W3
- Target plots for week 3 for Ztoee, Ztotautau, Ztomumu
- The number of leptons, taus, electrons, muons, jets (already there)
- PtCone20, PtCone30, PtCone40 for leptons, electrons, muons (done - not done taus because no value there)
- The pT distribution for leptons, electrons, muons, taus, jets (done)
- The Eta distribution for the above (done)
- dR between objects (lepton and jets for ex)
- Can we deduce the reconstruction efficiency for individual e, mu, tau based on the distributions we have?
W4
- First find out whether 'leptons' vector includes only electrons or muons, or taus too - confirmed as just electrons and muons.
- Investigate ΔR at small values where there is a big peak around zero to see where to make cut. Look at the following values of ΔR:
- between lepton 0 and nearest tau
- between electron 0 and nearest tau
- between tau 0 and nearest lepton
- between tau 0 and nearest electron
- between tau 0 and nearest jet
- between jet 0 and nearest tau
- Make cuts to only class each particle as one type with the following order of preference
- Muons
- Electrons
- Taus
- Jets
- (i.e. an electron will only be included if it is not in the same place as a muon; a tau will only be included if it is not in the same place as a muon or an electron; etc.)
- Remake all plots, including sub-leading leptons and paying more care to graph limits, bin sizes (particularly using half-integer bins for integer quantities)
- Look at truth level objects:
- Consider, for example, ΔR between truth muon and reco muon in Z to mumu
- Look at reconstruction efficiency as function of other properties (such as eta)
W5
- Remove pT cuts in PhysicsProjectMPhys.py (done)
- Look at ΔR between:
- Reco tau (0,1,2,3) and closest truth electron
- Reco tau (0,1,2,3) and closest truth muon
- Reco tau (0,1,2,3) and closest truth tau
- (All done)
- Try removing taus with passEleOLR and then plotting ΔR again - is there any improvement?
W6
-Code
- Try to fix double-counting of truth taus
- Done by only including taus with status = 2 and barcode < 10000
- These properties were identified by looking manually at truth particle printout data
- Separate taus into matched and non-matched (done)
- Plot various things with matched and non-matched taus stacked
- BDTJetLoose, Medium, Tight - Jihyun
- Number of tracks - Robin
- pT, eta, etc.
-Analysis
- The ratio of efficiency cut to background cut
- Puzzle: BDTMediumJet = 1 in Z->tau tau showed no rectau matched with truth tau, dR>1 why?
W7
- Look at truth taus in ttbar
- Why are there maybe twice as many as there should be?
- Look at how many come from b quarks - either by looking at parents or by ΔR matching
- Combinations of jet BDT cuts and NTrack cuts
- In both Z to tau tau and (more importantly) ttbar:
- Every combination of No jet BDT cut, loose, medium, tight with and without NTrack cut in a table
- and replicate Week 6, Fig 4, "Number of tracks" after loose, medium tight BDT jet cut
- Few other points (consider later)
- In Z to tau tau events, after removing duplicated truth taus, there are still, even though very few, "some" 3 truth taus. (as shown in W6, Fig. 1)
- Mphys_vars.(py? file?): Add variables regarding ttbar
- For 3 tracks in tau, it is possible that there is only 1 charged pion but with 2 charged electrons, produced by photons along with neutral pions. Are there any ways that we can distinguish this event from the actual 3 charged pion tracks?(with Rustem)
- Thursday
- Try to remove pT cuts
- Plot truth lepton pT distributions
- DR between fake/real taus and truth b quarks
- Matching between reco taus and reco b jets?
W8
- To tackle actual b-jets being misidentified as taus:
- Plot DR between each reco tau and nearest real b-jet
- Tighten tau selection (jet BDT) and plot again
- 4 plots in total: no BDT cut, loose, medium, tight
- To tackle actual taus being misidentified as b-jets:
- Plot DR between each reco jet and nearest truth tau
- Tighten jet selection (cut on BTagMV2c00/10/20) and plot again
- Maybe 4 plots in total? (no cut, then eg BTagMV2c00 > 0.75, BTagMV2c10 > 0.7, BTagMV2c20 > 0.7)
- To see what difference these cuts make
- Plot DR between each reco tau and nearest reco jet
- 16 plots in total: each combination of tau cut and jet cut
- Thursday:
- Re-plot Fig 1 with a log scale to better identify a good place to cut
- Plot DR between truth taus and truth b-jets to see how many genuine taus we are getting from a b-decay (and thus how many come from a W decay)
- Related to this - maybe look at parents of truth taus. When parent is W, see how often this overlaps with a truth b.
- Look at reco taus that overlap with reco b-jets. Plot 2D hist of whether tau is real and whether b-jet is real. Do this for all combinations of tau cut and jet cut (except no jet cut)
W9
- Tuesday:
- Thursday:
- Re-plot Fig 3b (DR between truth taus with W as parent and nearest real B jets) with smaller bin widths to see what happens around zero
- Re-plot Fig 4c (4x4) only considering truth taus that have W as parent
- Do the same for electrons and muons, only considering those that have W as parent OR have tau as parent that itself has W as parent
- Look at truth flavour of particles that are reconstructed as taus but aren't truth taus or truth Bs (this is the 1518 in Fig 4c)
- Look at properties of real and fake taus, where the reco tau passes tight tau cut and the truth tau has W as parent
W10
- Tuesday:
- Look at fake taus: how many overlap with truth electrons and truth muons (that come from W) ? if they don't, then what is the truth jet flavour? Plot Fig. 3 again - hopefully now including electrons and muons
- Re-plot Fig 4 with log y - make sure we don't lose things off the bottom of the plots
- Sort out logic behind Fig 2 (4x4) plots so that they only consider reco (tau, e, mu) not jets that overlap with these
- Plot Fig 1 (DR between truth taus to nearest real b jets) for truth elecs and truth muons to see if there is a larger overlap (which would suggest overlap is due to B decay) or the same overlap (which would suggest it's due to spin stuff)
- Look at parents of W which are parents of tau again - is there anything special about the W when the tau overlaps with a real b jet?
- Thursday:
- Manually remove reco taus that overlap with reco electrons or muons.
- Then re-plot figs:
- 1a
- 2a
- all of 5 - separate lines for different types of background (e, mu, jets)
- Plot DR between truth lepton and nearest truth b quark for each flavour of lepton
- (additional thing if we finish: replot 1b and 1c for different levels of electron/muon ID)
W11 Tuesday
- What we discussed:
- We are already using loosest possible electron selection (anything reconstructed as electron) for vetoing tau candidates
- The problem of muons reconstructed as taus can be ignored for now as it is small relative to other backgrounds
- Electrons reconstructed as taus are mostly from Ws - this is a problem. Hopefully tighter tau cut can reduce this.
- Total charge of tracks for reco taus could in principle be anywhere in range -3 to +3, although may already be selected to be +/- 1
- Why are there so many more entries in DR between truth taus and nearest truth b quark than for truth electrons and truth muons? Could be a mistake in code
- What we plan to do:
- Double check code for DR between truth taus and nearest truth b quark. If it seems correct, try to find what is going on. Could be to do with incomplete truth b quark data.
- Remake total charge of tracks plot with range -3 to +3
- Select events with 2 b-jets (tight ID), 1 tau (tight ID), and 1 electron or muon. (we will start by requiring at least these numbers, as we are not yet requiring any particular conditions on electrons or muons). Re-plot the 2D histogram and fake tau flavour to see how it has improved/changed
- When we receive Ntuples with new tau ID variable: plot this variable for the matched taus and different backgrounds
W11 Thursday
- What we discussed:
- DR between truth electrons/muons and nearest truth b - plots still don't look right. Why 70,000 entries and there appears to be overlap at 0? Looks like a mistake.
- Appears that there is already a selection on taus requiring total charge of tracks = +/- 1
- By requiring 2 b jets we have reduced the b background in taus, elecs and muons to the extent that it is now less of a concern than light jets and electrons faking taus (this in particular is still high and we're not sure why?)
- Fake electrons and muons from other sources are high.
- Why so many muons? Could possibly be a bug in code, or due to low pT muons from pion decay in detector
- A few events with 3 b jets. Could be other flavours but this is relatively unlikely with the tight b-jet ID, could be genuine b jets from ttbb events
- Not many events with 2 taus or 2 electrons
- Properties of real/fake taus haven't really changed with event selection
- BDTJetScore - we will be able to decrease background by cutting tighter on this, although we will also lose signal
- What we plan to do:
- Check DR between truth electrons/muons and nearest truth b - hopefully this is a bug and we can get more consistent numbers
- Check if there is a tau track charge cut in python file that we didn't spot
- Plot BDTJetScore for all reco taus, not just those passing tight ID
- Check if there is a bug in muon selection
- For each of the following cuts, reproduce: all 3 2D histograms, fake tau flavour, number of each object (don't need combined lepton no)
- Tighter tau ID (maybe 0.8)
- Tau pT > 25 GeV
- Medium muon ID
- Tight muon ID
- Muon pT > 20 GeV
- Medium electron ID
- Tight electron ID
- Electron pT > 20 GeV
- Also do these with no new cuts because it seems like the new NTuples are slightly different (dramatically different in the case of the muons)
- (do all these in python file so that they are all produced simultaneously - may need more events eg 5 or 10 million)
- Check the proportions and number of reconstructed taus, fake taus, real taus after such cuts.
- Homework: Find the momentum of pion that it is likely to escape the detector, given that the pion lifetime is of order 26 nanosecond and the length of the detector is ~2m.
W12 Tuesday
- What we discussed:
- We're not sure exactly how the tight tau cut works. Might have some pT dependence. We should try to check the documentation.
- The BDTJetScore plot after event selection (caption is wrong - requires 1 tau not 2) is a sum of 2 distributions: for the tight tau required for selection, and for the (usually non-tight) taus that also happen to be in the event.
- Electron/muon pT and ID cuts look good. We should use tight ID, and for pT we might have to go higher than 20 GeV.
- Plots for tau pT cut are identical to no cut (this is just a mistake)
- What we plan to do:
- Remake plots for tau pT cut
- Look at tau documentation - try to find how tight tau ID works
- Find a way to use separate uncut electrons and muons for purpose of vetoing taus
- Check that removing electron/tau overlap is working
- For electrons and muons, use tight ID and Pt > 20 GeV as a starting point for the following:
- Plot PtCone (var) divided by Pt and also EtCone (topo) divided by Pt for electrons, muons, taus to see if this discriminates well between signal and background
- Plot the signal/background ratio of the integral of Pt as a function of Pt cut
W12 Thursday
- Brief summary:
- Change OLR to use loose ID and 10 GeV pT for both electrons and muons (currently there are no cuts on those used for OLR)
- Would be good to use full samples (we think 20 million)
- Fix pT distributions
- Only plot pT ratios up to 50 GeV
- Also plot BDTJetScore in same way
S2 W1 Tuesday
- Decide on second round of baseline cuts:
S2 W2 Tuesday
- pTCone/pT < 0.3 is very loose. We will try changing this to 0.1
- Plot ETCone for electrons and muons on top of ptCone cut
- For plotting electron and muon properties, try removing tau event requirements to increase better statistics so we can see the shapes of distributions more easily
- Look at D0 and Z0 variables, and also D0/sigma_D0 and Z0*sin(theta)
- Lower priority: look at nearby objects to taus. Plot distance to closest electron, muon, (non b-) jet for real taus and fake taus.
S2 W2 Thursday
- Try plotting signal efficiency vs background efficiency for:
S2 W3 Tuesday
- Choose a point to cut on D0Sig for electrons and muons
- Run with all cuts including this one to see what current signal and background numbers are
- Then disable one cut at a time to plot:
- the distribution
- signal eff vs background eff
- S/sqrt(S+B) as a function of the cut
- for:
- Electrons:
- pT - J
- ID - R
- pTCone - J
- ETCone - J
- D0, D0Sig, Z0 - R (remember to get D0 and Z0 from the all-cuts-enabled root file, but D0Sig from the NoElecD0SigCut file)
- Muons:
- pT - J
- ID - R
- pTCone - J
- ETCone -J
- D0, D0Sig, Z0 - R (see above comment)
- Taus:
S2 W3 Thursday
- Plot BDTJetScore from zero for:
- Tight taus
- Medium taus (that are not also tight taus)
- Do efficiency plots not using 2D hist and include:
- Line of futility
- Marker of where current baseline cuts are
- FakeElecFlavour
S2 W4 Thursday
Robin:
- Plot children of fake electrons that are truth taus from W - done
- pT distribution (up to maybe 100-200 GeV) of real electrons vs fake electrons that are truth taus from W - done
- Fake muon flavour - done
- Plot children of fake muons that are truth taus from W - done
- Plot parents of fake muons that are truth muons (but aren't from W) - done
- pT distribution (up to maybe 100-200 GeV) of real muons (reco muons that are truth muons from W) vs reco muons that are truth muons from truth taus from W - done
Jihyun:
- Reclassify fake electrons/muons that are truth taus from W as real electrons/muons - do this in PrepTopEvent by changing definition of TruthMatchW
- Re-plot the following properties with the current baseline cuts, and signal vs bg efficiency and S/B:
- pT
- pTCone, ETCone
- d0, d0Sig, z0sin
- any more? ID?
S2 W5 Tuesday
- Remake Fig. 4-7 with all cuts and event selection disabled - Robin
- Change BDTJetScore cut back to 0.9 before the following: - done
- Add number in brackets in table 1 which is number of fake taus that are leptonic truth taus - Jihyun
- From sections 7 and 8 choose different set of electron (Robin) and muon (Jihyun) cuts
- Add lines in plots to indicate current cut point
Agreed cuts:
- Electrons:
- d0Sig: 2.5 (no change)
- ID: tight
- pT: 20 GeV
- ETCone30/pT: 0.14
- pTCone20/pT: 0.14
- Muons:
- d0Sig: 2.5 (no change)
- ID: tight
- pT: 20 GeV
- ETCone30/pT: 0.1
- pTCone20/pT: 0.18
S2 W6 Tuesday
- pT and eta distributions for leading 2 jets in fake tau events and real tau events (hopefully fake tau event jets will have large pT and small eta)
- Invariant mass of:
- leading 2 jets
- leading jet + tau
- 2nd jet + tau
- ... for fake tau events and real tau events. Hopefully for fake taus will see some sort of peak at W mass
- If these have been successful then can make some jet cuts and re-plot missing pT and maybe number of jets again
- and (later) for those which reconstruct close to W mass we could plot invariant mass of these plus b-jet to see if it gives top mass
- Look at possible background MCs of events with a single W to lepton plus bbar?
S2 W6 Thursday
- Plot JVT for jets 1 and 2 in fake and real tau events
- Plot transverse mass of electron-neutrino system and muon-neutrino system (should partially reconstruct W mass)
- Plot transverse mass of each of the above systems combined with each b-jet (should partially reconstruct top mass)
Extra W6-7
- Remake W6Tues Fig 2&3 putting events in 0 bin when there is no appropriate jet. Then fix the S/B plot. - J
- Remake Fig 8 using jet overlapping with tau rather than tau itself. - R
- Remake Fig 11 using nearest b-jet. - R
- Change electron/muon pT cut in python - J
- Look at events with 1 tau and see how many have a 2nd (meeting the cuts Rustem mentioned in email) - J
- Write a list of proposed overlap removal - R
W7
W8
- Apply event cut to exclude all events where (jet1 passes jet cuts AND m(tau, jet 1) between 55-100 GeV)
- Implement final event/object selection in CERN python. Make sure to include:
- Event selection (1 lep, 1 tau, 2 b-jets using same variables as Manchester code, plus the invariant mass veto)
- Object selection (elec, mu, tau, plus loose elec/mu)
- Overlap removal
- At Manchester:
- Make candidate with exactly 1 tight e/mu and at least 2 tight b-jets
- Call CandEvent.getReco to access all objects in event without selection
- Find e/mu away from the primary e/mu
- Make plot for number of these found
- and another for how many are close to one of the b-jets
W8 Thurs
- Remake all of the distributions of the cut variables, having applied all other cuts, but now incorporating the full statistics of all available MC samples.
- This involves:
- DONE - Re-enable plotRecoLep in CERN code to show subfolders for real electrons, fake electrons etc.
- DONE - Add functionality in CERN code to disable one cut at a time
- DONE - Check histogram dimensions are correct
- DONE - Check this is working by running CERN code at Manchester
- DONE - Commit to SVN and update at CERN
- IN PROGRESS - Run at CERN for small number (1000) to check that disabling cuts works
- Run at CERN for full statistics (>= 20m)
- While running, write a python script to make the combined histograms when finished
- Make a first data-MC comparison. The first “control” I’d suggest would be to select events with exactly one tight electron and exactly one tight muon and two b-jets. this is the “classic” ttbar -> dilepton signal and should be totally dominated by ttbar. (Include in the plots the equivalent of fig 1).
- This involves:
- Add new event selection for exactly one tight elec and exactly one tight muon and 2 b-jets
- Make basic event-level plots like elec0 pT, mu0 pT
- Test using CERN code at Manchester. Commit to SVN and update at CERN
- Run at CERN for 20m
- Make stack plots
W10 Tues
- Try to match data preselection. Put it before other event cuts
- Include other MCs and try to separate out ttbar into signal and background
- possible idea for how to do this: make event selection that selects events where one of the object used to select the event is a fake. then at the end, go in this directory in TBrowser and manually copy out the values from CutCand into the table
- Remake plots
- (Already done) plot number of additional taus in both 1elecExact_1muonExact_2bjet and 1lep_1tau_2bjet
- Plot number of additional leptons (including taus) in 1lep_1tau_2bjet event selection
- Plot same sign/opposite sign for elec_muon
To do
- Understand sources of truth taus in Ztoee, Ztomumu samples (low-ish priority)
- and also events with 3 truth taus in Z to tau tau
Topic revision: r52 - 2016-04-27
- unknown