee:
/DYToEE_Tune4C_13TeV-pythia8/Fall13dr-tsg_PU40bx25_POSTLS162_V2-v1/GEN-SIM-RAW (97.44)

Jobs have been submited but it is taking forever with CRAB...

And these are the rate results.

After investigating the high rates, the conclusion was that by default the new release CMSSW_7_4_0 does not include by defautl the adequate Global Tag for these samples. The study has to be repeated. The Global tag has to be still tweaked by hand, and the auto-version is not the most suitable one for the rate estimation following these instructions https://twiki.cern.ch/twiki/bin/view/CMSPublic/SWGuideGlobalHLT#Running_the_HLT_with_CMSSW_7_AN1

Results 2: (Removing HCAL Isolation) /users/carrillo/hgg_740/V26 and MCRUN2_72_V3A as Global Tag

Global information about these results
- release CMSSW_7_4_0
- global-tag MCRUN2_72_V3A
- hltIntegration_command: hltIntegrationTests /users/carrillo/hgg_740/V26 -s /dev/CMSSW_7_4_0/HLT -i /store/user/carrillo/13TeV_60GeV/8C589E9C-694A-E411-99B6-00266CF9B59C.root --mc -x "--l1-emulator stage1,gt"

Changes w.r.t previous version:
For this version of confDB the L1-HLT paths for the study of the L1 curves has been included
For the bit overlap studies the electron paths have been also included
The integration test can be seen here /afs/cern.ch/user/c/carrillo/public/for_steam/users_carrillo_hgg_740_V26/, and the efficiency results for the integration test:

TOTAL rates

TOTAL rate.

UNIQUE rates w.r.t eletrons


UNIQUE rates w.r.t resonant


UNIQUE rates w.r.t (eletrons OR resonant)



Results 3: tighter sieie and new mass cuts

A new update for the rediscovery paths has been done, the sieie cuts were changed from: EB(EE)=0.024(0.040) to EB(EE)=0.015(0.035), the implementation of the new version of the paths are, here /users/carrillo/hgg_740/V29
- apply the new sieie 0.015/0.035
- check the effect of cutting the mass at >55,60
The integration test was successfull and can be found under /afs/cern.ch/user/c/carrillo/public/for_steam/users_carrillo_hgg_740_V29/:

TOTAL rates

these are the latest (23/04/2015) values for TOTAL rate.

UNIQUE rates w.r.t eletrons


UNIQUE rates w.r.t resonant


UNIQUE rates w.r.t (eletrons OR resonant)


Results 4: Last Synchronization with rediscovery path /users/carrillo/hgg_741HLT1/V15

- Release: CMSSW_7_4_1 - GlobalTag: MCRUN2_72_V3A - ConfDB: /users/carrillo/hgg_741HLT1/V15 (with manueal tweak) - New values for isolation and HoE in this version of the rediscovery paths
- AND+SinglePV path has been dropped
- All paths migrated to the pt-rediscovery cuts, the minimum pt-cuts before bummping in L1
- New Mass cuts have been explored for all the strategies

The integration test was NOT successfull,
Illegal parameter found in configuration. The parameter is named: 'l1CenJetsTag'
This input was commented by hand in the config file after the failed attempt of the integration test. Onces this is done cmsRun works.

Efficiency

The signal efficiency for the 55 mass cuts can be seen here for 60GeV+80GeV+125GeV and here for 60GeV

Two low-mass strategies have been studied:
lowmass = low-mass paths (in yellow) EB OR AND-PV OR AND-Solid
lowmass2 = EB OR AND-PV

cross-bit efficiency for 60GeV signal sample



TOTAL rates

cross-bit rates



rates table


UNIQUE rates w.r.t eletrons


UNIQUE rates w.r.t resonant


UNIQUE rates w.r.t (eletrons OR resonant)



Conclusions:
- The efficiency for the low mass region is lower after the changes.
- The rates have been cut by a factor 3 with these new changes.
- The rate seems to be the same for lowmass and lowmass2.