1 / 51

Summary physics WG

Summary physics WG . Stan Bentvelsen Trigger and Physics week, June 7 th , 2007. CSC endgame has started. Targeting toward finishing the CSC notes Figures and tables available by September Closing date for MC samples Having preliminary results for the late summer conferences

nevina
Download Presentation

Summary physics WG

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Summary physics WG Stan Bentvelsen Trigger and Physics week, June 7th, 2007

  2. CSC endgame has started • Targeting toward finishing the CSC notes • Figures and tables available by September • Closing date for MC samples • Having preliminary results for the late summer conferences • First draft of the notes ready by October • I.e. ready for (first) review • Final papers ready by the end of the year 2007 • Remember to use the LaTeX and ROOT standard templates • Atlas detector paper based on rel 12 • Getting on steam for doing the physics • Most of the data sets are available in rel12 • Crucial samples can use rel13 reco – justify the choice • Fix on AOD for 1 mm bug is understood and under control for a number of practical physics cases • Common tools being developed / cross checks • No excuse for ‘waiting’ anymore!

  3. CSC endgame • CSC notes trajectory has an end • Extremely useful exercise to prepare tools to understand the detector data • Extremely useful exercise to get the first realistic physics results out of the data • To my mind the ultimate reach on Atlas’ physics performance is not the main goal of the CSC notes • Scope based on initial data sets • E.g. the ‘ultimate top mass’ precision for me is outside the scope of the CSC notes. Although effects of mis-calibrations and mis-alignments should be assessed. • Gap between performance and physics communities? • Clearly got smaller in many areas over the last period • Trigger aware analysis took off in impressive way • But improvement possible • If object not reconstructed correctly – go and fix it

  4. Atlantis display • JiveXML interface to Atlantis • Now reads some EventView information in addition to AOD info. • Visualise Eventview objects, labels etc. • Use predefined cuts to select event to write-out for later viewing in Atlantis • When was the last time you visualized your favorite data? Q. Lu, M. Stockton, J. Thomas

  5. A.Shibata

  6. CSC data sets: the 1 mm bug • 1 mm bug fixed on AOD level • Allow comparison between 30mm simulation, 1 mm bug and 1 mm bug with fix • Electron scale correctly corrected • Run 5200 ttbar events Electron scale in ttbar events A. Shibata

  7. Data sets: the 1 mm bug • Same plot for the jets in ttbar events • Smaller shifts in jets but forward region slightly overscaled • Run 5200 ttbar events A. Shibata

  8. 1 mm bug • Reconstructing top • ‘commissioning analysis’, i.e. reconstructing top w/o b-tag • Backgrounds included • For both 30 mm, 1mm and 1 mm bug fix • Fitted top mass shifts almost 3 GeV due to bug • But AOD fix is effective A. Shibata

  9. Cookbook generic trigger strategy C. Potter

  10. Estimating trigger rates at 1031 J. Baines • B-group strategy for 1031 : ~100 pb-1 • Mainly for understanding the detector using J/y(mm), J/y(ee), U(mm), plusstart to look at B->mm, B->mmK*/f and B->K*/fg. Use fullscan to find K*/f • LV1 muon rates 1031 • Estimated by downscaling 1033 numbers • EF numbers • Lots more work to do to complete tables. • Input needed from you for rates & comments & suggestions on the menus

  11. Trigger objects & environment

  12. Trigger redundancies • Trigger overlaps in ttbar events

  13. Trigger redundancies in ttbar • For uncorrelated trigger objects with redundancy in physics channel: can efficiencies be determined from data?

  14. Trigger Aware Example: H→WW→lnqq • Trigger efficiencies • Around 80% after alloffline selections are applied • Trigger does not significantlychange the signal mass distribution • Limited by MC statistics W. Quayle

  15. Charged Higgs trigger • EF object characteristics wrt offline • Mode tt→(bH)(bW)→(btn)(bln) • Most problematic is the MET • Resolution not so good. • Without MET the trigefficiency will drop byfactor two • MET broken in 12.0.6 • MET is only ‘global’ trigger, all others based on ROI’s • LV1: currently performing. Loop over all cells, no muons taken into account • LV2: add muon to LV1 MET – available in rel13 • EF: problematic - no noise suppression applied. Timing issue in unpacking LArg cells. C. Potter EF trigger MET turn-on

  16. Triggering Susy • Many studies based on MET + multijets • Preselection cuts typically • ≥ 4 jets, PT > 50 GeV • Leading jet PT > 100 GeV • MET > 100 GeV • Lepton selection, if applicable • Suggest to de-emphasize MET or keep it very loose • Also historically MET triggers take time to establish • MET has good discrimination power for signal versus background • E.g. do multijet triggers to select SUSY? • Not relying on MET trigger altogether? • Thresholds guided by minbias rates (assuming 65 mb) • 1jet > 115 GeV, rate: 9.5 ± 3.9 Hz • 4jets > 25 GeV, rate: 9.5 ± 3.9 Hz

  17. Example of SU3 point Efficiency after offline jet selection cuts ≥ 4 jets, pT > 50 GeV Leading jet pt> 100 GeV Efficiency without any offline selection cuts Need alternative in case the offline requirements are loosened on Njet E.g. rely more on single jet tigger 88% acceptance for SU3 for L1jet>115 GeV 4jet trigger dead Efficiency for Susy G. Redlinger

  18. MET trigger in HLT in development For now, assuming only L1MET To get rate down to EF limit implies MET>50 GeV To go lower in MET: must combine MET with jets or leptons Susy and MET trigger G. Redlinger • Dijet+MET efficiencies • May work, but eats in Susy efficiency rapidly • Color coding according to bsolute efficiencies for SU3 • Also provide good sample for jet response, needed for QCD bkg estimation

  19. Estimate QCD contribution to MET • Determine QCD background in MET, including normalisation, from data • Measure jet smearing using multijet events with one jet pointing to MET direction in phi • MET>60 GeV • |Df(MET,jet)|<0.1 • Estimate true jet pT as: pTest=pTrec+MET • Select multijet events with small MET/sqrt(ETsum) • Dominated by QCQ • Smear each jet by resolutionfunction as determined fromdata • Expect to reproduce the tailof the MET D. Tovey Plot PTrec/pTest

  20. Method looks promising Shape of MET distribution from this procedure equivalent to the QCD MC Normalisation estimated from data by selecting regions in (Njet-MET) space QCD dominant at small MET, large Njet Estimate QCD background to MET D. Tovey

  21. Eample in ttbar analysis: Isolated lepton >20 GeV 4 jets with pT>20 GeV 3 of them: >40 GeV MET>20 GeV Dropping the requirement on the missing ET: Increase of QCD background Estimated with Atlfast QCD samples 1,2,3,4,5,6+ parton samples in Alpgen, MLM matched Dropping MET requirement LMU Munich

  22. QCD background in ttbar Verkerke&vVulpen • QCD non prompt leptons in ttbar events • Isolated leptons from semi-leptonic decay in jets • Yield estimated as ‘good’ electrons in ttbar muon events with full simulation • Fraction of events with isolated leptonswith pT>20 GeV: 2∙10-4 Standard Top Standard Top, no MET requirement

  23. Talking about tops…. Udine/ictp rel 12 AShibata rel 12 • Top peaks of various groups • Start to speak the same language • Setting the samples & selections is almost debugged • Current issues: • Combinatry versus non-ttbar background • How to separate these? Nikhef rel11

  24. p p ~ c01 ~ ~ ~ q ~ c02 l g Missing energy A sample SUSY decay chain q q l l Jets Leptons Susy in Di-lepton+jet+MET • Selection straighforward • number of leptons=2, • Pt(lep) > 10 GeV • number of Jets >= 4, • Pt(J1)>100 GeV, • Pt(J2)>50GeV, • Pt(J3), Pt(J4)>20 GeV • Bgk: ttbar, Wbb+jets, W+jets, WW+jets. Same Sign

  25. Trigger and event selection • Many presentations involve trigger menu • Creative proposals for multiple object trigger • MET+jets, MET+tau, tau-tau, etc etc. • One of the urgent questions: • How do we assess the trigger efficiencies from data? • Can we utilizing redundancies of uncorrelated trigger objects? • Somewhat linked: How to assess the (offline) event selection efficiencies • Both needed for absolute normalization – cross section determinations • Assume that efficiencies and selections are not what they are predicted by MC

  26. Pile-up and cavern background • Its clear now that next year will be setup with beam, first collisions and 75 ns running • Min bias pile-up will not be a major issue • Cavern background will certainly be present • Especially analysis with muons will suffer. By how much? • Nevertheless: Example pile-up study: VBF H→tt(lh) • Compare pileup events • Lumi 2∙1033, pileupCollisions=2.3, include det noise, cavern bkg = 2 S. Tsuno

  27. Pile-up in VBF H→tt(lh) • Jet trigger very active • Big difference by variousgenerators (fragmentation) • Large systematics in rate estimations • Mostly very forward • Muon trigger • L1 low pT muon rate increased • At EF level increase ~10%

  28. Ex New analysis: di-boson production H ZZ SUSY signal

  29. Separating signal from background

  30. WW and WZ analyses with BDT H. Yang How to build a decision tree ? For each node, try to find the best variable and splitting point which gives the best separation based on Gini index. Gini_node= Weight_total*P*(1-P), P is weighted purity Criterion = Gini_father – Gini_left_son – Gini_right_son Variable is selected as splitter by maximizing the criterion. How to boost the decision trees? Weights of misclassified events in current tree are increased, the next tree is built using the same events but with new weights. Typically, one may build few hundred to thousand trees. Sum of 1000 trees How to calculate the event score ? For a given event, if it lands on the signal leaf in one tree, it is given a score of 1, otherwise, -1. The sum (probably weighted) of scores from all trees is the final score of the event. Ref: B.P. Roe, H.J. Yang, J. Zhu, Y. Liu, I. Stancu, G. McGregor, ”Boosted decision trees as an alternative to artificial neural networks for particle identification”, physics/0408124, NIM A543 (2005) 577-584.

  31. Results for WWem+X H. Yang • BDT do seem to work • Background event sample compared to Rome sample increased by a factor of ~10; compared to post Rome sample increased by a factor of ~2. • Improvement: • Simple Cuts: S/B ~ 1.1 • Boosted Decision Trees with 15 variables: S/B = 5.9 • But how to asses the systematics? S/B =5.9

  32. Streaming Samples • Streaming samples • 10 runs with (nominally) 1.8 pb-1/run • Same events stored two ways: • Inclusive & exclusive • From production: 12.0.6.5 HPTView • Caveats about streaming samples: • Trigger table is STR-01 • no muon endcaps • streaming decision from release 12.0.3+patches • trigger decision from 12.0.6 production • csc11 simulation, 12.0.6 reco/calibrations • Sample composition is ‘correct’ only for high-pT processes: • expect an unnaturally low fake rate • We ‘don’t know’ top kinematics in data: • MC@NLO with no weights

  33. Trigger e in stream samples (csc11 W MC) L1*L2 efficiency w.r.t. truth (streaming sample) L1*L2 efficiency w.r.t. reco LBL • First check (comparing with earlier studies using MC truth): • Overall efficiency vs. electron ET, η, φ: denominator is the number of tight (isolated) reconstructed electron ‘probe’ candidates from Z decays • Clearly also need to measure reconstruction efficiency • this uses 75% of inclusive electron dataset • streamtest data has insufficient statistics for a map in (ET, η, …): can we use other triggers?

  34. reco ESD/AOD/TAG files Trigger config LB #s used to make file Faked Trigger DB for Streaming test CondDB Trigger config Run-lumi DB prototype LB: luminosity, prescales Lumi-calc tool  L Det status Calculating luminosity diagram: Richard Hawkings A. Holloway • 18 pb-1 is delivered sample luminosity • recorded is less (sample prescales, online ‘deadtime’ ) • on disk may be even less (reconstruction job errors) • How to find delivered luminosity from AOD files: • keep track of used luminosity blocks • create LumiBlock metadata in tags or ntuple files • How to apply prescale/deadtime corrections: • LumiCalc: uses metadata ntuples, database prototype • Still to-do: (release 12 user tools), validate the DB

  35. Streamtest data: dilepton mass Udine/ICTP • Offline cuts on jet, lepton, missing ET are sensitive to calibrations • we can measure some efficiencies in data: lepton identification, isolation: reconstructed Z candidates • Derive corrections to apply to MC for energy measurements • calibrate electron energy: Z mass peak • missing ET: W missing ET/ MT • Jet energy calibrations:

  36. Top mass and x-sec Udine/ICTP 92 selecte evts Good agreement in shape with CSC data The exclusive ele sample has about 27% to 28% of evts compared to inclusive ele for commissioning selection.

  37. Outlook • To large extend a Trigger & Physics week • CSC efforts at high intensity • Large Atlfast data sets (800M) will be run next week

  38. backup

  39. Stream test data Stan Bentvelsen June 6th, 2007

  40. Reconstruction e & calibrations • Offline cuts on jet, lepton, missing ET are sensitive to calibrations • we can measure some efficiencies in data: • lepton identification, isolation: reconstructed Z candidates • Derive corrections to apply to MC for energy measurements • calibrate electron energy: Z mass peak • missing ET: W missing ET/ transverse mass • Jet energy calibrations: • trickier -- constrain errors on the overall scale by comparing to the hadronic W mass • W+0 jets • 5104 (12.0.6) • vs. inclEle (11.0.X) • pT e > 25 GeV • no MET cut • Cone 0.4 jets LBL W+1 jet

  41. MC production Stan Bentvelsen June 6th, 2007

  42. SUSY MC production

  43. SUSY MC production

  44. Top MC: 30 micron samples Single top resimulation running P. Ferrari

  45. Top MC production: systematics • Study of ISR/FSR samples for top mass systematics: • AcerMC ttbar + various pythia parameters: • 500 K each are available • Different samples might be needed for cross-section • UE systematics dataset 5565 (same as 5200 without UE) 500 k ATLFAST • B-fragmentation systematics: • with ATLFAST • looking at the best choice of parameters using as baseline sample 5205. • Flavour tagging group will also produce some fullsim for that • Pile-up,100 K events with fullsim have been submitted with version 12.0.6.5. The jobs are running.

  46. Top MC production: background • QCD multijet background events with ALPGEN ATLFAST/FULLSIM: official production for CSC notes • -2pb-1 |h|<2.5 and Njets >=4 (no pT cuts), for single top analysis as request for hard lepton reduces the contribution and 2 pb-1 is enough to evaluate fake leptons from QCD. • 9pb-1 of |h|<2.5 and Njets >= 5 jet pT>15 • waiting for joboption files. • Wbb and Wcc 200 pb-1 i.e. 10K events ALPGEN FULLY SIMULATED RUN numbers 6280-6287: will enter in 12.0.7 that will start within days. • Wc sample 10k events will be produced in the flavour tagging quota: job optioons not ready yet

  47. Top MC: single top signal • simulating with ATLFAST t-channel ( and s-channel) single top events with MC@NLO3.3, which do have spin correlations to compare with toprex. • sytematics: -ISR samplesa. ISR Lambda_QCD = D*2.0 AND ISR_cutoff = D*0.5  --> tend to increase the jet multiplicity & efficiencyb. ISR Lambda_QCD = D*0.5 AND ISR_cutoff = D*2.0 --> tend to decrease jet multiplicity & eff. -FSR samplea.  FSR Lambda_QCD = D*2.0 AND FSR_cutoff = D*0.5b.  FSR Lambda_QCD = D*0.5 AND FSR_cutoff = D*2.0This would mean4 configurations for 2 (3) single-topchannels,  ie 8 (12) dataSets The numbers in parenthesis include the Wt channel Decision to be taken soon.

  48. Exotics MC production

  49. B-physics: MC production J. Catmore

  50. Trigger issues • Optimize the low luminosity trigger menu • Based on 1031 luminosity • See S Rajagopalan last Monday • The draft menu is still not complete • Missing trigger items, especially topological triggers • Rates/performance are slowly coming in. • Some datasets not yet available • Physics input needed

More Related