Skip to content

Releases

As far as we are aware, there is no dependence on the release used to analyse the data, only the release and conditions used to reconstruct that data. Running VID and scale & smearing counts as analysis here. As an example the 2016 legacy scale factors are valid for working in 80X, 94X and 102X releases. Further more, the reminiAODs do not alter the quantities of e/gamma objects, they only add additional information. Therefore there is no difference between running VID yourself on V1 MiniAOD or reading it directly in V2 miniAOD. So rest assured, the scale factors for a given data (2018, 2017, 2016 Legacy) are the same no matter what you do. Feel free to ask us for further clarification on hn-cms-egamma@cern.ch.

E/gamma DataFormat

We have now use the E/gamma MiniAODV2 format which embeds all the energy corrections and the IDs inside the pat::Electron/pat::Photon. Please see EgammaMiniAODV2 for more information. We provide tools to update the format with new IDs and energy corrections, please see EgammaPostRecoRecipes. This is probably the easier way to use E/gamma objects.

E/gamma Trigger

L1 Prefiring

The darkening of the ECAL crystals causes the light reaching the photodetectors to become increasingly delayed. Due to the interaction of this effect with the trigger electronics, it is possible for a trigger tower to be assigned to the previous bunch crossing, with the likelihood of this happening increasing with increasing opacity of the crystal. This effect is was not corrected for and in 2016 and 2017, it is possible for L1 ECAL trigger primatives corresponding to an object to end up in the previous bunch crossing. As there is unlikely to be anything in this event cababile of passing the HLT, the event is rejected. Even worse, due to CMS trigger rules, the L1 can not fire on two successive bunch crossings so even if there are other objects to trigger on at L1 in the correct bunch crossing, the event is rejected. This manifests itself as an event level trigger inefficiency and is not included in tag\&probe measurements.

There exists a preliminary recipe by Laurent Thomas and rest of the pre-firing team: L1ECALPrefiringWeightRecipe.

Please be aware that the recipe is only guaranteed to work when there is only one prefirable object (ie |η|>2.0). The recipe may work with >=1 prefirable objects but this is strongly topology dependent. In this case you should definately check with the experts, do not blindly assume it works for this case.

As this recipe is still preliminary, we & the l1 pre-firing group would greatly appreciate feedback.

Selecting a Trigger Path

The twiki EgHLTRunIISummary gives a complete listing of all E/gamma triggers (including cross triggers) for 2016, 2017 and 2018. It also gives some basic suggestions for the typical uses cases for electrons. The detailed selection of each trigger is summarised in EgHLTPathDetails (its a large twiki, can take a while to load).

HLT Zvtx Scale Factor

  • there is an allowed Z region for an electron to be in the HLT which is obtained by the online beamspot. There were problems in the begining of 2017, which means that the z region was too small and efficiency suffered. This can not be determined with Z->ee tag & probe as the tag & probe have the same zvtx. Care must be taken with e-mu tag and probe as this will include it as the muon triggers were not as effected
  • the MC efficiency is one so the efficiency of the online Z region cut in the HLT and the scale factor for MC are identical
  • events selected using electron triggers which use pixel matching require an additional scale factor. As it is event level, it should be applied only once per event, no matter how many electrons were required to be triggered
  • this only needs to be applied in 2017, 2016 & 2018 are fine
  • these numbers are considered final by E/gamma (and are unchanged since last year)
    • the 2018 number measured to be > 0.99976 and is taken to be 1.0 as a 0.02% inefficiency dwarfed by other efficiency systematics

| Era | 2016 | 2017B | 2017C | 2017DEF | 2017BCDEF | 2018 | | SF | 1.0±0.0 | 0.934±0.005 | 0.992±0.001 | 1.000 | 0.991±0.001 | 1.0±0.0 |

E/gamma RECO

Photon Scale Factors

The scale factor to reconstruct a supercluster with H/E\<0.5 is assumed to be 100%

Electron Scale Factors

This is the scale factor to get a GsfTrack once you have a supercluster with H/E\<0.5.

E/gamma IDs

E/gamma maintains several IDs, both cut based and MVA. In terms of cut based, there is also the HEEP (high energy) ID which aimed to be a simple straighforward ID that is safe for high energy electrons. The HEEP ID for historical reasons operates on a different versioning than other cut based IDs. All these IDs are accessible as part of the pat::Electron/pat::Photon however depending on the miniAOD version you run over, you may need to run EgammaPostReco tools to embed it first. It is important to note due to the way our ID framework functions, if you find an ID in the pat::Electron/pat::Photon with the requested name, that ID will be the correct. So if you find the ID you are looking for, you should not worry that its an out of date version.

More information can be found at EgammaIDRecipesRun2

Example accessing IDs

We recommend using the E/gamma miniAOD V2 format which embeds all the IDs in the pat::Electron,pat::Photon which means if you know the name of the ID, you can access it straightforwardly like:

pat::Electron::electronID("cutBasedElectronID-Fall17-94X-V1-tight")==true //passes Fall17 V1 tight
pat::Photon::photonID("cutBasedPhotonID-Fall17-94X-V1-tight")==true //passes Fall17 V1 tight

We try to have our IDs in a release so no PR is necessary. However for the latest and greatest ID, there will be a period of time where the ID has to be merged in via a PR. For each ID, the first CMSSW release where the ID is included is indidicated as "release availiblity". The miniAOD which has this ID embedded is also noted as "miniAOD availiblity".

Electrons IDs:

Spring16 (mva) & Summer16 (cut based)

Fall17v1

Fall17v2

HEEPV7.0

  • summary: a simple robust ID designed to be safe for high electrons. The Et evolution of this ID must be well described in the MC therefore this ID is designed so its scale factor is flat vs Et. As a result the HEEP differs in that it provides just a single number for the barrel and a single number for the endcap.
  • release availibity: CMSSW_9_4_0, CMSSW_10_2_0
  • miniAOD availiblity: 2016 v3 miniAOD, 2017 v2 miniAOD, 2018 v1 miniAOD
  • recommended for: 2016 (prompt and legacy), 2017
  • scale factors:

HEEPv7.0-2018Prompt

There was a retune of HEEPv7.0 ID for 2018 due to HCAL data/MC disagreements. The cuts in EB remain same as HEEPv7.0, but there were some changes in cuts in EE. The H/E cut and EM+Had_depth1 isolation cut changed in 2018 for EE only.

For more details, see

1) Decicated study by W. Fang etc al. https://indico.cern.ch/event/787315/contributions/3434898/attachments/1847223/3031172/HEEP_2019_0517_v3.pdf . In slide 16, the choice 2 was considered the final choice, and was used in Z'->EE search.

2) Z'->EE approval talk https://indico.cern.ch/event/831669/contributions/3485543/attachments/1871797/3084930/ApprovalSlides_EE_v3.pdf

This ID is not centrally available in https://github.com/cms-sw/cmssw/tree/CMSSW_10_2_X/RecoEgamma/ElectronIdentification/python/Identification , which means that users need to manually apply this ID. Users can make use of individual cuts as described here https://twiki.cern.ch/twiki/bin/view/CMS/CutBasedElectronIdentificationRun2#Applying_Individual_Cuts_of_a_Se , so that the other cuts that did not change in 2018 can still be used from centrally available VID.

  • For 2018 rereco (Autumn 18) the SF are the following, 0.969 +/- 0.000 (stat) (EB), and 0.984 +/- 0.001 (stat) (EE). The systematic uncertainty is same as 2017 prompt, as indicated in slide 15 of Z'->EE approval presentation, linked above. So, the following should be used
  • uncertainty: EB E\<sub style="background-color: transparent;">T \< 90 GeV: 1% else min(1+(E\<sub style="background-color: transparent;">T-90)*0.0022)%,3%)
  • uncertainty: EE E\<sub style="background-color: transparent;">T \< 90 GeV: 2% else min(1+(E\<sub style="background-color: transparent;">T-90)*0.0143)%,5%)
  • For more details see here https://twiki.cern.ch/twiki/bin/view/CMS/HEEPElectronIdentificationRun2#Scale_Factor. As always, HEEP ID SF are just two numbers, one for EB and one for EE.

Photon IDs:

If you are using any of our photon IDs for high pT (>200 GeV) photons, then read this twiki for detailed EGM recommendations: https://twiki.cern.ch/twiki/bin/view/CMS/EGMPhotonIDHighPtPhotons

Spring16

Fall17v1

Fall17v2

E/gamma Energy Corrections

Electron and Photon Energy Regression Corrections

The electron and photon energy corrections are applied by default to the object. 2016Legacy & Prompt uses the V1 or "74X" regression while 2017 and 2018 use a modified version of the V2 or "80X" regression. This modification is that the high pt ecal training is only used for saturated electrons and photons.

Residual Scale & Smearing Corrections

The data / MC energy rescale and resolution have a small residual miss match. Therefore we scale the data to the MC and the smear the MC to match the data resolution. This is known as scale & smearing corrections. The scale and smear corrected energy can be accessed inside the miniAOD object via pat::Electron::userFloat('ecalTrkEnergyPostCorr"), pat::Photon::userFloat('ecalEnergyPostCorr'). Some analyses use the ECAL only energy of electrons and therefore use pat::Electron::userFloat('ecalEnergyPostCorr'). Currently, these final corrections are only availible for 2016 and 2017, although preliminary ones are now availible 2018 Preliminary Energy Correction

To access the scale and smearings please see EgammaMiniAODV2#Energy_Scale_and_Smearing

Recommendations on Combining Systematics between years

What follows are some simple recommendations on correlating the systematics between the years which we feel are a resonable starting point. The method for deciding the correlation depends on whether the quantity is measured in its kinematic phase space or extrapolated its kinematic phase space. The systematic uncertainty of a measured value is dominated by biases associated with the measurement method. As long as the method remains the same, we can reasonably assume that the method bias will be the same each year. The systematic uncertainty of an extrapolated value is dominated by the modelling of that extrapolation by the Monte Carlo simulation and can change between years if new cuts, new variables or new detectors are introduced. We absolutely encourage feedback here and will update the recommendations based on this.

  • trigger efficiency:
    • uncertainties are negligible w.r.t to reco & ID unless there are bugs
  • RECO efficiency:
    • uncertainties are correlated between the years as they are measured with the same method
    • due to the larger backgrounds, we assume the RECO uncertainties are uncorrelated w.r.t to the ID uncertainties
  • ID efficiency
    • 20 to 200 GeV (near the Z peak): small stat errors, systematics dominate. This is a measurement and given the method doesnt change, its 100% correlated between the years
    • >200 GeV: extrapolation based on confidence of the Monte Carlo simulation capturing the energy evolution of the ID variables
      • we decided 100% correlated between 2016 & 2017. A 50 GeV electron looks a lot like a 1 TeV electron pixel wise so any differences are already accounted for. Any energy disagrement would come from something else which is constant between the two years
      • it is yet unclear if 2018 is correlated or uncorrelated due to the HCAL endcap upgrade
        • UPDATE as of 14th Dec, 2020: Since most of the variables are ECAL related, we suggest to treat the ID systematics correlated between all the years unless there are strong points against it.
  • conversion safe electron veto:
    • as this is arguably an extrapolation, we uncorrelate 2016, 2017 and 2018 due to the different pixel detector / pixel running conditions between the years
  • energy scale:
    • the errors are: stat, systematic, gain
    • stat: neglibible and can be ignored
    • syst: a measurement, correlated between the years
    • gain: the error on the gain category mostly coming from the statistics of each gain category electrons and therefore it can be taken to be uncorrelated between the years. Really only important for gain1 which are our highest energy electrons. Also helps that 2016 and 2017 have different energy regressions which may be modelled differently in MC so they really should be uncorrelated.
  • energy resolution:
    • the errors are correlated between the years as they are a measurement and the same method is used in all years

Summary of available Egamma POG ntuples

2016
Rereco version: /*/Run2016*17Jul2018/ (legacy rereco)
MC samples: /*/*RunIISummer16MiniAODv3/
egamma ntuples: /eos/cms/store/group/phys_egamma/soffi/TnP/ntuples_04162018-Legacy2016
Regression: PF/SC/Egamma 74X /74X/74X trained
Scale and smearing: embedded in MiniAOD V2
JSON: Cert_271036-284044_13TeV_23Sep2016ReReco_Collisions16_JSON.txt
2017
Rereco version: /*/Run2017*31Mar2018/
MC samples: /*/*12Apr2018/
egamma ntuples: /eos/cms/store/group/phys_egamma/soffi/TnP/ntuples_01162018/Moriond18_V1
egamma ntuples with same tag but additional variables: /eos/cms/store/group/phys_egamma/micheli/TnP/ntuples_2017_20181116
Regression: PF/SC/Egamma 94X /74X/80X trained
Scale and smearing: embedded in MiniAOD V2
JSON: Cert_294927-306462_13TeV_PromptReco_Collisions17_JSON.txt
2018
Rereco version: //17Sep2018 (for A-C), promptReco (for D)/
MC samples: /*/*RunIIAutumn18DRPremix -102X_upgrade2018_realistic/
egamma ntuples:/eos/cms/store/group/phys_egamma/swmukher/NtupleForRecoSF/Rereco2018Data_Autumn18MC_AOD
Regression: PF/SC/Egamma 74X /74X/74X trained
Scale and smearing: embedded in MiniAOD V2
JSON: Cert_314472-325175_13TeV_PromptReco_Collisions18_JSON.txt